The purpose of this work is the study of urban areas based on texture analysis methodology that may improve the urban investigation through remote sensing, dedicated to seismic vulnerability in correlation with the geometrical properties of building blocks and the vulnerability data, taken from surveys. Most of the algorithms for urban classification in the image processing domain are based only on the use of spectral information, not taking into account the geometric characteristics given by the spatial distribution to roads, buildings and green areas on the image. The study of these local properties, as the texture, may enrich the available information on the image, providing a further information layer which may improve the differentiability among suburban areas originally belonging to homogeneous areas. In order to overcome those limitations, a new methodology, based on the analysis of connected regions, was studied. At first the image is thresholded in order to detect built pixels. Then these pixels are connected into sub-regions and these sub-regions are investigated to extract features.

This paper deals with texture segmentation in images of wooden plates. These plates are used in pencil industry and its quality reflects on the quality of the industrialized product. We have used the spatial gray level dependence method for analyzing the plate surface, detecting nodes, stripes, and defects. This statistical method has many applications in several fields with textures. We classify the pates based on the industry requirements. With images of 128 X 256 pixels and 256 gray levels, we have used a 64 X 64 window placed over 8 non- overlapping areas. For each position the window computes the co-occurrence matrices for the directions 0, 45, 90 and 135 degrees with a chessboard distance equal to 1. The values of he angular second moment average, computed in the four directions versus the plate area, are plotted in a feature space. This shows clusters of similar textures around constant values. The good plates have a homogeneous visual appearance. The unusable plates have nodes, stripes and defects. Between then exist the high stripped plates and the medium stripped plates. We can classify then in good one or unusable depending on the industry demand. The method shows a good efficiency for discriminating between textures of the same material but with a random distribution established by the nature.

A new method is presented for focused image recovery from two blurred images which are taken by a CCD camera with different camera parameter settings. The focused image is obtained through actual point spread function (PSF) using the constrained least squares filter. Subbarao et. al. proposed an approach that the focused image is obtained through deconvolution in the Fourier domain using the Wiener filter. Subbarao's method requires the estimation of noise parameter, which is not known as a priori knowledge. This causes less accurate results in focused image recovery. On the other hand, the application of constrained least squares filtering to image restoration is optimal for each given image and requires knowledge only of the noise mean and variance. This is very useful for its flexibility. This approach fives more accurate focused image than the previous method.One of the difficulties of the image restoration problem is finding exact PSF which makes the blurred image. The actual PSF is obtained by the calibration of camera levels is used to find line spread function (LSF). The obtained LSF can be used to compute its actual PSF directly using the separable optical transfer function. The new method has been implemented on an actual camera system, and the experimental results of focused image recovery are provided and discussed.

This paper deals with the problem of restoration of images blurred by relative motion between the camera and the object of interest. This problem is common when the imaging system is in moving vehicles or held by human hands, and in root vision. For correct restoration of the degraded image it is useful to know the point spread function (PSF) of the blurring system. In this paper we propose a straightforward method to restore motion blurred images given only the blurred image itself. The method first identifies the PSF of the blur, and then use it to restore the blurred image. The blur identification here is based on the concept that image characteristics along the direction of motion are effected mostly by the blur and are different from the characteristics in other directions. By filtering the blurred image we emphasize are presented for both synthetic and real motion blur.

Low temporal frequency vibrations involve random image degradation depending on the instant of exposure. Exact restoration requires calculating a specific filter unique to each vibrated image. In order to calculate the restoration filter for each image the specific degradation function is needed. Therefore, the instant of exposure has to be measured or estimated by some other means. In this work a general restoration filter for a single vibrated image is developed. The filter is independent of the instant of exposure. Assuming that receiving each image from a vibrated image ensemble is equally likely, the filter is designed to give best performance averaged over the ensemble. Even though when applying the new filter to any vibrated image the restoration achieved is slightly poorer than with an exact filter based on the specific motion function, the new filter has the advantage of simplicity. Since the instant of exposure is not needed a new appropriate filter does not have to be calculated for each exposure, the proposed filter is more practical and suitable for real time restoration.

In inertial confinement fusion experiments, images of neutron distributions provided by a large aperture imaging system are degraded by multiplicative and signal-dependent noise. In this case, it is difficult to separate the object's fluctuations from those of the noise. Wiener filtering is a well-known method which takes noise into account. Unfortunately, the Wiener filter is a low-pass filter which blurs the contours. The propose of this paper is to introduce two new approaches which restore most of the object's fluctuations. The first approach is a Wiener adaptive filtering method that consists of locally adjusting the value of the cutoff frequency. This adjustment is based on the basic properties of the human visual system where the visibility of noise can be reduced considerably in the neighborhood of a 'strong' contour. The contour detection is realized by means of a special class of quadratic Volterra filters. These filters are approximately equivalent to the product of a local mean estimator with a high pass filter. The second approach consists in applying a multiresolution decomposition of the raw image followed by a wavelet thresholding that changes for each resolution layer. Results and comparative evaluation of these two methods will also be presented.

We report on an evaluation study of a ship classifier based on the principal components analysis (PCA). A set of ship profiles are used to build a covariance matrix which is diagonalized using the Karhunen-Loeve transform. A subset of the principal components corresponding to the highest eigenvalues are selected as the ship features space. The recognition process consists in projecting a profile on this eigen-subspace and performing a similarity measure. We have measured the recognition performance of the classifier using various sets of range-profile signatures of ship silhouette images and simulated synthetic aperture radar images of ships under various aspect angles. It is found that the PCA-based ship classifier design offers good class discriminacy when trained with a limited number of ship classes under an aspect angle range of 60 degrees about the ship side view. Additional tests are however necessary to validate the classifier on large data sets and real images.

