Image reconstruction in electrical impedance tomography (EIT) is a non-linear inverse problem. The linear model is always used in most of the reconstruction algorithm in dynamic EIT, which causes large errors of image reconstruction. In this paper, we proposed a new image reconstruction method in dynamic EIT. In the method the artificial neural network based on the error back propagation algorithm (BP ANN) is used to express the non-linear relation between the impedance change position inside the measured object and the voltage change value measured on the surface of the object. Thus, the location ofthe impedance change can be decided by the measured voltage change on the surface, and then the impedance change image will be reconstructed with linear approximated method. The reconstructed error will be decreased largely, because the impedance change position can be detected precisely by our proposed method. The experimental results indicate that the precision of the reconstructed image with our method is greatly higher than that with the back projection method.

Using low-resolution (LR) multiframes from the same object with subpixel displacement to each other is a better solution for the high-resolution restoration of undersamped images. A super-resolution (SR) image restoration algorithm based on multiframes or multisensors is proposed for blurred, noisy and undersampled images. Poisson-MAP image restoration algorithm with Markov constraint (MPMAP) is used in the proposed approach. The high-resolution restored images are obtained, and the performances of the proposed approach are presented in the simulated experiments.

According to the Mars Express mission , the MARSIS primary scientfIc objectives are to map the distribution ofwater, both liquid and solid, in the upperportions of the crust ofMars. Detection of such reservoirs of water wifl address key issues in the hydrologic, geologic, climatic and possible biologic evolution of Mars, including the current and past global inventory of water, mechanisms of transport and storage of water. Three seconda,y objectives are defined for the MARSIS experiment: subsurface geologic probing, surface characterization, and ionosphere sounding. According to the previous scientific objectives, this paper provides a description of the design approach and expected performances of the MARSIS, taking into account of Mars Orbital Laser Altimeter (MOLA) data. As matter of fact the analysis of MOLA data, from the current Mars Global Surveyor Mission, have shown indeed that Mars surface shows a fractal behavior in a range of scale going from 100m to 3÷ 10km. In this paper the results of the analysis of MOLA data by mean also a mapping and a statistical distribution of the fractal parameters performed all over the surface ofMars are shown. Then a new analytical model for the surface correlation function will be introduced taking into account the fractal behavior ofthe surface over a finite range of scales. Hence for the predicted range of variation of the surface parameters and accordingly to the instrument wavelengths the radar backscattering ofthe surface will be evaluated according to the kirchoffapproximation; in order also to predict the strength of the clutter signal and then of the penetration depth that is possible to reach with MARSIS instrument. Moreover the difference between results obtained with the fractal model and with the two scale model will be evaluated.

The purpose of the work reported here is to show that spectral information could be used with high efficiency in various science investigations. The results obtained allow to describe the procedure of forming chromaticity windows with minimum dimensions preassigned by probabilities of correct identification of image elements by their chromaticities. The ability of a color camera to distinguish small chromaticity differences for image processors have been discussed. These data are important in many applications of color computer vision systems. A multispectral image processing automated system were developed for and applied to the analytical methods that have been used for several years to control the ecological and biological processes.

Structure-based algorithm, using the Gabor filters, captures rich discriminatory texture information contained in the gray level fingerprint image. It can implement matching for the limited area image of fingerprint, when the performance of traditional minutiae-based algorithm not very well because of lack of sufficient feature points in the overlap area. This method not only suits for the small area image from widely used solid-state sensor, but also acts as the complement of minutiae-based matching. Moreover, it generates a feature code with the same length, which is benefit for quickly matching. Two methods involved are described in this paper: one is the optimal design of Gabor filters, theory and methodology; the other is whole matching strategy and procedure. The core point and its director are detected to achieve the translation and rotation invariant. Many comparison experiments are also considered carefully, such as the size of tessellation cells, etc. The experiment result shows the efficiency ofthe Gabor filters and matching method. The analysis and future work is given in the end.

In this paper, a new approach to multi-objects motion segmentation in image sequence based on cellular neural networks (CNN) is proposed. As the core ofthis approach, difference merged image algorithm is presented. In order to realize the algorithm, the reverse CNN template, the addition CNN template, the patch-filled CNN template and the composition CNN template are presented and designed. For based on CNN, this approach can improve the capability of real-time in motion segmentation. And on the other hand, because difference merged image algorithm we improved is directly used in gray-scale image processing instead of in binary image processing, it can get more information of motion and increase the accuracy of segmentation. Finally, we show the experiment results, which prove that this approach has a good capability in multi-objects motion segmentation.

Coherent contours generation requires a judicious combination of processing at di erent levels of details, making this process very complex to achieve. Indeed, one has to analysis locally surfaces irregularities, to select at an intermediate level, the right edges among a set of candidates, and to obtain at a higher level, a coherent semantics that represents the global topology of the surface. We propose in this paper an adaptive approach, using the multi-agents paradigm to achieve this task. In our approach, three kinds of agents are used, each associated with a level of analysis. Agents of a same level as well as of di erent levels, interact together using the shared environment, in a cooperative way. At a same level, intermediate results are exchanged whereas inter-levels interactions, allows higher levels agents to guide to processing of lower levels agents, by giving them feedback of their work, through reinforcement/inhibition mechanisms.

The traditional edge detection gets the edge by one-off process and has no multiresolution character. Wavelet transform can decompose images into various multiresolution sub-images. And the multiresolution edge image can be detected by calculate maxima of the sub-images. The isolated edge sub-image in only one scale cannot cany the sufficient edge information. Considering correlations in these sub-images, a new method by taking advantage of the correlation is presented. In the new method, the edge sub-images which got from the LL1,LL2,...,LLN were endued different weights and inputted into a voting-algorithm. After calculating the total weight of each pixel in the different sub-images, the pixels whose weight exceed a threshold act as the final edge point.

Scene matching is of great interest in the pattern recognition and computer vision domain. Some researchers investigated the problem of scene matching, by employing phase correlation and neural network. Although varying good results have been obtained, noise and occlusion appear to affect the robustness of these matching algorithms. In order to develop a scene matching system that is robust to adverse condition (i.e. occlusion and added noise) and produces intuitively reasonable results, a robust scene matching system based on line segments is proposed in this paper. Since line pattern is effective for scene representation and matching, the proposed system employs a two-stage hierarchy, i.e. line segmentation and matching. In the first stage, the raw scenes are transformed into line segment maps (LSM); in the second stage, the Line Segment Hausdorff Distance (LHD) measure is applied to generate the matches. The line segmentation approach is based on robust shape feature and tends to generate more consistent LSM, while the LHD has the advantage to incorporate structural and spatial information to compute dissimilarity between two sets of line segments rather than two sets of points. Encouraging results have been obtained with aerial images.

Motion estimation plays an important role in real-time video coding because it can improve compression efficiency greatly. Block matching algorithms are popular in motion estimation. The full search algorithm is the most obvious and simplistic block matching algorithm, but its high computational cost limits its practical use. Many fast algorithms are constantly proposed. Fast algorithms can greatly reduce the computation by only searching selected positions in the search window. In this paper, several block matching algorithms (including FS, TSS, TDL, and CS) are introduced and compared by compute simulation. Experiment results show that the full search algorithm provides high compression with low distortion, but it costs much time; fast algorithms reduce the computational time significantly while suffer performance loss; the three-step search algorithm is the most practicably algorithm in the fast algorithms.

This paper proposes a watermark, W, a set of independent and identically distributed Gaussian pseudo random signals, which is embedded into the coeÆcients of the high-low band and ?W is embedded into those of the low-high band at level 3 in a 3-scale wavelet transform. A watermark generation, insertion, extraction, and veri cation after a variety of attacks via image operations such as scaling, smoothing, cropping, noise adding, JPEG, SPIHT, and fractal compression, are demonstrated by using Haar and Daubechies' four wavelet transforms on the image Lenna. Experiments reporting the PSNR value of each attacked image with its corresponding detected level show that the proposed watermarking strategy is promising.

In the context of multimedia retrieval, the goal of accuracy is to a certain extent contradictory with that of efficiency. The former relies on exploiting sophisticated features, whereas the latter favors using simple features with reduced dimensionality. As an endeavor to strike the balance between these two goals, this paper presents a self-adaptive semantic schema mechanism (SSM) for multimedia databases. The SSM is implemented based on an object-oriented data model, with classes being organized into a semantic hierarchy. As its most distinguishable feature, when the conditions of certain ECA-rules are satisfied, SSM supports adaptive evolution of a schema in the form of expansion with new classes and/or compaction by removing inefficient ones. This self-adaptive evolution strategy allows a schema to optimize for the requirements of each specific application, thereby achieving a dynamic, application-specific balance between accuracy and efficiency. A prototype system for multimedia retrieval, 2M2Net, has been built based on this mechanism and validated for its feasibility.

