In recent years, the acquisition of image and video information for processing, analysis, understanding, and exploitation of the underlying content in various applications, ranging from remote sensing to biomedical imaging, has grown at an unprecedented rate. Analysis by human observers is quite laborious, tiresome, and time consuming, if not infeasible, given the large and continuously rising volume of data. Hence the need for systems capable of automatically and effectively analyzing the aforementioned imagery for a variety of uses that span the spectrum from homeland security to elderly care. In order to achieve the above, tools such as image segmentation provide the appropriate foundation for expediting and improving the effectiveness of subsequent high-level tasks by providing a condensed and pertinent representation of image information. We provide a comprehensive survey of color image segmentation strategies adopted over the last decade, though notable contributions in the gray scale domain will also be discussed. Our taxonomy of segmentation techniques is sampled from a wide spectrum of spatially blind (or feature-based) approaches such as clustering and histogram thresholding as well as spatially guided (or spatial domain-based) methods such as region growing/splitting/merging, energy-driven parametric/geometric active contours, supervised/unsupervised graph cuts, and watersheds, to name a few. In addition, qualitative and quantitative results of prominent algorithms on several images from the Berkeley segmentation dataset are shown in order to furnish a fair indication of the current quality of the state of the art. Finally, we provide a brief discussion on our current perspective of the field as well as its associated future trends.

Crosstalk, also known as ghosting or leakage, is a primary factor in determining the image quality of stereoscopic three dimensional (3D) displays. In a stereoscopic display, a separate perspective view is presented to each of the observer's two eyes in order to experience a 3D image with depth sensation. When crosstalk is present in a stereoscopic display, each eye will see a combination of the image intended for that eye, and some of the image intended for the other eye-making the image look doubled or ghosted. High levels of crosstalk can make stereoscopic images hard to fuse and lack fidelity, so it is important to achieve low levels of crosstalk in the development of high-quality stereoscopic displays. Descriptive and mathematical definitions of these terms are formalized and summarized. The mechanisms by which crosstalk occurs in different stereoscopic display technologies are also reviewed, including micropol 3D liquid crystal displays (LCDs), autostereoscopic (lenticular and parallax barrier), polarized projection, anaglyph, and time-sequential 3D on LCDs, plasma display panels and cathode ray tubes. Crosstalk reduction and crosstalk cancellation are also discussed along with methods of measuring and simulating crosstalk.

Visual tracking can be treated as an optimization problem. A new meta-heuristic optimal algorithm, Harmony Search (HS), was first applied to perform visual tracking by Fourie et al. As the authors point out, many subjects are still required in ongoing research. Our work is a continuation of Fourie's study, with four prominent improved variations of HS, namely Improved Harmony Search (IHS), Global-best Harmony Search (GHS), Self-adaptive Harmony Search (SHS) and Differential Harmony Search (DHS) adopted into the tracking system. Their performances are tested and analyzed on multiple challenging video sequences. Experimental results show that IHS is best, with DHS ranking second among the four improved trackers when the iteration number is small. However, the differences between all four reduced gradually, along with the increasing number of iterations.

A context-adaptive-variable-length-coding (CAVLC) decoder cannot determine an exact start position of the k'th syntax element until it decodes the (k - 1)'th syntax element, which makes parallel CAVLC decoding difficult. We propose a new bitstream structure to maximize parallelism of CAVLC decoding. Our algorithm enables us to simultaneously access multiple points from different MBs in a bitstream. Then, a CAVLC decoder can concurrently read multiple symbols from the multiple access points and decode them in parallel. Experimental results show that the proposed algorithm significantly increases decoding speed without sacrificing coding efficiency.

A methodology for automatic facial expression recognition in image sequences is proposed, which makes use of the Candide wire frame model and an active appearance algorithm for tracking, and support vector machine (SVM) for classification. A face is detected automatically from the given image sequence and by adapting the Candide wire frame model properly on the first frame of face image sequence, facial features in the subsequent frames are tracked using an active appearance algorithm. The algorithm adapts the Candide wire frame model to the face in each of the frames and then automatically tracks the grid in consecutive video frames over time. We require that first frame of the image sequence corresponds to the neutral facial expression, while the last frame of the image sequence corresponds to greatest intensity of facial expression. The geometrical displacement of Candide wire frame nodes, defined as the difference of the node coordinates between the first and the greatest facial expression intensity frame, is used as an input to the SVM, which classify the facial expression into one of the classes viz happy, surprise, sadness, anger, disgust, and fear.

This paper presents a new model of a complementary metal-oxide-semiconductor (CMOS) camera using combinations of several pin hole camera models, and its validity is verified by using synthesized stereo images based on OpenGL software. Our embedded three-dimensional (3-D) image capturing hardware system consists of five motor controllers and two CMOS camera modules based on an S3C6410 processor. An optimal alignment for capturing nine segment images that have their own convergence planes is implemented using a pi controller based on the measures of alignment and sharpness. A new synthesizing fusion with the optimized nine segmentation images is proposed for the best 3-D depth perception. Based on the experimental results of the disparity values in each of the nine segments, the multi-segment method proposed in this paper is a good method to improve the perception of 3-D depth in stereo images.