Energy minimization approaches in the form of deformable model fittings have recently attracted considerable attention. Such models are particularly effective when the sought after shapes undergo elastic deformations that cannot readily be accounted for by RST effects. Deformable models are by their nature iterative processes. Such approaches are computationally costly and require a detailed model building step, frequently in the form of cubic Bsplines. It is reported that almost all misclassifications in the energy minimization approaches can be attributed to two problems: local minima and modeling difficulties[3]. In this paper we have taken handwritten character recognition problem and proposed a solution that bypasses the two bottlenecks above. The algorithm requires no parametric shape models and is non-iterative. Character classes are modeled from a library of handwritten digits and described by discrete spatial processes. Contour classification is then performed by a MAP rule implemented on an unknown observed digit. We have shown classification rates that match or exceed those obtained by complex deformable templates. Keywords: deformable models, character recognition

In recent years the wavelet paradigm has become a popular area of research and it has been found to be a very powerful technique for a variety of applications ranging from communication protocols to image compression. While most of the published literature deals with real valued wavelets recently there has been some interest in wavelets for finite fields, especially GF(2). This paper describes a framework for a logical-wavelet based multiresolution analysis for binary images.

A technique for polychromatic image restoration is proposed in the context of high-resolution synthetic aperture imaging. Noise and aberrations from the atmosphere are included in the analysis. The technique is simulated and successful image restorations are shown for objects with spectrally-varying radiance profiles. It is concluded that color restoration requires some additional information, e.g., phase closure or multiple-frame averaging,t o overcome aberrations for robust image formation.

The analysis and an efficient scientific exploration of the digital Palomar observatory sky survey represents a major technical challenge. The input data set consists of 3 Terabytes of pixel information, and contains a few billion sources. We describe some of the specific scientific problems posed by the data, including searches for distant quasars and clusters of galaxies, and the data-mining techniques we are exploring in addressing them Machine- assisted discovery methods may become essential for the analysis of such multi-Terabyte data sets. New and future approaches involve unsupervised classification and clustering analysis in the Giga-object data space, including various Bayesian techniques. In addition to the searches for known types of objects in this database, these techniques may also offer the possibility of discovering previously unknown, rare types of astronomical objects.

Current efforts to perform automatic galaxy classification using artificial neural network image classifiers are reviewed. For both digitized photographic Schmidt plate data and newly obtained WFPC2 imagery from the Hubble Space Telescope, a variety of 2D photometric parameter space produce a segregation of Hubble types. Through the use of hidden node layers. a neural network is capable of mapping complicated, highly nonlinear data space. This powerful technique is used to map a multivariate photometric parameter space to the revised Hubble system of galaxy classification.

We have developed a method for the linear reconstruction of an image from undersampled, dithered data, which has been sued to create the distributed, combined Hubble Deep Field images - the deepest optical images yet taken of the universe. The algorithm known as variable-pixel linear reconstruction, preserves photometry and resolution, can weight input images according to the statistical significance of each pixel, and removes the effects of geometric distortion both on image shape and photometry. In this paper, the algorithm and its implementation are described, and measurements of the photometric accuracy and image fidelity are presented. In addition, we describe the use drizzling to combine dithered images in the presence of cosmic rays.

The use of maximum entropy image restoration has, heretofore, required recursive, or iterative, search routines. This paper, by contrast, describes a one-pass, closed-form maximum entropy algorithm. The approach derives from minimization of a Log-L² error norm between object and reconstruction. The resulting output has the form of the exponential of a Wiener-type filtering of the image data. The logarithmic nature of the norm gives rise to a tendency toward constant relative error across the output field.

Low-level features such as the color, texture and shape of objects have been widely studied for similarity search in image indexing and retrieval. A new color indexing scheme based on the octree quantization scheme is proposed in this research to achieve efficient multi-resolution image retrieval. The new color feature not only integrates commonly used color features such as the color histogram and dominant color, but also support a selective filtering strategy to speed up the retrieval process. It can also be further combined with other visual features to facilitate similarity searching. Extensive experiments are performed to illustrate the performance of the proposed approach.

In this paper a comparative study of the application of four neural networks for the segmentation of magnetic resonance images of brain, is proposed. The segmentation of MRI images enable one to present tissues of the same category with an equal gray level resulting in a more clear image for future diagnosis and treatments. Results of using three supervised networks, i.e. multi-layered perceptron, probabilistic neural network and radial basis functions and one unsupervised network, i.e. adaptive resonance theory 2 will be reported.

In this paper, a method will be described for reduction of an n-dimensional feature vector into a 2D feature vector. Reaching for this goal, a structure is introduced, referred to as the chaining structure, which is generated from the initial n-dimensional feature vector. The proposed technique can be though as a feature extraction method. The simplicity and the consistency of the technique beside the fact that the resulted feature set is of 2D, are the main advantages of the proposed method. It will also be illustrated how a specially designed neural network can be used to implement the proposed method. The efficiency of the proposed feature extraction algorithm will be illustrated by applying the method to the OCR of handwritten Persian digits. Finally, it will be compared with principal component analysis.

In comparison to present security applications, pattern recognition techniques can be categorized as 'hard' automatic target recognition (ATR) and 'soft' ATR. The first category has been established for years and deals with specific object recognition. On the other hand, the second, less established category operates on very fast object class-level recognition only. The second category usually employs very fast processing and an image database. In this paper, we introduce a novel method to integrate a compression technique based on logic data representation with soft ATR. This new compression method applies Arnold's Differential Mapping Singularities Theory in the context of 3D object projection into the 2D image plane, and takes advantage of the fact that object edges can be interpreted in terms of singularities, which can be described isomorphically by simple polynomials. Therefore, compared to state- of-the-art still image compression, such as JPEG, there is no information los in high contrast, high-dynamic range image areas; as a result, the global peak signal noise ratio can be very high By using linked edges to represent an object, it is possible to use a simple set of parameters for real-time 'soft' ATR. This publication discuses various security applications of this new scheme, which is integrated with ATR compression.

Segmented object shape is an important visual feature and one of the content based image retrieval techniques is the shape based technique which allows users to ask or objects similar in shape to a query object. We describe a method for shape representation and similarity measure which we call the grid based method and evaluate its performance by comparing it with two popular methods, namely, the Fourier descriptors method and the moment invariants method. The comparative study is done by performing the same queries on the same database for all the methods. We also test the methods for robustness to noise by performing the queries on the database of noisy shapes. Experimental results show that the method proposed by us performs favorably compared with the other two methods.