This paper presents a content-based image retrieval technique based on interest points matching and geometric hashing. We estimate points with significant luminance variations as interest points. A small region around the interest point is located as an image patch. Low-level features are extracted to describe each image patch. To provide geometric invariant image matching, we index the image patches into a 2-D hash table by geometric hashing technique. Thus, the matching is invariant to global and local geometric transforms. In addition, since we use the image patch to capture the local information, the indexing can effectively handle partial matching. We formulate a matching criterion by weighted voting technique to incorporate the spatial interrelationship into consideration. We have performed a series of experiments to confirm the effectiveness of our method. Images are globally transformed and locally manipulated to examine the efficiency of our indexing scheme. Experimental results indicate satisfactory retrieval in the case of partial matching and geometric transformation.

Most Watermarking in copyright protection is usually concentrated on how to=make the=watermark robust to resist= any intetional or unintentional attacks. They often do not consider how to resolve the rightful ownership if several= watermarks are detected.= In this paper,=we use the block wavelet transform that is included in the compression standard ? jpeg=2000=to= compute=coefficients.= The image=is partitioned into non-overlapped=blocks and then=each=block is decomposed=into 10= subbands. For each block, we find a list of significant wavelet coefficients to embed the watermark. Then, we use a= cryptography=algirithm=rc6 to select the coefficients=from=the=list of significant=wavelet=coefficients=for watwrmarking.= Finally,=we=embed=the watermark=in the=image.=The purpose of using=the encryption=algorithm=is=that=the watermarking= algorithm can be public while the wavelet coefficients for embedding the watermark can be known only through the= secrete key of the encryption algorithm.=

Picture Archiving and Communication System (PACS) is a system which connects imaging modalities, image archives, and image workstations to reduce film handling cost and improve hospital workflow. Handling diagnostic ultrasound and endoscopy images is challenging, because it produces large amount of data such as motion (cine) images of 30 frames per second, 640 x 480 in resolution, with 24-bit color. Also, it requires enough image quality for clinical review. We have developed PACS which is able to manage ultrasound and endoscopy cine images with above resolution and frame rate, and investigate suitable compression method and compression rate for clinical image review. Results show that clinicians require capability for frame-by-frame forward and backward review of cine images because they carefully look through motion images to find certain color patterns which may appear in one frame. In order to satisfy this quality, we have chosen motion JPEG, installed and confirmed that we could capture this specific pattern. As for acceptable image compression rate, we have performed subjective evaluation. No subjects could tell the difference between original non-compressed images and 1:10 lossy compressed JPEG images. One subject could tell the difference between original and 1:20 lossy compressed JPEG images although it is acceptable. Thus, ratios of 1:10 to 1:20 are acceptable to reduce data amount and cost while maintaining quality for clinical review.

Modern andrology laboratories or departments evaluate sperm to analyze male ejaculate and to determine male infertility. The evaluation includes the sperm's physical properties, sperm morphology and sperm progression. Among all these characteristics, sperm progression is a predicator ofmale fertility and is used in assisted reproduction programs. The analysis of sperm progression is based on the kinematic measurement of the sperm's motion. With the purpose of quantitatively measuring sperm's progression by using image processing methods and then providing an assessment of sperm motility without the need for complex equipment, in this paper a low cost, high efficient PC-based CASMA system is presented that is suitable for laboratorial reproduction research and clinical requirements.

Clustered microcalcifications (MCCs) on mammograms are an important early sign of breast cancer. An intelligent computer-aided diagnosis system can be very helpful for radiologist in detecting and diagnosing MCCs earlier than typical screening programs. In this paper, the detection algorithm is able to extract high-frequency signal and remove low-frequency background by exploiting a difference-image technique in which a signal-suppressed image is subtracted from a signal-enhanced image to remove the structured background in a mammogram. The difference image is thresholded to detect these MCCs in mammograms. The algorithm is tested with a series of clinical mammograms. A true positive rate ofmore than 75.5% is obtained at a false-positive (FP) detection of 2.18 per image

Periodic broadcast techniques have been demonstrated to effectively deliver the popular media contents with short startup delay. Due to the broadcast nature, this approach optimizes the server streams sharing by default. With essentially no overhead in scheduling user requests, these server-push schemes easily accomplish better service scalability than other elaborate client-pull designs. Unfortunately, there has been no support for interactive VCR-like services. Users are only allowed to playback the media title in a regular fashion. In this paper, we propose a novel streaming protocol to compensate for such shortage. The idea is to recursively split the successive video frames into several disjoint groups. Yet, different groups are served by the corresponding clusters of server channels at the reduced delivery capacity. With such design, each client is allowed to individually fulfill the specific interactivity requirement by tuning to the proper channels in the clusters with no server intervention. Consequently, the scalability of interactive services is maximized with no sacrifice on the support of the regular service demands. The performance study demonstrates significant server bandwidth saving can be realized using the proposed technique.

Recently established high definition television (HDTV) standard is expected to replace the conventional analog television standards such as NTSC, PAL and SECAM in the next few years. However the high cost of HDTV is proving to be a major factor impeding its popularity. Integrating HDTV with other home appliances is likely to increase its usefulness and popularity. In this paper, we propose an efficient scheme to use HDTV as a computer monitor in addition to its entertainment role. In the proposed scheme, we assume that the main computer is situated in a remote location. The computer raster in the remote server is compressed using an MPEG-2 encoder and transmitted to the home. The built in MPEG-2 decoder in HDTV decompresses the bit stream, and displays the raster. The HDTV will be fitted with a mouse and keyboard, through which the interaction with the remote computer server can be performed. The HDTV can thus be used as a high-resolution computer terminal. The experimental results show that the performance of HDTV as a remote terminal is very good, with marginal degradation in text quality due to compression noise.

Multicast streaming has been extensively investigated to tackle the network bandwidth challenge. However, the ubiquitous deployment of IP multicast on the Internet is expected only after several years to come due to the manageability and scalability problems. We study a pragmatic solution to enhance the feasibility of multicast media streaming by focusing on an one-to-many streaming over the source specific multicast (SSM) service. In [1], we propose a synchronized multicast media streaming framework employing server-client coordinated adaptive playout. The adaptive playout mechanism controls the playout speed of audio and video by adopting the time-scale modification of audio. Based on the overall synchronization status as well as the buffer occupancy level, the playout speed of each client is manipulated within a perceptually tolerable range. In this work, we are extending the framework to incorporate the error recovery. Each client performs an interactive error recovery with the assistance of adaptive playout. RTCP-compatible signaling between the server and clients is performed, where the cumulative feedback for retransmission (to address the bandwidth restriction on the control messages) is assisted by the adaptive playout. The simulation results demonstrate several enhancements in terms of playout discontinuity and error recovery.

Works on QoS-enabled IP networks have led to two distinct approaches: the integrated service (IntServ) and the differentiated service (DiffServ) architectures. To address the tradeoff between service guarantee and scalability, a resource manager (a.k.a. bandwidth broker: BB) can be employed to complement the IntServ/RSVP with the DiffServ in the pursuit of end-to-end QoS. One major component of the resource manager is a decision mechanism for resource allocation, which enables hosts to request per-flow, quantifiable resources along the end-to-end path and to obtain feedback regarding the acceptance. However, most of existing resource manager implementations are still adopting the decision mechanism that makes a decision on the immediate availability of resources, especially bandwidth. Considering the variations in the demand over the time, we can easily expect some level of inefficiency due to this in terms of resource utilization and management. Thus, we are investigating the methods to support the request scheduling and advance reservation for the dynamic resource management. We use a time slot manager to ensure that the committed resources never exceed a specified limit and to predict the unused (but reserved) bandwidth. Network simulations are conducted to evaluate the enhanced performance of the proposed mechanism (i.e., with respect to the acceptance rate, the resource utilization, and others).

Hydrogenated amorphous silicon is known for its large area imaging applications because of its high photoconductivity and high absorption coefficient in the visible light range. This material can be also applied to X-ray imaging when coupled with a uniform scintillation (e.g. Gd2O2S phosphor) film integrated on top of a 2-D detection array. A thick phosphor layer is the prerequisite for high X-ray conversion efficiency. In reality, however, there may be significant crosstalk between adjacent pixels thus undermining spatial resolution. This paper introduces a high aspect ratio microstructure with the new photoresist SU-8 epoxy, which limits the phosphor to regions above the photodiodes. The differences between the above scheme and that of a continuous phosphor layer are compared in terms of the absorption efficiency, the conversion efficiency, and the modulation transfer function (MTF). The measurements are carried out in a medical testing environment with X-ray source voltages of 40-120kVp. The results show a great improvement in the spatial resolution.