The recognition of moving 3D objects is one of the most important tasks of future vision systems. Several techniques have been proposed using a sequence of images. Image difference thresholding, optical flow, token tracker...are a few examples. The main inconvenience is the amount of calculation that cannot meet the real time constraints, even on parallel computers. The snakes approach, also called 'active edges' seems more suitable for 3D-object recognition and tracking. We used this concept as a basic tool in our model-based recognition method. The proposed approach uses an elementary geometric property: if the 2D projections of 3D objects are similar, then 3D objects can be identical. A conveniently defined series of 2D projections can allow us to find correct associations of the representative projections of known objects memorized in image database. The snakes are used as an efficient means for location/tracking of a given object in a sequence of images. The good results obtained with this technique on a SUN Sparc workstation allows us to expect similar satisfaction on dedicated computer under strong real-time conditions.

The purpose of this report is to propose a new restoration technique, based on the Tikhonov regularization approach, including local properties about the original image into the restoration process, with the use of an a priori model of the solution. In order to prove the effectiveness of the proposal, we compare it with three restoration methods of images: usual Tikhonov regularization, Markov-fields and maximum entropy. In image restoration, the problem is usually addressed under the assumption that the blur operation is shift-invariant. Since real- world blurs are often shift-variant, we will either consider the shift-variant problem and its approximation, or we will use a simplifying approximation, by an invariance blur. A criteria will be defined to validate, in terms of quality restoration, the approximation of a spatially- variant blur by an invariant one. Simulation results show that the proposed method, with an accurate a priori model, out-performs the conventional Tikhonov regularization. The influence of the space-variability will be illustrated on images.

Compression is based on the removal of redundancy inherent in most images. Most of the work in the area of image compression has been directed at removing the redundancy within the same image. However, if we examine similar images from MRI scans, or CT scans, or aerial images, from surveillance photos; there is often similarity between these images. Removing this inter-image redundancy is the focus of 'set compression' techniques. However, when similar images are registered their compression ratios are much higher, using set compression techniques. In order to register images we consider the Wavelet Modulus Maxima to determine the important control points which provide shift invariant structures. Using these structures and control points form different images in a similar set of images, we register the images using the homologous set of points. By suing the difference image we are able to exploit the inter-image redundancy. The difference image can be further compressed by any other available compression technique to further increase the compression ratio. Since we are dealing with sets of images, these applications are well suited for using parallel architectures. For example in similar sets of medical images, there is a great potential to exploit the inherent parallelism is the process, by working on several images simultaneously. We make use of several nodes on each image as well as process several images in parallel. This paper discusses the issues related to the parallel implementation of our image registration and compression algorithm. This paper also discussed the results and details the gains obtained in parallel implementation using several nodes.

Optimal performance in a data compression scheme is very hard to obtain without inordinate computational complexity. However, there are possibilities of obtaining high performance with low complexity by proper exploitation of the characteristics of the data. Here, we shall concentrate on two recent low complexity algorithms for image compression which exploit data characteristics very efficiently. We shall try to discover principals for realizing high performance with low complexity and explain how these recent algorithms utilize these principles We shall also present image compression result with actual codecs which realize the promise of high compression with low complexity.

We present a technique to compress scalar functions defined on 2-manifolds of arbitrary topology. Our approach combines discrete wavelet transforms with zerotree compression, building on ideas from three previous developments: the lifting scheme, spherical wavelets, and embedded zerotree coding methods. Applications lie in the efficient storage and rapid transmission of complex data sets. Typical data sets are earth topography, satellite images, and surface parameterizations. Our contribution in this paper is the novel combination and application of these techniques to general 2-manifolds.

We present a technique to compress scalar functions defined on 2-manifolds of arbitrary topology. Our approach combines discrete wavelet transforms with zerotree compression, building on ideas from three previous developments: the lifting scheme, spherical wavelets, and embedded zerotree coding methods. Applications lie in the efficient storage and rapid transmission of complex data sets. Typical data sets are earth topography, satellite images, and surface parametrizations. Our contribution in this paper is the novel combination and application of these techniques to general 2-manifolds. Keywords: compression, wavelets, manifolds, 30 objects, zerotree, multiresolution analysis, progressive coding

We study the performance difference of the discrete cosine transform (DCT) and the wavelet transform for both image and video coding, while comparing the other aspects of the coding system on an equal footing. Based on the state-of-the-art coding techniques, we point out that, for still images, the performance gap between DCT and wavelet based coding is within one dB in PSNR at the same bitrate. This is in contrast to the common perception that the wavelet transform is much superior to the DCT for image compression. For video coding, the advantage of using the wavelet transform over the DCT is even less pronounced.

This paper presents a new method for an efficient coding of image data. Based on the wavelet transform, the algorithm utilizes the remaining correlation between subbands. A successive approximation of wavelet coefficients yields a hierarchical symbol stream, which is highly compressed with a prediction of significant descendants. The coding algorithm distinguishes itself by a high adaptivity to the image content. The resulting bitstream contains all image information in order of its significance. Therefore it is possible to truncate the bitstream at any point, without endangering the decoding process. The advantages of such an embedded bitstream are spatial and rate-distortion scalability. Further improvement is obtained using anew adaptive wavelet transform known as wavelet packets. Contrary to earlier techniques, relevant statistical properties of the current subband are first analyzed. Dependent on that, the decomposition decision is made, whether the subband should be decomposed or not. This procedure yields not to a best-basis selection but to a near-optimal decomposition structure. The main advantage is the reduction of computational cost.

The need for effective medical image compression and transmission techniques continues to grow because of the huge volume of radiological images captured each year. The limited bandwidth and efficiency of current networking systems cannot meet this need. In response, Physical Optics Corporation devised an efficient medical image management system to significantly reduce the storage space and transmission bandwidth required for digitized medical images. The major functions of this system are: (1) compressing medical imagery, using a visual-lossless coder, to reduce the storage space required; (2) transmitting image data progressively, to use the transmission bandwidth efficiently; and (3) indexing medical imagery according to image characteristics, to enable automatic content-based retrieval. A novel scalable wavelet-based image coder was developed to implement the system. In addition to its high compression, this approach is scalable in both image size and quality. The system provides dramatic solutions to many medical image handling problems. One application is the efficient storage and fast transmission of medical images over picture archiving and communication systems. In addition to reducing costs, the potential impact on improving the quality and responsiveness of health care delivery in the US is significant.

One important field of interest in medicine and biology is the evaluation of DNA repair and cellular DNA damage after physical or chemical treatment. Manual analysis has some disadvantages such as a decrease in recognition ability during the time consuming observations as well as a requirement of experts for microscopic investigations. Therefore, automatic inspection and recognition of biological structures in several applications such as fluorescence in situ hybridization (FISH), fluorescence immuno-assays, comet-assay, chromosome karyotyping and micronucleus assay have been considerably advanced in the last decade. This presentation will give an overview of the image analysis and pattern recognition methods employed in different automated cell inspection systems which have been developed in our institute during the last years. Depending on the kind of assay, different experimental setups must be used in order to extract the respective measurement quantities. For example FISH technique requires a very sensitive fluorescence microscope combined with an image intensified target or time integrating camera. The major algorithms for image preprocessing and image segmentation based on mathematical morphology are briefly introduced. Feature classification is carried out with different methods.

A major problem in image processing and analysis is the segmentation of its components. Many computer vision tasks process image regions after segmentation, and the minimization of errors is then crucial for a good automatic inspection system. This paper presents an applied work on automatic segmentation of cell nuclei in digital noisy images. One of the major problems when using morphological watersheds is oversegmentation. By using an efficient homotopy image modification module, we prevent oversegmentation. This module utilizes diverse operations, such as sequential filters, distance transforms, opening by reconstruction, top hat, etc., some in parallel, some in cascade form, leading to a new set of internal and external cell nuclei markers. Very good results have been obtained and the proposed technique should facilitate better analysis of visual perception of cell nuclei for human and computer vision. All steps are presented, as well as the associated images. Implementations wee done in the Khoros system using the MMach toolbox.

Computed tomography fluoroscopy (CTF) has attracted significant attention recently. This is mainly due to the growing clinical application of CTF in interventional procedures, such as guided biopsy. Although many studies have been conducted for its clinical efficacy, little attention has been paid to the temporal response and the inherent limitations of the CTF system. For example, during a biopsy operation, when needle is inserted at a relatively high speed, the true needle position will not be correctly depicted in the CTF image due to the time delay. This could result in an overshoot or misplacement of the biopsy needle by the operator. In this paper, we first perform a detailed analysis of the temporal response of the CTF by deriving a set of equations to describe the average location of a moving object observed by the CTF system. The accuracy of the equations is verified by computer simulations and experiments. We show that the CT reconstruction process acts as a low pass filter to the motion function. As a result, there is an inherent time delay in the CTF process to the true biopsy needle motion and locations. Based on this study, we propose a generalized underscan weighting scheme which significantly improve the performance of CTF in terms of time lag and delay.

Contact lens' fitting evaluation is of critical importance in the contact lens' prescription process. For the correction of eye's refraction problems the use of contact lens' is very appealing to the user. However its prescription is far more demanding than the one of eye glasses. The fitting of a contact lens to a particular cornea must be carefully assessed in order to reduce possible user's physical miscomfort or even medical situations.The traditional way of easily checking the fitting of a contact lens is to perform a fluorescein test. The simple visual evaluation of the 'smoothness' of the color/brightness distribution of the fluorescence at the contact lens' location gives the optometrist an idea of the fitting's quality. We suggested the automation of the process simply by the substitution of the optometrist's eye by a CCD camera, and the use of appropriated simple image processing techniques. The setup and the digitalization and processing routines will be described in this communication. The processed images may then be directly analyzed by the optometrist in a faster, easier and more efficient way. However, it is also possible to perform an automated fitting evaluation by working out the information given by the image's intensity histograms for the green and blue RGB' channels.

Images are statistical in nature due to random changes and noise,therefore it is sometimes an advantage to treat image functions as a realization of a stochastic process. An advantage of stochastic random field models is that they need only a few parameters to describe a region or texture. In this paper textural images are modeled as a realization of Markov Random Fields such as binomial and autoregressive Markov random fields. Parameters of each model are estimated and considered as features of the textural images. The extracted features are incorporated in either a probabilistic neural network (PNN) or a deterministic back propagation neural networks for the purpose of classification and differentiation between various textural images. The PNN and the learning algorithm are discussed in this paper in details. To train back propagation neural network, a hybrid training algorithm is proposed. This hybrid algorithm takes advantage of both simulated annealing and deterministic learning algorithms. The former algorithm is more reliable since it locates a more likely global minimum, but it is slow. The latter algorithm is fast but less reliable, and it can converge to a local minimum. There are many practical uses for the above proposed textural analysis tool such as remote sensing, mineralogical analysis, and medical image processing. In this paper, successful applications of the present stochastic model to synthetic texture images and MRI tongue and brain images are described.

Among the various skin diseases skin tumors are the most serious ones and skin Melanoma is particularly dangerous. Its malignant evolution lasts about 5 or 6 years and ends with the death of the patient. Early diagnosis is a powerful means of preventing this evolution allowing sudden intervention, which increases probability or recover and survival. Aim of the paper is to present the result of an active support system for early diagnosis of melanoma and related skin diseases. The system is based upon a digital acquisition camera with a dedicated illumination system digitally controlled in order to achieve best performance in color and feature discrimination reaching best signal to noise ratio especially in blue band. A polarization framework allows for reflected ray rejection maximization. A new classification approach is presented. It allows for a quantification of morphological patterns and standard parameters in order to implement a computer aided dermatological system. The image information extraction is based on minimal descriptor set of parameters in order to classify chromatic texture and morphological features. The results obtained allow for determination of standard reference grids for pathological cases and reliable and objective classification procedure. We adopt, as reference, the approach used by Stanganelli and Kenet. Through a bioengineering analysis we can organize reference grids that offer the possibility to extract the maximum information content from dermatological data. The classification takes into account the spread and intrinsic descriptors and correspond to the best operative description. Therefore these grids are the more suitable tools for applications which requires active support system for diagnosis. In fact it is possible to obtain quantitative evaluations too. We propose a method based on geometrical synthetical descriptors. All that permits a reliable early diagnosis of melanotic disease and to follow its evolution in time. The first result and the incoming work points to the realization of a system for general dermatological applications.