In this paper the performance of a GaAs detector based on a pixel matrix architecture is experimentally evaluated. Several spectral distribution have been obtained for a single GaAs pixel and compared with that of a reference Si sample in order to determine the Charge Collection Efficiency and the Energetic Resolution. The obtained results indicate the detector ability to interact with low-energy X-ray to generate good quality images for high-contrast details

This paper describes a new method of the optimal design of electrical capacitance sensor, which is based on genetic algorithm, and a modified BP algorithm of image reconstruction for electrical capacitance tomography system. In the optimal design of sensor, the evenness of sensitivity distribution, the ratio of maximum capacitance to minimum capacitance and the ratio of full-pipe capacitance to empty-pipe capacitance are considered. In image reconstruction we use the curve of components ratio and threshold valve of pixel gray to improve the performance of BP algorithm. Better quality and accuracy is obtained in experiment.

The volumetric data set is important in many scientific and biomedical fields. Since such sets may be extremely large, a compression method is critical to store and transmit them. To achieve a high compression rate, most of the existing volume compression methods are lossy, which is usually unacceptable in biomedical applications. We developed a new context-based non-linear prediction method to preprocess the volume data set in order to effectively lower the prediction entropy. The prediction error is further encoded using Huffman code. Unlike the conventional methods, the volume is divided into cubical blocks to take advantage of the data’s spatial locality. Instead of building one Huffman tree for each block, we developed a novel binning algorithm that build a Huffman tree for each group (bin) of blocks. Combining all the effects above, we achieved an excellent compression rate compared to other lossless volume compression methods. In addition, an auxiliary data structure, Scalable Hyperspace File (SHSF) is used to index the huge volume so that we can obtain many other benefits including parallel construction, on-the-fly accessing of compressed data without global decompression, fast previewing, efficient background compressing, and scalability etc.

In this study, a simple and efficient scheme to improve the compression ratio of the block–based motion fields in the video compressing is proposed. In the proposed scheme, each motion field is represented by a small set of coded motion vectors and a binary bitmap. Some of the motion vectors in a motion field, named as significant motion vectors (SMVs), have the property of being significantly different from the motion vectors of their spatial or temporal adjacent blocks. Instead of coding all the motion vectors in a motion field, only the SMVs need to be coded in the proposed scheme. For each block in a frame, one bit, corresponding to its position on the binary bitmap, is used to record if the motion vector is significant or not. And then, the joint bilevel image experts group (JBIG)image coding standard is applied on the bitmap to improve the bit-rate. The experimental results reveal that more than half of the motion vectors are non–significant and can be omitted on coding procedure. Although there exists an overhead of JBIG coded binary bitmap, the coding improvement ratio of the proposed scheme is over 40% on average when comparing with the conventional variable length coding (VLC)sc heme.

We propose two modi ed versions of the original wavelet di erence reduction (WDR) algorithm proposed by Tian and Wells. The rst algorithm encodes an image without entropy coding and achieves comparable PSNR performance with the original WDR at low bit rate while exceeding the PSNR performance of original WDR algorithm at medium to high bitrates. The second algorithm achieves slightly higher PSNR performance than that of original WDR with similar computational complexity. The modification to the original WDR is based on the observation that encoding the symbol streams generated in the sorting pass (significance map) and re nement pass (magnitudes) separately can be beneficial. We can either encode the symbol streams under different contexts using entropy coder, or output the raw symbol streams without entropy coding stage. It is shown in the experiments that up to 1/3 of both the encoding time and the decoding time can be saved without sacrificing any PSNR performance when compared to the original WDR.

This paper presents an error resilient H.263 video compression system over noisy channels. We develop a video segment regulation algorithm at the decoder to efficiently identify and correct erroneous start codes and block addresses. In addition, a parity-embedded error detection technique is also implemented to enhance the error detection capability of the decoder at the macroblock-layer. After performing above two approaches, the decoder can report the accurate addresses of detected corrupt blocks back to the encoder via a feedback channel. With these negative acknowledgments, the precise error tracking algorithm is developed at the encoder to precisely calculate and trace the propagated errors for INTRA refreshing the contaminated blocks. Simulation results show that the proposed system yields significant video quality improvements over the motion compensated concealment by PSNR gains of 4 to 6 dB at bit rate around 32 kbps in error-prone DECT environments. In particular, this system complies with the H.263 standard and has the advantages of low memory requirement and computation complexity that are suitable for practical real-time implementation.

This paper introduces our recent research work on the development of a scalable foveated visual information coding and communication system, which follows two emerging trends in visual communication research. One is to design rate scalable image and video codecs, which allow the extraction of coded visual information at contin- uously varying bit rates from a single compressed bitstream. The other is to incorporate human visual system models to improve the state-of-the-art of image and video coding techniques by better exploiting the properties of the intended receiver. The central idea of the proposed system is to organize the encoded bitstream to provide the best decoded visual information at an arbitrary bit rate in terms of foveated visual quality measurement. Such a scalable foveated visual information processing system has many potential applications in the eld of vi- sual communications. Signi cant examples include network image browsing, network videoconferencing, robust visual communication over noisy channels, and visual communication over active networks.

In the field of human motion analysis, recovering 3D human postures from image sequence may encounter many difficult issues due to the little understanding of how visual mechanisms works. This topic has motivates increasing interests among image processing researchers. In this paper, a method of recovery of 3D human posture from single view is proposed. The recovery process is as follows: First, the image sequences of human motion are acquired by single camera. The body region is extracted from the image by background subtraction, and body silhouette is represented by FCC(Freeman Chain Code); then each part of the body is labeled, and the coordinates of joints connecting the body parts are retrieved. Thus the sequence of 2D human skeletons is recovered from the image sequence. At last, 3D human skeleton can be recovered by perspective projection model of the camera and knowledge about human structure. This paper gives the tracking results for three different kinds of human motion. The 3D human skeletons can be successfully retrieved in the whole course ofthe complex movement. This proves that the system have certain advantages and fascinating applications. It is a basic and important step toward the recognition of human motion.

A new method is proposed, in this paper, to magnify a portion of an image based on the fact that fractal attractor has details at every scale. The formulae for magnifying image portions with fractal method are deduced. The fractal codes, the coefficients of a set of contraction mappings, are determined by encoding the portion to be magnified. The fractal codes are then modified according to the formulae deduced in this paper, and the magnified image is obtained by decoding the modified fractal codes. A portion of Lenna image is enlarged by a factor of 8 at both horizontal and vertical directions with 2 methods respectively, one is the new method described and the other is pixel duplication. Experimental results show that the new method is good for partial image magnification with no block effect.

A new technique based on feature localization is developed for segmenting and tracking objects in videos. A video locale is a sequence of image feature locales that share similar features (color, texture, shape, and motion) in the spatio-temporal domain of videos. To exploit the temporal redundancy in digital videos, two algorithms (intra-frame and inter-frame) are used to grow locales efficiently. Multiple motion tracking is achieved by tracking and performing tile-based dominant motion estimation for each locale separately. Hence, the difficulty of multiple non-dominating motions is avoided. Tests on natural videos have shown very good results.

Due to the low resolution and few gray-levels in B-scan ultrasonic images, it is difficult to recognize their contours. This paper presents a new method to extract the contour of B-scan ultrasonic images. To get the image's edge, the discrete dyadic wavelet transform has been used on the image's histogram, and by detecting the zero-crossings after wavelet transform in the histogram, the image contours are obtained. An edge mould has been constructed based on the artificial extraction of concerned contours, which can be used to extract the particular contours of sequential images automatically. This algorithm has been successfully employed in contour extraction of human kidney B-scan ultrasonic images thus demonstrating its practical and satisfactory application potential.

Edge detection is important in many fields such as pattern recognition and computer vision. Many edge detection methods are sensitive to noises because gradients are used to enhance edges. To solve this problem, a new edge detection method is proposed in the paper based on local self-similarity. For any pixel in an image, a metric, called as local self-similar coefficient, is defined on its square neighborhood. The square neighborhood blocks are classified into three types: edge block, smooth block and random block. Two theorems have been proven according to the self-similar metric definition and the image block classification. The theorems and experimental results demonstrate that the local self-similar coefficients on edge blocks and smooth blocks are much greater than that on random blocks. Fortunately, it is quite easy to distinguish edge blocks from smooth blocks. A new edge detection algorithm based on these properties is provided in the paper. Several kinds ofimages, including human pictures and natural scenery, are used to detect edges with the new algorithm, and satisfactory results are obtained. The results show that under noisy conditions, the new algorithm extracts better edges than Sobel method.