Compression of digital image and video data plays an important role in reducing the transmission and storage cost for visual communication. However, when bit rates become very low, most compression algorithms yield visually annoying artifacts that highly degrade the perceptual quality of image and video data. To achieve high bit rate reduction while maintaining the best possible perceptual quality, postprocessing techniques provide one attractive solution. There has been a significant amount of research activities in post- processing of compressed image/video over the past decade. Some of them ar developed from the image enhancement point of view while others are from image restoration. The goal of this paper is to provide a review and analysis of recent developments in post-processing techniques. A discussion on compression artifacts is given first. Then , we survey several algorithms that reduce compression artifacts. Finally, we describe the current bottleneck and future directions in this field.

A new embedded wavelet coder called the multi-threshold wavelet coder (MTWC) was proposed in this research. The new coder was motivated by a careful analysis of the rate- distortion performance and the coding efficiency of existing embedded wavelet coders. Unlike other proposed embedded wavelet coders which use a single initial threshold in their successive approximate quantization, MTWC uses different initial thresholds in different subbands and applies a subband decision scheme to decide which subband should be coded in sequence. It is shown with experimental results that MTWC has a better rate-distortion tradeoff at higher bit rates and a lower computational complexity than other embedded wavelet coders. Furthermore, MTWC can be easily applied to an image coder with the wavelet packet transform, whose performance is significantly better than their age coder with the pyramid wavelet transform for texture-dominant imagery.

The impact of lossy compression was considered for transmission of environmental image data. In the investigated implementation scenarios, it was found that compression at rates approaching 16:1 has minor impact on the exploitation and assessment of the ultimate derived automated cloud analysis, due to the robustness of this algorithm. Additional work is needed to evaluate the impact of compression on other products, such as sea surface temperature. The implementation included high dynamic range satellite imagery.

An adaptive overlapped block motion compensation (OBMC) scheme, which takes the rate, distortion and complexity factors into account, is proposed in this research. With this scheme, the encoder uses the classification information to determine which blocks could be distorted after OBMC. Then, the BMC search rather than the OBMC search is applied to these blocks. Depending on the allowed computational complexity of the decoder, the encoder chooses the best set of OBMC macroblocks which requires a higher complexity. Other macroblocks simply adopt BMC. Furthermore, the motion bit rat can be reduced by soothing the motion field of BMC blocks during motion search in the encoder. An effective group OBMC motion search scheme is applied to this new adaptive OBMC motion scheme. Experimental results show that the complexity of both the encoder and the decoder is reduced while the visual quality remains the same due to ten improvement in the oversmooth region.

The generalized recursive interpolation (GRINT) algorithm was recently proposed and shown to be the most effective progressive technique for decorrelation of still image. A nonlinear version of GRINT (MRINT) employs median filtering in a nonseparable fashion on a quincunx grid. The main advantage of both these schemes is that interpolation is performed from all error-free values, thereby reducing the variance of interpolation errors. MRINT is embedded in a simplified version of the context-based encoder by Said and Pearlman. Coding performances of the novel context-based coder are evaluated by comparisons with GRINT, and a variety of other multiresolution lossless methods, including the original scheme by Said and Pearlman. The modified scheme outperforms all the other algorithms, including the latter, especially when dealing with medical images.

Classical approaches of video scalability may produce some problems for medium or high cell loss ration and when traffic load on the transmission network is rapidly varying In this paper, the standard MPEG-2 video codec is modified to manage packet losses during the transmission over an ATM network. A layered coding scheme is adopted, which sues an adaptive data- partition method to generate two compressed video layers to be transmitted with different priority. The MPEG-2 slice structure of the upper layer is completely redefined and a new synchronization procedure between the two layers is introduced. The experimental results prove the superiority of the two-layer model with respect to the standard approach, particularly for increasing values of cell loss probability.

The purpose of this paper is to compare low complexity LEI padding and MxN DCT method with 'macroblock-padding' technique for encoding the textures around the boundary blocks of an object. For this experiment, MxN DCT and LEI are applied to the boundary blocks of a video object plane. And then, the performance of two techniques will be compared separately with those of repetitive padding.

In this paper, we introduce the updated polygonal representation method for coding binary shape information. The geometric representation method supports a wide range of the requirements for natural and synthetic video. In addition to providing high coding efficiency, this method is unique in its effectiveness at addressing a number of other requirements. These unique capabilities are due to its geometric nature, its inherent bitstream scalabilities, and its highly-efficient intra coding more. And, we can obtain noticeable improvement in terms of objective and subjective quality by applying some algorithms.

In this paper we extend a segmentation method aimed at separating the moving objects from the background in a generic video sequence by means of a higher order statistics (HOS) significance test performed on a group of inter-frame differences. The test is followed by the motion detection phase, producing a preliminary binary segmentation map, that is refined by a final regularization stage. The HOS threshold and the temporal extent of the motion detection phase are adaptively changed on the basis of the estimated background activity and of the detected presence of slowly moving objects. The regularization phase, imposing a local connectivity constraint on the background-foreground map by basic morphological operators, plays an important role in eliminating misclassifications due to motion estimation ambiguities, of the original video sequence. The algorithm performance is illustrated by typical results obtained on MPEG4 sequences.

Pre-processing images prior to encoding can remove noise, or unimportant detail, and thus improve the overall performance of the coder. Typical objective image quality metrics are obtained by computing a single or several numbers as a function of the difference image between the original and coded images. Such metrics may ont reflect the improvement in performance. In a recent paper we have presented a methodology that allows the quantitative determination of image quality when the image has been processed prior to encoding. We now present an extension of the work showing global objective quality measures that quantify the value of pre-processing for image coding using a wavelet coder. Because many options are available in wavelet coder design, we limit our study to a 'best' coder obtained in previous work, and determine what further performance improvement can be achieved by image processing.

We have developed a new architecture for computer vision which is a multi-level, MIMD machine. It is a homogenous system where the control unit, processing arrays and image store are all built using the same type of digital signal processor (DSP). DSPs provide satisfactory performance at low-level and intermediate-level image processing and make the construction of a homogenous system possible the processing array of the system consists of three linear arrays. Each linear array is dedicated to one image processing type. The system has a modular design and it can be expanded easily. The overall system design utilizes distributed memory MIMD paradigm. The system is benchmarked with some low- and immediate-level image processing algorithms such as convolution, edge finding. Hough transform etc. and results are presented.

This paper describes the development of a novel automated and efficient vision system to obtain velocity and concentration measurement within a porous medium. An aqueous fluid lace with a fluorescent dye to microspheres flows through a transparent, refractive-index-matched column packed with transparent crystals. For illumination purposes, a planar sheet of laser passes through the column as a CCD camera records all the laser illuminated planes. Detailed microscopic velocity and concentration fields have been computed within a 3D volume of the column. For measuring velocities, while the aqueous fluid, laced with fluorescent microspheres, flows through the transparent medium, a CCD camera records the motions of the fluorescing particles by a video cassette recorder. The recorded images are acquired automatically frame by frame and transferred to the computer for processing, by using a frame grabber an written relevant algorithms through an RS-232 interface. Since the grabbed image is poor in this stage, some preprocessings are used to enhance particles within images. Finally, these enhanced particles are monitored to calculate velocity vectors in the plane of the beam. For concentration measurements, while the aqueous fluid, laced with a fluorescent organic dye, flows through the transparent medium, a CCD camera sweeps back and forth across the column and records concentration slices on the planes illuminated by the laser beam traveling simultaneously with the camera. Subsequently, these recorded images are transferred to the computer for processing in similar fashion to the velocity measurement. In order to have a fully automatic vision system, several detailed image processing techniques are developed to match exact images that have different intensities values but the same topological characteristics. This results in normalized interstitial chemical concentrations as a function of time within the porous column.

A machine-vision sorting system was developed that utilizes the difference in light reflectance of fruit surfaces to distinguish the defective and good apples. To accommodate to the spherical reflectance characteristics of fruit with curved surface like apple, a spherical transform algorithm was developed that converts the original image to a non-radiant image without losing defective segments on the fruit. To prevent high-quality dark-colored fruit form being classified into the defective class and increase the defect detection rate for light-colored fruit, an intensity compensation method using maximum propagation was used. Experimental results demonstrated the effectiveness of the method based on maximum propagation and spherical transform for on-line detection of defects on apples.

This paper deals with the recovery of a scene from a pair of images, where each image is acquired from a different viewpoint. The central problem is the identification of corresponding points in all views. We use the feature-based approach to find corresponding points.Various types of features have been sued previously, where Gabor features showed significant advantages in terms of accuracy and the complexity/accuracy trade-off. The accuracy is measured as the rate of correctly associated pixels The accuracy measures are found by comparing the disparity maps produced by the matching program with the correct disparities. These correct disparities must be known, and are typically produced by expensive photogrammetric techniques. In this paper we show a method of gauging the performance of a stereo by expensive photogrammetric techniques. In this paper we show a method of gauging the performance of a stereo matcher without the necessity of such a reference disparity date set. We show that statistics on the back-matching distances can be used instead. these are a by- product of the matching process. This opens the door to extensive testing and optimization, since we no longer have to rely on the existence of the reference disparities.

Recent efforts in image morphing research aim at improving both user interface and warping results. To specify feature points in tow images, the user interface takes up much time and it allows the warping specification by the user which represents very tedious work. In this paper, we propose a semi-automatic algorithm based on active contour model to specify the feature correspondence between two given images. It allows a user to extract a contour that defines a facial features such as lips, mouth, profile, etc., by specifying only endpoints of the contour around the feature that serve as the extremities of a contour. The proposal algorithm uses these two points as anchor points, and automatically computes the image information around these endpoints to provide boundary conditions. Then we optimize the contour by taking this information into account close to its extremities. During the iterative optimization process, the image forces are moving progressively from the contours extremities towards its center to define the feature. Once the feature correspondence points are paired, the intermediate images are generated by interpolating the positions of feature points linearly. The proposal algorithm helps the user to define easily the exact position of a feature. It may also reduce the time taken to establish feature correspondence between two images.

An investigation is under way to develop a control system for an industrial process which uses a vision systems as a sensor. The research is aimed at the improvement of product quality in commercial injection molding system. A significant enhancement has been achieved in the level of application of visually based inspection techniques to component quality. The aim of the research has been the investigation, and employment, of inspection methods that use knowledge based machine vision. The application of such techniques in this context is comprehensive, extending from object oriented analysis, design and programming of the inspection program, to the application of rule based reasoning, to image interpretation, vision system diagnostics, component diagnostics and molding machine control. In this way, knowledge handling methods are exploited wherever they prove to be beneficial. The vision knowledge base contains information on the procedures required to achieve successful identification of component surface defects. A collection of image processing and pattern recognition algorithms are applied selectively. Once inspection of the component has been performed, defects are related to process variables which affect the quality of the component, and another knowledge base is used to effect a control action at the molding machine. Feedback from other machine sensor is also used to direct the control procedure. Results from the knowledge based vision inspection system are encouraging. They indicate that rapid and effective fault detection and analysis is feasible, as is the verification of system integrity.

Large power stations are usually chosen as the target for missiles guided by IR seeker. This paper presents an aimpoint finding algorithm. The high light blob of boiler and straight lines of chimneys or water cooling towers are taken as the reference feature objects for locating and keeping the aimpoint form long to short distance. Some experiment results are given to show the validity of our method.

A new hardware and software platform based on Windows for processing image and graphic information is developed, some advanced external equipment such as CCD camera, color scanner, printer and plotter are integrated together under the platform. The software package developed with Borland c++ includes four kinds of software for image and graph processing, an advanced text/image management information library, a GUI and an external control program. The platform may be used to perform the image input for the actual objects and plane diagram, processing for the complicated image, and specially, image and graphic output with large size.

This paper describes a real-time correlation tracker realized with field programmable gate array (FPGA). Primarily a method for correlation tracking target has been developed. The main point of this method is to present the reference pattern updating methods we implemented. Then, information on system hardware is given. The correlation tracking circuit is realized by FPGA so autotracker is small, low cost, and it could be used for real time application. Subsequently, system software can be found. System operation is controlled by high speed digital signal processor TMS320C30. In the end, the result of field-tests with our real-time correlation autotracker show that under complex background and foreground, such as clouds, trees, grasses, our approach can track a flying plane or a moving vehicle steadily.