The Gabor transform has been recognized as being very useful in diverse areas such as speech and image processing, radar, sonar and seismic data processing and interpretation; however, its real time applications were limited due to its high computational complexity. To reduce the computational complexity, the real-valued discrete Gabor transform (RDGT) was presented in our previous work. In this paper, firstly, the 2-D RDGT and its simple relationship with the 2-D complex-valued discrete Gabor transform (CDGT) will be briefly reviewed; secondly, time-recursive algorithms for the efficient and fast computation of the 2-D RDGT coefficients of an image and for the fast reconstruction ofthe original image from the coefficients will be developed; thirdly, two-layer parallel lattice strLictures for the implementation of the algorithms will be studied; and finally, the computational complexity of the proposed algorithms will be analyzed and compared with that of the existing 2-D CDGT algorithms, which points out that the parallel implementation ofthe proposed algorithms are very attractive for real time image processing.

Since chip shape directly affects the normal running ofthe automatic machining system, such as CNC, FMS, CIMS and so on In order to ensure their normal running, the supervision of tool cutting state was necessary and significative research task. In this paper, we analyze the method ofindirectly supervising chip shape by signal changing, for example cutting force, cutting temperature, Acoustic Emission, ray-electronic signal and sound signal when chip effuses out etc. and point out their flaw in theory or application and bring forward chip recognizing method based on chip shape features. The differentiate method according to gray-scale of chip image, can get chip shape feature what need by setting appropriate value. Then the feature is extended and transformed and get new shape feature. Finally recognize chip shape by calculating euler number of new feature. This paper puts forward recognizing theory's algorithm and develops computer program. The method can recognize typical cutting chip, such as "C" chip, spiral chip and abnormal chip and was proved to have a good precision by recognizing chip experiment in practical machining.

Automatic segmentation of echocardiographic sequences can decrease the processing time required for calculation of LV volumes and systolic function parameters, reducing at the same time the intra- and inter-observer variability. This paper proposed an approach to enhance quality of noisy echocardiographic images, and gain robust segmentation of cardiac structure during motion cycle. The approach is composed of preprocessing with PDE-based nonlinear diffusion filtering, followed by endocardial border segmentation using curve evolution algorithm. Initial experiment show encouraging results on analyzing echocardiographic image sequences based on the framework of PDE.

Frame rate up-conversion is used to reduce large area flicker. In this paper, frame average interpolation and interpolation based on motion compensation are analyzed and the results are compared. Based on the differences that the interpolated pixel projects on the two connective frame, a simple and effective interpolation algorithm which is sum of the result of frame average and motion compensation is presented. Experiment results show that this mixed interpolation can maintain the advantages of motion compensation interpolation and at the same time use the motion vectors more safety, that makes it more appreciable in the still areas than motion compensation interpolation.

In the new MPEG-4 video coding standard, the semi-automatic video segmentation plays a key role in supporting object-oriented coding and enabling content-based functionalities. A novel hierarchy optical flow based semi-automatic video segmentation method is presented in this paper. The proposed segmentation method contains spatial and temporal segmentation. For the spatial segmentation, a point-based graphic user interface (PBGUI) is presented, with which the user can input easily, and then active contour model and tracking bug algorithm are applied to precisely define the video object of interest to be segmented. With the result of spatial segmentation, the temporal segmentation involves non-rigid object contour tracking and rigid object whole-tracking by hierarchy optical flow algorithm based on the Lucas-Kanade algorithm. And the tracking point selection algorithm is proposed to greatly improve the tracking performance in the rigid object whole-tracking. The experimental results show that the proposed algorithm can precisely segment video objects from video streams.

With the increasing number of people who can afford to make videos to record their lives, home videos play more and more important role in people's lives. Video abstraction is an efficient way to help review such a huge amount of home videos. In this paper, an automatic home video abstraction method mainly using audio contents is presented. The audio contents are first segmented and classified as speech, music, silence and special sounds basing on audio short-time features and morphology. Then special sounds are further categorized as songs, laughter, applause, scream and others using Hidden Markov Model (HMM). After that, motion level and blur degree are acquired using the video contents. Finally, video segments containing special effects, such as speech, laughter, song, applause, scream, and specified motion level and blur degree, are extracted as the main parts of the abstract. The remaining parts of the abstract are generated using key frame information. The experimental results show that the proposed algorithm can extract desired parts ofhome video to generate satisfactory video abstracts

Eye detection is one ofthe principal tasks in face analysis. The eye location is very helpful in many applications. This paper regards eye detection as two-class problem and employs SVM filter for finding the eye locations. Unlike other eye detection methods, we do not exploit any of explicit feature extraction approaches. Instead, the grayscale values of eye sub-image are fed to the filter directly. This is based on the observation that SVM has good learning capability in high dimensional feature space and incorporates the mechanics similar to feature extraction in its own architecture. Experimental results show the effectiveness ofthe presented system, even for low quality images.

Chromosome karyotyping is an important procedure in clinical and cancer cytogenetics research. In this study, based on efficient chromosome image processing and approximate knowledge-based geometry classification, a new automatic karyotyping technique is introduced which comprises some efficient intelligent sequential phases, namely chromosome image collector, chromosome image processor, chromosome analysis result database. In this paper, detailed techniques including image preprocessing, chromosome incision and chromosome classification have been presented. By testing on a large number of clinical data, excellent result has been reached and the performance of automation has been greatly improved.

Color histogram has been used as one of the most important image descriptor in a wide range of content-based image retrieval (CBIR) projects for color image indexing. It captures the global chromatic distribution of an image. Traditionally, there are two major approaches to quantize the color space: (1) quantize each dimension of a color coordinate system uniformly to generate a fixed number of bins; and (2) quantize a color coordinate system arbitrarily. The first approach works best on cubical color coordinate systems, such as RGB. For other non-cubical color coordinate system, such as CIELAB and CIELUV, some bins may fall out of the gamut (transformed from the RGB cube) of the color space. As a result, it reduces the effectiveness of the color histogram and hence reduces the retrieval performance. The second approach uses arbitrarily quantization. The volume of the bins is not necessary uniform. As a result, it affects the effectiveness of the histogram significantly. In this paper, we propose to develop the color histogram by tessellating the non-cubical color gamut transformed from RGB cube using a vector quantization (VQ) method, the General Loyld Algorithm (GLA) [6]. Using such approach, the problem of empty bins due to the gamut of the color coordinate system can be avoided. Besides, all bins quantized by GLA will occupy the same volume. It guarantees that uniformity of each quantized bins in the histogram. An experiment has been conducted to evaluate the quantitative performance of our approach. The image collection from UC Berkeley's digital library project is used as the test bed. The indexing effectiveness of a histogram space [3] is used as the measurement of the performance. The experimental result shows that using the GLA quantization approach significantly increase the indexing effectiveness.

In the context of product search in information intermediary or infomediary, text- and nevigation-based searching mechanisms such as keyword search are usually adopted [13]. Google [5], WebSeer [4], and Alta Vista Photo Finder [1] are some prominent examples. However, such search mechanisms are not efficient for feature-based products and the major problem is that the feature-based products are difficult to be described with textual expression. A potential candidate for the search of feature-based products is query-by-example (QBE). However, our study reveals that QBE is not an ideal searching method for feature-based products. This paper proposes an image browsing technique for the search of feature-based products in infomediary. The image browsing technique allows the users to access feature-based products through a two-dimensional map constructed with self organizing map (SOM) technique. The technique overcomes the problem of describing feature-based products. Simple view and pick operations can drive the user to the desired group of products. A task-based user evaluation was conducted to examine the usability of the proposed technique and the experimental results show that the proposed browsing technique is more practical and efficient compared with QBE.

Under the condition of being covered by anti-infrared smoke screen, the grayscales of infrared image becomes fairly concentrated, thus the contrast will be worse. In order to improve the image quality of thermal image of target and decrease the time of being covered by anti-infrared smoke, a new method of improving quality of thermal image by evening dynamic local histogram is presented in this paper. A slide window is used for selecting the interested range of target. The histogram ofthe window is calculated by a program first, then the local histogram is evened so as to modify the grayscale of the center pixel of the window. The grayscales distribution of the infrared thermal image changes continuously under the condition of being covered with anti-infrared smoke, so the program will calculates the transformation function of evening dynamic local histogram of the selected window continuously, and modify the whole displaying image. The infrared image quality ofthe target in selected area is improved by using this method. The effective covering time ofanti-infrared image decreases 3 to 4 seconds, and this is very valuable for tracking target.