In this paper, we will discuss the issue of recognizing the runway target in IR images and choosing the aimpoint. At first, we detect out the target from complicated background, Then we use tow characteristics to recognize it and choose the optimal aimpoint. Some results are presented. From the experiment, our method operates successfully.

This paper presents an image analysis by basis pursuit, such analysis, depending the dictionary employed, can provide an image multi-scale edge representation such as the Mallat and Hwang's representation. When a stationary dictionary is employed while processing images by basis pursuit, the coefficients selected from the dictionary can characterize very well the most important image features, because of this well characterization we obtain directly an image edge representation from the coefficients selected. According to the results presented in this paper, we can say processing images by basis pursuit provides a new alternative to multi-scale image edge detection which should be exploited and expanded to other image processing applications such as image compression, de-noising, segmentation, image enhancement or image interpretation.

Endoscopy has become a crucial diagnostic and therapeutic procedure in clinical areas. Over the past four years, we have developed a computerized system to record and store clinical data pertaining to endoscopic surgery of laparascopic cholecystectomy, pelviscopic endometriosis, and surgical arthroscopy. In this study, we developed a computer system, which is composed of a frame grabber, a sound board, a VCR control board, a LAN card and EDMS. Also, computer system controls peripheral instruments such as a color video printer, a video cassette recorder, and endoscopic input/output signals. Digital endoscopic data management system is based on open architecture and a set of widely available industry standards; namely Microsoft Windows as an operating system, TCP/IP as a network protocol and a time sequential database that handles both images and speech. For the purpose of data storage, we used MOD and CD- R. Digital endoscopic system was designed to be able to store, recreate, change, and compress signals and medical images. Computerized endoscopy enables us to generate and manipulate the original visual document, making it accessible to a virtually unlimited number of physicians.

We present a method for detecting concentric strias which can be found in different natural 'objects'. One of the major problems encountered during an automatic image processing is the lack of continuity perception in strias. We propose an approach to this continuity perception based on a multiagent system made up of reactive agents. These agents can move around on their environment which consists of an image made up of light and dark rings set out concentrically. Our multiagent system is made up of a set of agents named darkening agents and lightening agents. These agents follow either the light rings or the dark rings and act on the image. Their actions aim to reinforce the rings by stressing the contrasts allowing, thus, a reliable detection of these rings, even if they are discontinuous. Each agent has three sensors allowing it to obtain information about the environment. The sensor ar made up of unit sensors returning the value of a pixel. The three sensors of an agent are: (1) a unit sensor allowing the agent to know if it is located on an already detected stria, (2) two disk-shaped sensors made up of unit sensors. These two sensors return the sum of their unit sensors. They are located in front of the agent and distant one from the other. Those two disk-shaped sensors are used for determining the movements of the agent and, therefore, for detecting the rings. An agent, indeed, always tries to move where the values returned by the senors is minimal. The suggested multiagent system has shown great robustness and adaptability, for it can detect strias, even if they are discontinuous and the image noisy.

Ocean surfaces should be considered as randomly rough surfaces at the IR and optical bands. The geometric model of sea surfaces has been obtained by using image processing. Based on the electromagnetic scattering theory, the Laser Radar Cross Section for the sea surfaces is analyzed at IR band and the scattering properties of the sea surfaces have been gotten.

Since the turn of the century there has been widespread acknowledgement that Alphonse Mucha inspired the birth of the Art Nouveau movement when he created the 'Gismonda' poster advertising the opening of the Sarah Bernhardt play of that name in Paris in 1894. At an estate sale in 1954 a small collage bearing a likeness of the Mucha 'Gismonda' was offered. It was composed of fragments of sixty postage stamps glued to a small ceramic tile. Digital computer image processing has been applied to the collage design, scratches on the handle of a walking stick in the same collection, and the Mucha poster. From comparative analyses of the enhanced 'Gismonda' images it is revealed that the little collage is considerably more detailed and compete than the Mucha 'original'. Thus, it is concluded that the poster is a hasty photographic plagiarism of the intricate collage. Further image processing of the scratches on the handle of the walking stick and the collage reveal them to conform to the famous and enigmatic 'P GO' monogram signature of the artist Paul Gauguin. Thus, it follows that the original design was created by Gauguin rather than by Mucha. It may be that while Gauguin was in Brittany recovering from injuries sustained in a brawl his former lover, Annah la Javanese, pilfered his belongings and took them with her to Paris and her next lover, photographer-designer, Alphonse Mucha, who copied the collage and offered it as the Gismonda Poster.

In this paper we report the application of neural trees for image segmentation of magnetic resonance (MR) images. The network, built up during training, effectively partitions the feature space into subregions and each final subregion is assigned a class label according to the data routed to it. As the tree grows, the number of training data for each node decreases, which results in less weight update epochs and decreases the time consumption. The growing algorithm is based on depth-first search, which is guaranteed to find deep solutions, i.e. linearly non-separable classes. We also introduce a measure for estimation of the best-fit neuron to split the feature space at each tree node. This eliminates the necessity for postponing perturbance of the hyperplanes and proves essential for solving linearly non-separable difficult tasks. The network performance is compared to the multilayered perceptron (MLP) over the white/gray matter MRI segmentation problem.

Turbulence conditions affect video images in tow ways. It causes local blur, and it distorts the geometry of the scene. A video sequence of a still scene recorded under turbulence, appears to contain local random motion of small neighborhoods in the images. The blur is an accumulated results of the imaging point spread function and the local motion. The geometric distortion is due to the fact that small neighborhoods move in different directions. The restoration scheme reported here takes care of the geometric distortion as well as the blur. The geometric distortion is reduced by averaging the gray levels of relatively long video segments. The averaging reduces the geometric distortion, but it increases the blur. The second stage is the estimation of the global point spread function. The blur in the average image is a combination of the effects of the imaging system transfer function,the turbulence, and the averaging of the sequence. The global non-isotropic point spread function is estimated based on edge responses in the average image. A Wiener filter is used for the restoration of the image. The presented experimental result are superior to the results obtained by a previously proposed majority-vote technique.