MPEG-4, with its superior compression, interactivity and universal access, is the most promising future standard. MPEG-4 relies on a content-based representation so it can provide new functionalities such as content-based interactivity and scalability. Content-based coding, which encodes the objects within a picture, is one ofthe most important functionalities proposed by MPEG-4, and it has attracted considerable attention. Moreover, encoding the texture ofan arbitrarily shaped object is an important part ofobject-based coding. Therefore, many researchers have being trying their best to solve this problem, and they proposed various methods to encode the texture ofan arbitrarily shaped object. In this paper, we summarize several methods of arbitrarily shaped object coding including SA-DCT (Shape Adapted-DCT), EA-DCT (Expanded Arbitrary-Shaped DCT), Mean Replacement Padding and Low Pass Extrapolation Method (LPE). We also propose two new padding method ofarbitrarily shaped object coding. Experimental results showed that the proposed methods have almost the same effect as LPE method, which MPEG-4 has adopted, so that it can improve the coding efficiency of DCT.

Due to processing of multi-channel two-dimension data, the realization of the image fusion algorithm makes a high requirement to the I/O and process capacities of the image fusion system. The high performance digital signal processor (DSP) combined with the parallel structure can provide the possibility for the realization of the real time image fusion. This paper introduces a parallel system composed by three high speed DSPs-- TMS320C6201, which can accomplish two-channel multi-spectrum image fusion real-time by pyramid algorithm at the pixel level. The system analyses the characters ofthe pyramid algorithm, and makes fully use ofthe hardware parallel structure and software pipeline. As a result, the system fairly balances heavy computation burden of some fusion algorithms, mass data storage and real-time request for hardware, which ensures the real-time realization of complex image fusion algorithm.

The network that is formed by optical transmission system and the electronic nodes can increase the transfer delay, and make the load that the electronic nodes handles overweightly , and also restrict the handling capacity of the network nodes . Optical network is constituted by optical transmission system and optical nodes which switch or select in the optical domain. The bottle neck that electronic devices causes can be eliminated, the need for the router scale is diminished, the cost of constituting network, maintenance and management can be reduced, and the utilization ratio of network resources can be raised. Hence the optical network possesses nimble and efficient ability to constitute network and quick restorative capability to deal with great malfunction. In this text, the formative course of the optical network, some international trials and OXC are described and analysed.

Because ofthe inherent features ofdetectors, the lower contrast between object and background, ambiguous image edge and great noise have widely existed in the infrared image. It's hard to get the better result by the general method when detecting and recognizing infrared images. The recognition method for infrared objects based on multiple features by integrated neural networks, which is proposed in this paper, not only has improved the reliability, but avoided the system halting because ofthe invalidity on some feature. This paper describes and implements this method from the following aspects: infrared object image processing, image segmentation, feature abstraction, and object recognition by integrated neural networks. According the experience, the image preprocessing has improved image signal noise ratio by close frame accumulation, and smoothing and decreasing noise based on the space variant scale in deformable model has guaranteed the nicer edge effect and established the good foundation for the further image segmentation. Image segmentation and feature abstraction are important steps in the course of image recognition. Segment the object image by the integrated consideration of difference operators and histogram switch, then abstract the features from it, we can find ten aspects relating to the infrared image feature and object. Finally, it fulfils information fusion by processing abstract object features with integrated neural networks, realizes the infrared object recognition, and avoids the whole system halting when some feature information is lost.

Neural networks have pretty adaptability on the multifarious features ofrecognized objects, which can fulfill the multi-feature information fusion by combining multiple neural networks linearly and enhance the performance of recognition system. For the linear combination, it can't select the best dynamically to regulate the contribution of individual subnets because it combines the static weights ofthe output in subnets, which has limited the whole network performance. This paper puts forward an optimized linear combined method on multiple neural networks. This method determines the optimized combination weight by constructing estimate function ofthe whole network performance, gives the computable mathematical model for this optimized combination weight estimated method, and discusses the robust in the multiple neural networks system by optimized linear combination. From the simulation experience, this method is used on object recognition by multi-feature information fusion and gains more satisfying result than general multi-neural network linear combined method.

The paper presents an improved robust watermarking algorithm against rotation, scale and translation (RST). The watermark is embedded in magnitudes of DFT and detected based on principal axis. The detection algorithm doesn't require the original image. The results demonstrate that the method is robust to any rotation angle, wide scale ranges, JPEG compression and some collusion attacks. This algorithm is more timesaving and feasible than other algorithms in detection processing.

In the paper, according to analyzing features of flight multi-targets in the far distance, a new multi-target recognition algorithm based on conformability degree principle for multistage searching process is introduced In the cursory searching period, all possible targets are quickly detected with the round template in the filtering binary image. After the cluster analysis for all possible targets based on distance conformability degree principle is finished, the centroid of all possible points in a same target is used as center of a local entropy region. In the precision searching period, the position of maximum entropy value of the local entropy region being labeled is gained to use as a growing seed for implementing region growing. In order to decrease operation amount and improve system real-time character, with entropy conformability degree principle, an improved region growing algorithm being combined simple linking algorithm with sub-region combination algorithm is used to rebuild image target

The high-speed features of DSP device enable itself very suit for digital video processing. In this article, a new kind of high-speed DSP low-light-level video processing system is studied. Its basis of theory, main functional units, and the features of the circuitry are all discussed in detail. At last a flow-chart of the system operation is given.

In this paper, the implementation of H.263 real-time encoding on DSP-TM13XX is described. This implementation method can be easily extended to other standards such as H.263+, MPEG-i, MPEG-2, MPEG-4 etc., because these standards are essentially very similar (i.e. Motion Compensate, Discrete Cosine Transform, Quantization and Entropy Coding). In addition, TM13XX series of DSPs utilize a very powerful, general purpose VLIW (Very Long Instruction Word) processor core which makes them ideal for the high performance embedded multimedia applications. The architecture and the characteristic of DSP that enable real-time multimedia processing are firstly represented. Second, We discussed the major problem of H.263 real-time encoding on the high performance DSP. In order to ensure the real-time video processing, the main coding time-consuming modules of the H.263 video standard and the idea of optimization are discussed in details. These optimal algorithms make fully use of the DSP architecture and play very important role for the realization of real-time video encoding on DSPs. Finally, We show the H.263 encoding experiment results based on the DSP for standalone application. For CIF format sequence image, the average encoding frame rate about 2Ofps is achieved, it is expected to be used in videophone, mobile video communication, remote monitor and so on.

In this paper, a self-adaptive method is designed to compress infrared thermal image fast. This method adopts three techniques and mainly includes two steps. Firstly, we make use of infrared thermal image's faint visual effect to cut down data quantity and increase compression ratio to a higher level. We decrease the image's resolution from 8 bits-per-pixel to 5 bits-per-pixel. Secondly, improved Run Length Coding (RLC) is applied to realize further compression. Judging threshold is introduced to overcome traditional RLC's disadvantage of low compression ratio. This judging threshold is decided by the statistic of structure abundance degree of the specific image, thus this method is self-adaptive. At the same time, the introduction of threshold makes it possible to adjust the compression ratio and the reconstructed image quality. In order to shorten processing time, interlaced pixel statistic is adopted in calculating structure abundance degree instead of usual means of the whole image statistic. Coding results prove that this novel compression method has good performance and is suitable for infrared thermal image coding.

Holographic Transformation (HT) of a black-white bitmap image was successfully realized using the free transferring principles of light and the principle of computer generated holography. The original image was successfully reconstructed on the computer. The possibility of using HT for image encryption in image communication is experimentally analyzed and testified. It is shown that five parameters in transformation and their sequence can be taken as the key to regenerate the original image, thus a certain degree security is realized.

The importance of phase and amplitude in the field of image processing is compared by using the method of Fourier transform. The digital image is encoded with only-phase encode and the computer simulation results are given. The research shows that phase is more important than the amplitude, and phase-only code of image is effective for image encryption and we should make full use ofthe phase of the image.

Videokymography (VKG) is a powerful and cost-friendly method to observe the variability of the vocal fold vibration. A new quantitative method based on image processing, which introduces snakes model and genetic algorithm to improve precision and speed, is presented to analyze the vibration information in VKG automatically. To verify the precision of the proposed algorithm, an indirect simulation setup of vocal folds has been performed. One hundred and twenty images from twelve subjects have been analyzed, and the result shows that the vibration characteristics of vocal folds can be recognized more exactly, and the diseases in vocal fold can be diagnosed quantitatively.

For high-speed data stream from a high performance CCD image sensor, optimizing data channel and extending bus bandwidth are essential to ensure real-time storage of image, based on present computer architecture and SCSI technology. This paper put forwards a method of bypassing the main memory to directly transfer data from an image acquisition device to a SCSI hard disks array in this paper. Moreover, It describes the process of creating a high-speed data channel in detail. In order to simplify the structure of CCD image storage system, an embedded sample has been designed. A series of tests show that the storage rate will be up to 130MB/s if designed on embedded RTOS platform and 64-bit/33MHz PCI bus, so it can meet the needs of some high-speed CCD applications where speed and image quality are crucial.

The integration time limits the performance of CCD imaging-system, especially when it is used in MAVs (Micro Air Vehicles), so it is important to regulate the integration time of CCD in order to improve the quality ofmoving images when MAVs is flying. In this paper, we analyzed the relationship between image smear and the flying status of MAVs, and the degradation of the image quality caused by the variety of the integration time of CCD. We have proposed a method to ascertain the integration time of CCD imaging-system on the basis of analyzing the relationship between the integration time and the quality of moving images, and constructed a kind of experiment system to validate the method. The experiment result indicates that the method is valid for ascertaining the integration time of micro CCD imaging-system used in MAVs.

An automatic tracing system applied in USB video camera is introduced. The computer calculates the video image that captured and delivered via USB interface with a tracing target which expressed as a template, and judges whether it is in the middle of the view, if not, then gives commands via USB interface either to drive the camera tracing the target. In order to shorten the procedure, we implement an SSDA algorithm (Series Similar Detection Algorithm) on template matching and modified it. The system shows both the efficiency of the algorithm and convenience of the USB interface in the project.

CPLD (Complex Programmable Logic Device) is an effective device to realize real-time parallel processing of batches of video data. In the paper, the real-time method for realizing some image enhancement algorithms in CPLD is described. It is based on Altera's ACEX 1K device that is modular enough to be used in many scientific and industrial applications and powerful enough to maintain the throughout required for real-time video enhancement.

Although gray-level fused images can optimally integrate the modalities of low-light CCD and infrared imager, operators cannot tell from which modality the details originate. Thus the fundamental that human eyes can discern much more color categories than gray levels has been used to assigns a distinct color to each sensor modality. But the color fused image which has no natural appearance will fatigue operators greatly. Our approach is building on MIT scheme and aims at achieving natural appearance in the color fused image. MIT scheme derives its basis from biological models of color vision and utilizes the feed-forward center-surround shunting neural network to enhance and fuse low-light and infrared images. We bring forward linear fusion architecture, and composite architecture that comprises the enhancement part of MIT scheme and the linear fusion architecture. Furthermore, enhancement and combination methods for low-light and infrared images of different properties have been specified.

In this article, a kind of video processing system with FPGA is introduced. First, we introduce the typical application of video processing system, then several video processing methods are compared, at last we discussed the unit of real-time video processing system and its function.

Dual-band uncooled Focal Plane Array (FPA) thermal imaging system adopts an Archimedes spiral cord chopper and a matched dual-band light filter to achieve two single-band IR images in one imaging system. Traditional methods of getting two bands images need two single-band thermal imagers, this system only needs one detector and one optical imaging system, so the system's structure becomes smaller and the cost can be reduced. This paper studies the dualband light filter and the realization of capturing dual bands images, it also researches the algorithm of dual-band temperature measurement, using this algorithm, two bands infrared images can be fused into a temperature image.

Because of the phenomenon that zooming will make the resolving power of CCD imaging sensor decline, we analyzed the character that forming the imaging signal, deduced the every frame's image signal's expressions of typical object during the focus' changing. Based on them, we presented the calculated result during changing the multiple parameters, found the rule that zooming process affects the image signal of the camera's CCD detector, and proved it by experiment.

Based on the idea of multi-resolution model, this article proposes a quarter-tree multi-resolution model of three-dimension terrain and a simplification algorithm. The author's starting point is: Firstly, it is necessary to normalize the basic model and re-sample the normalized grid by means of quarter-dividing them. Secondly, to obtain a refined model with encrypted sampling technology. With the grid being sampled, the normal vector of each mesh and MSD (mean square deviation) of normal vectors in the adjacent four meshes are also calculated. The process of encrypted sampling will continue until the requirement of the system resolution is fully satisfied. And, finally, a quarter-tree multiple resolution model of 3D terrain has been set up by the author from top to down, thus making the model much simplified with error control and mesh emerging technology on the real-time drawing basis. Tested by a series of experiments and measurements, the algorithm has proved to be highly efficient in model simplification and convenient to be put into practical application.

In the industrial process of welding, there are kinds of stress that deform the weidment and reduce the precision of welding. Through sampling and processing the ESPI stripe-images, which generate from the process of welding, finally the skeletons of these images can be extracted. Since these skeletons represent the movement of dynamic displacement fields, the ultima aim of processing ESPI stripe-image is to control the process of welding, to debase the degree of distortion for the weldment, consequently to heighten the precision of welding. On the basis of analyzing the characteristic of ESPI stripe-images, this paper bring forward the thought of integrated design, thin these images and extract its skeletons after gray transform, histogram equalization, filtering noise and binarization.

Machine vision technology is used in extend field, mostly in industrial manufacture and assembly product line. A simple industrial robot vision system consists of a CCD camera and an image processing system. This paper introduces a machine vision system for the automatic identification of the real parameters of an industrial robot based on an assembly product line. The main of this system is a digital image processing and analysis system. Through the image processing and analysis, the exact places ofaccessories should be ascertained and then the robot can pick up the wished one tinder control. To enhance the speed of the image processing, a variety of image processing functions including Hough transform are applied. Adopting corresponding hardware, real-time inspection and manipulation can be realized.

In this paper, taking the new-type medical X-ray TV system which is developed by our laboratory as the object of study, the imaging principle of the new-type medical X-ray TV system is analyzed and the characteristics of X-ray image is described in detail. On the basis of them, the studies of medical X-ray image processing are carried out. By using Microsoft C language, the softwares of medical X-ray image processing are designed, including image noise suppression, interpolation zoom, image gray transform and image sharp processing. In order to enhance the detection capability of the new-type medical X-ray TV system and improve the X-ray image, novel time domain, space domain and frequency domain X-ray image noise processing techniques are adopted. After debugging and running in the CA-C540 image processing system, the linking of the new-type X-ray TV system and the CA-C540 image processing system is realized successfully. Some good experiments and processing results are obtained. The theory and experiment results show that the methods discussed in this paper have obviously effect on X-ray image. The research results of this paper have significant meaning on biomedical and medical image processing.

This thesis presents an digital image stabilization system that is capable of real-time stabilizing a sequence of images taken by electro-optic sensors that is in a vibrating platform. The system achieves the goal of using hardware which uses TI C6000 DSP as main processor to realize real-time image processing. The application of TI C6000 DSP in the system achieves making use of the digital signal processing to realize the real-time and high-precision image stabilization. Our experimental system is capable of settling the conflict between a great deal of calculation in image stabilization and real-time request. What are discussed in this paper are the whole configuration and the advantages ofan initial designing ofreal-time DSP video processor. The circuit's configuration and main electronic devices(VIDEO DECODER CPLD DUAL-PORT SRAM etc.) used for every function block are also presented.

This paper presents a real-time digital image stabilization algorithm on PC to stabilize a sequence of images. The proposed algorithm can be divided into 3 steps. First step is to acquire an appropriate bit plane. It is easier and faster to compute with bit plane than original image. Second, through histogram processing, we can match cursorily. lfthere are error matches, we can use a block matching algorithm to match again to acquire exact result. Then the local motion vector can be obtained from it. If we apply inline MMX assemble language hybrid programming, we can make it faster on PC. Last, we eliminate the field motion vector produced in irregular conditions. Through this step, we can void local motion vectors affected by undesirable condition such as moving objects. The characteristic of this algorithm is fast and effective. By it, real-time digital image stabilization can be realized on PC.

In outdoor surveillance systems, block matching is usually used for image stabilization but prohibited for real-time video rate application due to its tremendous amount of calculation burden, though many strategies induced to save searching time. In this paper, a faster and robust image stabilization technique is presented to eliminate the unwanted video motion during watching time. Unlike other image stabilization algorithms using two projection vectors in each x or y direction, we use only one projection vector to estimate the translation motions in 2 directions between image sequences. Our stabilization algorithm compare the template vector with vectors in the referenced vector matrix constructed and find out the minimum vector-wise MSE. At last, inspired by the block matching method, we introduce its searching strategies into our algorithm to decrease the computation times, thus making real-time video stabilization could be implemented on normally available PC.