We studied the influence of an aperture shape of a diaphragm on the restoration of low quality out-of-focus video images. We measured PSF of three consumer video cameras which are equipped with different types of the diaphragm system. In this measurements, the appropriateness of the assumption of space-invariant PSF and the aperture-size-dependence of the geometrical shape of PSF were examined. Although rotationally symmetric PSF has been assumed in many studies, it was recognized that the shape of PSF is not circular but triangular or rhomboid in many cases. Moreover, the shape depends both on the camera type and the aperture opening ratio. We applied the measured PSF to the restoration of the blurred images taken by the home video cameras. The Wiener filtering technique was used to restore the images. More desirable images were obtained by approximating PSF to two triangular disks on the basis of the measurement than by approximating PSF to a circular disk or a gaussian function.

Recent industry efforts have focused on the Balanced Tree Clustering (BTC) approach to speed up the encoding process for fractal image compression. Since the BTC algorithm compares intensity pixel by pixel for blocks in a mother cluster, clocks are divided into different clusters when they have different orientations or when they have different amplitudes of intensity offset with the averaged values but otherwise the same intensity distribution, causing poor clustering. In this paper we present and discuss the improvement of the original BTC algorithm by devising a method to include all 8 isometric versions of a block without enlarging the size of the domain pool and without sacrificing the encoding time. The results of compression are compared using the BTC algorithm based on the three different pixel values, intensity, intensity offset by the average value, and intensity variance, with and without considering 8 isometries of each block. For a sample image of 256 X 256 X 8 with 4 X 4 block size and with about 10 blocks in each cluster, the BTC using intensity variance and considering the 8 isometries the encoding speed about 200 times while decreases PSNR less than 8 percent over the ordered quadrant intensity classification into three classes. The BTC based on intensity variance also improves the encoding time and the fidelity of compression over the BTC based on the other two pixel values. The BTC algorithm based on the three different pixel values also generate different fractal code books which may affect the utility of the code books for image classification and image matching.

Mathematical morphology (MM) is a very efficient tool for image processing, based on non- linear operators.In this paper MM is applied to extract the image's features. As a feature we understand specific information about the image i.e. location, size, orientation of certain image elements. Morphological operators are applied to find and measure objects on the image's surface. Two practical examples are considered. First is devoted to analysis of binary images, containing printed characters. Characters are separated and MM is used to extract some information from each character. These features are later measured and included in a feature vector. It contains the special kind of information - the number of elements of the character with its shape modified in different ways. Second examples shows how feature extraction by MM works on graytone images. Images for analysis contain human faces. Morphological operators extract some important elements of human face. This information is very important to identify the human face. Experiments show us how the morphological operators can be applied to the feature detection. The simplest operators as erosion and dilation, as well as more sophisticated tools like: morphological filtering, geodesic transformations are used for that purpose. Also directional operations are applied to extract some areas. This paper includes algorithms for feature extraction by MM, as well as the brief description of morphological tools, explication of experiments and the results of them.

A popular approach to the qualitative analysis of thermal patterns has been to identify anomalies through comparison of thermal images against a single baseline or reference image. However, this approach represents an oversimplification as significant variations of thermal patterns due to change in measurement position, changing equipment loading, environmental conditions and varying mechanisms of equipment deterioration are not catered for. To overcome these limitations, the use of neural net and statistically based classifiers has been investigated, in the latter case for both parametric and non parametric designs. An experimental thermal image database characterizing normal and abnormal load tap-changer operation of a 63 MVA, 22kV transformer provided the training data. The images were captured at different times, different locations and under varying loads.

Jay DeFeo's sculptural-painting was originally begun as a 'concept with a center'. This DeFeo work was pivotal to both her artistic recognition and development It epitomized the beat- culture's newly discovered creative freedoms encouraged during the art renaissance of the period that has come to be known as America's Chamelot. Jay DeFeo's Rose was the culmination of an almost ritualized performance-art process of creation and destruction that spanned the years 1958 through 1965 and resulted in a 0.4 X 2.5 X 3.2 meter, 100 kilogram composite-laminate painting/artifact. When the artwork begun to crumble twenty years ago it was totally encased in approximately 10cm of steel-reinforced plaster for structural support. Here we will describe the solids-imaging techniques considered and utilized in determining the painting's internal composition to aid in safe extraction of the artwork from its plaster tomb followed by complete restoration. Our imaging goals were first, locate the interior plaster-paint boundary and second, characterize the size, shape and relative geometry of voids and artifacts in the paint composite laminate to facilitate the positioning and embedding weight-bearing pins.

We outline a novel method for the estimating a fixed aberration that is in the image path, but not in the wavefront sensor (WFS) path of an adaptive optics (AO) imaging system. We accomplish this through a non-traditional application of the Gonsalves least-squares phase diversity technique using an ensemble of images and WFS data. The diversity phases required for the Gonsalves technique are provided by the temporal differences in WFS residual phase measurements for different members of the ensemble. The technique is demonstrated using actual observations from an operational AO system exhibiting such aberration. An estimate of this aberration was obtained using the proposed algorithm which agrees reasonably well with the observed point-spread function.

Image intensifiers on board UVSTAR, the Ultraviolet Spectrograph Telescope for Astronomical Research, did not seem to be functioning as expected. Therefore, a simplified version of the UVSTAR electronics was used to test the image intensifiers in the laboratory. All photosensitive locations of he intensifier array were exposed at once, and then each line read each sequentially. The images which were produced had an annoying intensity gradient superimposed on them. This gradient is due to the fact that line n of the intensifier array was corrupted by thermal signal n times longer than line 1. The imagery was further degraded by the presence of so-called 'hot-pixels'. These pixels correspond to locations on the intensifier's charge coupled device array which over react to the photon flux and produce abnormally high intensity readings. This paper descries image processing techniques which removed the gradient as well as hot pixels from the intensifier imagery so that an accurate performance assessment could be made. The methods describe can also be used to diagnose flaws in medical and industrial image intensifiers in general.