This paper offers a fast and simple design about real-time electronic image stabilization hardware system. This design produces a video-process daughterboard based on C6711DSK [1] (The DSK is a parallel port interfaced platform that allows TI, its customers, and third-parties, to efficiently develop and test applications for the C6711). It estimates and eliminates the frame-to-frame global pixel motion in an unsteady image sequence. The system achieves a real-time stabilizing sequence ofanalog signals taken by a phonograph that is on a vibrating platform. In this paper we present a CPLD system control operation on whole board. We adopt block-matching algorithm to greatly accelerate the velocity of processing. The system completed real-time electronic image stabilization. The experimental system has been simulated in computer. The system could complete to stabilize image. Now, the hardware system is being test and debugged.

Based on the photon noise fluctuation theory, the linear filter theory, the threshold specific characteristics and vision theories of the human eye, the new performance model and universal apparent distance detecting equation for LLL (low light level) imaging system will be established and further development in this paper.

Clustered microcalcifications (MCCs) in mammograms are an important sign in the detection of breast cancer. Nevertheless, it is a complex and difficult task for radiologists to detect the clustered MCCs from the tissue background of mammograms only by naked eyes. This paper presents a prototype of a computer-aided detection system to automatically detect MCCs in mammograms. The detection algorithm mainly comprises three modules. The first module, called the mammogram pre-progressing module, inputs and digitizes mammograms into 8-bit images of size 2048x2048, normalizes the images, manually extracts the breast region from the background. The second module, called the feature extraction module, is achieved by using mixed features consisting of two wavelet features and two gray level statistical features. The wavelet features are generated by a five-level wavelet decomposition and reconstruction algorithm. The gray level statistical features used in this paper are median contrast and normalized gray level value. Finally, the third module, called the MCCs detection module, discovers MCCs in the images by using a classifier. This paper uses a three-layer artificial neural network (ANN) as a classifier to segment MCCs from the processing image. The ANN takes these four features generated in the second module as inputs. The output of the ANN corresponding to the true MCC pixels is then thresholded to segment out the true MCC pixels. One advantage of the designed system is that each module is a separate component that can be individually upgraded to improve the whole system. The algorithm is tested with a series of clinical mammograms. A sensitivity of more than 78% is obtained at a relatively low false-positive (FP) detection of 2.09 per image. The results are compared with the judgement of radiological experts, and they are very encouraging.

We present an apparatus to support real-time imaging for color night vision through fusion of information derived from two channels of low-light CCD. The processing system is comprised of a bi-channel real-time video acquisition module, a low-light image pretreatment module, a core image fusion module, a real-time video output module and a user interface module. The bi-channel active videos are firstly captured and stored in the main memories of preprocessing DSP (Digital Signal Processor). Then, the low-light image pretreatment module will preprocess images buffered in the main memories. After pretreatment the real-time image datum are passed into the core image fusion module. In this module the bi-channel images are finally fused. The pretreatment and core fusion circuits are based on the special DSP TMI300 of Philips. The fused image data is then transferred to the real-time video output module for display. The user interface module is provided for easily adjusting the acquisition area of the image. The fusion system can run independently without presentation of computer.

Infrared imaging system performance models (IRISPM) have become increasingly important in providing performance estimates for developmental infrared systems. Many factors such as the target and background signature, atmospheric attenuation, sensor response, image processing algorithms contribute to detection performance. For accurate evaluation of the IR system performance, all these factors should be considered. Traditionally, IRISPM utilize the minimum resolvable temperature difference (MRTD) and the Johnson criteria to predict IR imaging system field performance. However, these kinds ofperformance models are not suitable for JR imaging system with machine vision. To make up for the default of the above performance models, this paper provides a generic JR imaging system performance evaluation model based on the machine vision, and introduces system power transfer function (PTF) to describe the total influence of subsystems on the performance. The emphasis is put on the effect of background clutter and imaging processing algorithm on the detection performance. And the theoretical expression of the Signal-to-Interference (background clutter and system noise) ratio (SIR) is derived. Then, system detection range, acquisition probability and false alarm probability are acquired according to the given detection algorithm threshold.

In many industries there exist high temperature and nuclear radiation and poisonous gas in the production, so the robot technology is emerge, it's asked to make proper response to the objective change by vision. The research on the auto-tracking-technology of objective based on video frequency possesses important meaning. Matching of the correlation tracking is fit for the tracking ofobjective in the complicated background, the template matching method is used. The operation of the image matching is very greatly, not only need to choose the digit signal processor at a high-speed, but also need to distribute the work ofsoftware and hardware reasonably.

Due to the existing problem ofdata loss and time delay, Internet Multimedia applications always couldn't achieve good real-time performance. Real-time streaming of audiovisual content over Internet is emerging as an important technology area in multimedia communications. This paper presents a system for transporting real-time video stream over Internet. A MPEG-4 software CODEC is used to compress and decompress digital video data in this system for MPEG-4's built-in flexibility and error resilience capability. New transportation architecture is designed in this system to achieve minimum delay and high network resource utilization. Real Time Protocol (RTP in short) is adopted for the purpose of real-time transportation. Experimental results show that this delivery system obtains good perceptual picture quality and a very short delay under medium bit rate. The presented delivery system will be beneficial to applications that demand high real-time performance under medium bandwidth network environment.

Coronary artery disease is greatly problematic to health in industrial countries. The information ofthe vessel diameter in angiogram is very important to detect heart vessel disease. The traditional diagnostic method suffers from inter-and intra-observer variability. Now there is no available auto-diagnosing system, our aim is to build up a objective system based on image processing technology to diagnose coronary disease.

In this paper, a real-time eye tracking technology for the adaptive human-computer interaction has been introduced. The principle of system adaptability is presented, and the system architecture is also described. The adaptive system can segment the interesting eyes' area in jittering images through utilizing the digital image processing technologies, which involve image filter, edge enhance, feature extraction et al. Then, the ideal detection algorithm for the eye-gazing direction is put forward, and the revising approach for the head slight movement is given which can calculate relative position deviation between the Purkinje image and the pupil. The system can accurately distinguish among as many as nine dialog boxes arranged in a 3x3 matrix on the display screen. At the same time, the manipulation ofperipheral equipment based on this kind ofinterface through eye-gaze is discussed. The results of the experiments have been given. The system will be used widely for disabled people.

In this paper, based on the analysis ofhead MR (Magnetic Resonance) image, an effective method is developed to automatically extract skull area and edges from MR image. Based on slip gray scale threshold and specifically qualifications, a region growing method is applied to automatically form the skull area candidates and confirm the true area. Then an improved boundary tracking method is applied to remove the pseudo-structures of the rough skull area. Limited tracking bug and some definite shape operations are applied to remove the redundant protuberances and dead zones, without any effect on the true skull area. The experimental result is provided and discussed.

In the imaging process of staring infrared Focal Plane Array (FPA), spatial-sampling frequency is determined by the physical size and spatial distance ofthe detecting unit. While the spatial frequency ofthe detected scene is always more than half of the sampling frequency of staring infrared Focal Plane Arrays, thus under-sampling happened, and the quality and utility of the image will be seriously degraded by aliasing. For the reduction of this effect, many scholars carried out a lot of theoretical analyses and bring forward some techniques. In this paper, theoretical analyses of forerunners are concluded, an integrated novel theoretical analysis on the reduction of aliasing is put forward, and the expectation of image reconstruction through uncontrolled micro-shift is discussed.

This paper presents a new Chain-Code algorithm based on eight-direction tracing. Whole code area is divided into four quadrants by this method. In each quadrant, the tracing order is different, but accord with holistic direction. Using the Chain-code tracing method and tracing the contour, a closed contour's information is obtained. By this tracing principle, all contours' information in the chain-code is gotten finally. Isodose curve with same dose, which is appeared several times in medical image process, is solved by this tracing principle. Using the chain-code in order, the original isodose curves can be drawn. However it need to be approximated and smoothed. The distance is taken as the intermediate variable, and the relationship of coordinate x and y translate into the relationship of distance and coordinate. It is convenient to smooth the isodose curve and get the ideal isodose information. The satisfactory result of isodose in 3-D medical image process is shown in this paper.

The performance evaluation model (MRCeye model) of CCD low light level imaging system is deduced on the basis of human eye models and experimental data. It is involved signal to noise in time frequency and modulation transfer function in spatial frequency. It is showed by calculation that the minimum value of MRC is obtained when spatial frequency equals to f0 C/ mrad , MRC increase with the decrease of spatial frequency when spatial frequency f is less than f0 C/ mrad , MRC increase with the increase of spatial frequency when spatial frequency f is larger than f0 C/ mrad , namely, The MRC has band-pass characteristic that is according to the characteristic ofhuman eye.

Research into limited resolution of EBCCD and experiments have been reported in the present paper. Rules in EBCCD imaging for resolution chart (stripe pattem) ofdifferent frequencies have been found. It's possible to output stripe pattern image even if the frequency of input chart is higher than the Nyquist frequency of the system (the phenomenon calls aliasing). It can be explained by frequency spectrum. Modulation transfer function has been tested. It shows that EBCCD system has a certain imaging ability to high frequency (higher than Nyquist frequency) image. This provides the possibility of the EBCCD system in imaging for resolution chart of high frequency.

The paper researched a detecting system of wheelset based on real time image acquisition and processing. The system used linear laser diode sources to illuminate the wheel treads, and the profiles are obtained by high sensitive CCD. The paper researched the dynamic image acquisition, spatial transformation, image enhancement, image segmentation and measurement algorithms. By analyzing the geometric shape of the wheel tread profile, the parameters such as the flange height, the flange thickness, the angle of flange and the back to back distance can be deduced. Then the wearing degree of the wheel can be calculated, and the more important is that the next maintaining date can be predicted. It's crucial for the safety and stability ofrunning wheelset.

In the measurement ofinterference fringe using the linear CCD, thinking ofthe separation of CCD, the integral effect of the pixel, and the sampling speciality of the interference fringe (cosine signal), three problems have been discussed seriously——"how to determine the advisable maximum of the spatial frequency ofthe fringe", "the effect on the sampling ofthe integral effect ofthe pixel", and "the effect on the contrast ratio ofthe fringe ofthe phase matching ofthe pixel".

Currently, image retrieval system are based on low level features of color, texture and shape, not on the semantic descriptions that are common to humans, such as objects, people, and place. In order to narrow down the gap between the low level and semantic level, in this study, we describe an efficient and effective image similarity calculation method for image comparison at object classes. It is not only suitable for images with single objects, but also for images containing multiple and partially occluded objects. In this approach, a machine learning algorithm is used to predict the classes of each of object-contour segments. The similarity measure between two images is been computed using Euclidean distance between images in the k-dimensional space. Experimental results show that this approach is effective, and is invariant to rotation, scaling, and translation of objects.

Visual tracking could be treated as target state representation and target state inference problem in an image sequence. Moreover, in cluttered and dynamic environments the better probabilities of accurate tracking depend on richer representation and more robust inference. Target state representation could be considered as color segmentation, contour detection and position mark. Target state inference could be treated as an evaluation from old states to new one in fuzzy logic at every step of an image sequence. This paper presents a special tracking system based on factored sampling model in order to resolve difficult and complicated visual tracking problem, such as a changing of target’s representation, a clutter of environments and an interaction of target and camera. This tracking system is applied to changeful target tracking by handling the related information to sample-set between every two time-steps in an image sequence and implemented in real time system at around 20Hz with 640*480 pixels image. Specially, color and position distributions of a target have been used in this system to estimate the target situation. The results show the robust, real-time system is able to track a target with enough accuracy to automatically control the camera’s pan, tilt and zoom in order to remain the object centered in the field of vision.

In this paper, we proposed a method, which is very fast and gives better moving information of the objects in the image sequences. The possible locations of moving objects are found first, and then we apply the Hough Transform only on the detected moving regions to find the optical flow vectors for those regions only. So we save lot of time for not finding optical flow for the still or background parts in the image sequences. The new Boolean based edge detection is applied on the two consecutive input images, and then the differential edge image of the resulting two edge maps is found. A mask for detecting the moving regions is made by dilating the differential edge image. After getting the moving regions in the image sequence with the help of the mask obtained already, we use the Hough Transform and voting accumulation methods for solving optical flow constraint equations. The voting based Hough transform avoids the errors associated with least squares techniques. Calculation of a large number of points along the constraint line is also avoided by using the transformed slope-intercept parameter domain. The simulation results show that the proposed method is very effective for extracting optical flow vectors and hence tracking moving objects in the images.

A frequency transform-based statistical method is proposed for shape matching for MPEG-7. Shape description and its corresponding matching algorithm is one of the main concerns in MPEG-7. The normalized frequency transform is invariant to translation and scaling. The image is transformed into frequency domain using Fourier Transform. Two similar images will have same power spectrum. Annular and radial wedge distributions for the power spectra are extracted. The annular and radial wedges can be set arbitrarily. Different statistical features, such as mean and variances are found for the power spectrum of each selected transformed individual feature. The Euclidean or Minkowsky distance of the extracted features are found with respect to the shapes in the database. The minimum distance is the candidate for the matched shape. The simulation results are performed on the test shapes of MPEG-7.

This paper proposes a new fast and high-resolution color structured-light-based 3D imaging system. Unlike the expensive laser-based 3D imaging system, our hand-held system contains a commercial digital camera and two special designed projectors, the cost is much lower. Only two images are necessarily to reconstruct the 3D model with textures. One projector illuminates white light for texture imaging, the other projector projects special designed pattern for 3D information imaging. 384 vertical color bars (comprised of six primary colors) with special coding are projected onto the target, our proposed algorithm detects edges between any two successive color bars. From decoding algorithm the correct locations of edges are derived. Then the calibration algorithm finds accurate positions of edges in 3D space with triangulation. Color bars can be seriously distorted by colorful objects. Edges and colors of the object may cause edges of color bars to be broken segments. And some bars even show very different colors. A curve reconstruction algorithm is developed to reconstruct the complete edge from residual edge segments. And a robust color decoding algorithm can recover the correct color sequence even when many color bars are seriously distorted. The suggested approach has been implemented and the preliminary results show the resolution can compete with laser 3D imaging systems and -0.2~+0.2 mm accuracy is achieved.

The paper presents a novel 3-D hand-held camera to get the depth information from a single snap shot. Traditionally, a stereo system needs two cameras to reconstruct the 3-D shape. We propose a technique which can be applied to the commercial digital signal camera, and make the 2D camera be able to capture 3D information. The proposed 3-D hand-held camera contains three major components: a flash, a commercial camera, and a triaperture lens. The flash projects the speckle pattern onto the object and the camera captures a single snap shot at the same time. In order to embed the 3-D information in one captured image, we use a novel lens containing three off-axis apertures, where each aperture was attached one color filter. Therefore, a captured image carries the information from three different viewing directions. A hierarchical pseudo sub-pixel correlation algorithm is proposed to compute the disparity vectors at a fast speed. It is an adaptive block-based correlation process based on the sparse array of the extracted features. The experimental results show that our approach is robust and convincing.

We present a multifunctional video stream processing peripheral component interconnect (PCI) card based on special digital signal processor (DSP). Real-time processing of the input video signal, such as 3D special effects, mixing graphic caption, making specially transform, and so on, is the core function of the system. The decoded or processed digital video can be compressed into MPEG2 video stream by MPEG2 encoder and captured to the hard disc of computer through PCI bus, and it can be encoded directly as analog or digital output video. Through the PCI bus MPEG2 videos saved on the hard disc can be read out and decoded by the card. Of course the system can transform video signal between analog and digital directly with the analog and digital encoder and decoder. Besides it also supports parallel digital video input and output interfaces. Software control command can be written to the system control module through the PCI multimedia bridge to select exact function of the system. Finally information for writing driver for the PCI card is provided plainly.

A novel adaptive method of automatic threshold selection is proposed for image segmentation in this paper. An optimal threshold is selected by the discriminant criterion acquired from the gray-level statistic of the resultant target and background region. The theoretical analysis shows that correct image segmentation can be obtained for a low-contrast image with complex background using this method, no matter the proportion of the target is small or big. Moreover, the computing quantity and the processing time can be decreased greatly by setting the gray level selection region. Several experimental results are also presented to support the validity ofthis method.

Accurate measurement of the MTF for discrete imaging system can be obtained by viewing a known frequency target. A slit target is frequently used to measure the MTF for an imaging system. Usually we have 4 methods to measure the MTF for a discrete imaging system using a slit. But for the discrete imaging systems of under sampling that is difficult to reproduce this type of target properly since frequencies above Nyquist are folded into those below Nyquist, resulting in aliasing effect. To decrease the aliasing problem, we have introduced a "super resolution technique" to our measurement, which gives MTF values both above and below Nyquist more accurately.

The information of temperature field on the back ofa weldment is analyzed by a computer vision system in this paper? A new method of seam recognition is proposed by using this information A self-adjusting fuzzy controller is used for seam tracking the system has been applied successfully in the production of spiral steel pipe with submerged arc welding? The practical production shows that the control system is reliable and stable.