The iterative regularization method proposed by Osher et al. for total variation based image denoising can preserve textures well and has received considerable attention in the signal and image processing community in recent years. However, the iteration sequence generated by this method converges monotonically to the noisy image, and therefore this iteration must be terminated opportunely with an "optimal" stopping index, which is difficult in practice. To overcome this shortcoming, we propose a novel fractional-order iterative regularization model by introducing the fractional-order derivative. The new model can be considered as an interpolation between the traditional total variation model and the traditional iterative regularization model. Numerical results demonstrate that with a fitting order of derivative, the denoised image sequence generated by this model can converge to a denoised image with high peak signal to noise ratio and high structural similarity index after a few iteration steps, and therefore we can terminate the iteration according to some most used termination conditions. Moreover, we propose an experience method to choose the order of derivative adaptively for the partly textured images to improve the performance of noise removal and texture preservation. The adaptive method has low computational cost and can improve the result visually efficiently.

Scalable video coding and multiple description coding are the two different adaptation schemes for video transmission over heterogeneous and best-effort networks such as the Internet. We propose a new method to encode video for unreliable networks with rate adaptation capability. Our proposed method groups three dimensional discrete wavelet transform coefficients in different descriptions and applies a modified embedded zero tree data for rate adaptation. The proposed method optimizes the bit-rates of the descriptions with respect to the channel bit rates and the maximum acceptable distortion. The experimental results in the presence of one description loss indicate that on average the videos at the rate of 1000 Kbit/s are reconstructed with Y-component of peak signal to noise ratio (Y-PSNR) value of 36.2 dB. The dynamic allocation of descriptions to the network channels is optimized for rate distortion minimization. The improvement in term of Y-PSNR achieved by rate distortion optimization has been between 0.7 and 5.3 dB in different bit rates.

In recent years, gradient-domain methods have been widely discussed in the image processing field, including seamless cloning and image stitching. These algorithms are commonly carried out by solving a large sparse linear system: the Poisson equation. However, solving the Poisson equation is a computational and memory intensive task which makes it not suitable for real-time image editing. A new matrix decomposition graphics processing unit (GPU) solver (MDGS) is proposed to settle the problem. A matrix decomposition method is used to distribute the work among GPU threads, so that MDGS will take full advantage of the computing power of current GPUs. Additionally, MDGS is a hybrid solver (combines both the direct and iterative techniques) and has two-level architecture. These enable MDGS to generate identical solutions with those of the common Poisson methods and achieve high convergence rate in most cases. This approach is advantageous in terms of parallelizability, enabling real-time image processing, low memory-taken and extensive applications.

We incorporate, evaluate, and assess the feasibility of using filter banks in automated pavement distress systems from a system level. We integrate a novel filter-bank-based distress segmentation method, which, unlike previously researched methods, does not depend on highpass data. In addition, we incorporate the standard Said Pearlman set partitioning in hierarchical trees compression coder into the automated pavement distress system, which is a first in this area of research. A third contribution of the research is a statistical detection algorithm that assists in overall system performance. Preliminary testing using images provided by the Georgia Department of Transportation demonstrate the promise of the proposed method.

Speckle reduction is a difficult task for ultrasound image processing because of low resolution and contrast. As a novel tool of image analysis, quaternion wavelet (QW) has some superior properties compared to discrete wavelets, such as nearly shift-invariant wavelet coefficients and phase-based texture presentation. We aim to exploit the excellent performance of speckle reduction in quaternion wavelet domain based on the soft thresholding method. First, we exploit the characteristics of magnitude and phases in quaternion wavelet transform (QWT) to the denoising application, and find that the QWT phases of the images are little influenced by the noises. Then we model the QWT magnitude using the Rayleigh distribution, and derive the thresholding criterion. Furthermore, we conduct several experiments on synthetic speckle images and real ultrasound images. The performance of the proposed speckle reduction algorithm, using QWT with soft thresholding, demonstrates superiority to those using discrete wavelet transform and classical algorithms.

In solving pattern-recognition tasks with partially labeled training data, the semi-supervised dimensionality reduction method, which considers both labeled and unlabeled data, is preferable for improving the classification and generalization capability of the testing data. Among such techniques, graph-based semi-supervised learning methods have attracted a lot of attention due to their appealing properties in discovering discriminative structure and geometric structure of data points. Although they have achieved remarkable success, they cannot promise good performance when the size of the labeled data set is small, as a result of inaccurate class matrix variance approximated by insufficient labeled training data. In this paper, we tackle this problem by combining class membership probabilities estimated from unlabeled data and ground-truth class information associated with labeled data to more precisely characterize the class distribution. Therefore, it is expected to enhance performance in classification tasks. We refer to this approach as probabilistic semi-supervised discriminant analysis (PSDA). The proposed PSDA is applied to face and facial expression recognition tasks and is evaluated using the ORL, Extended Yale B, and CMU PIE face databases and the Cohn-Kanade facial expression database. The promising experimental results demonstrate the effectiveness of our proposed method.

Within a pixel in a digital imager, generally either a charge-coupled device or complementary metal oxide semiconductor device, doping of the semiconductor substrate and application of gate voltages create a region free of mobile carriers called the depletion region. This region fills with charge after incoming photons or thermal energy raise the charges from the valence to the conduction energy band. As the signal charge fills the depletion region, the electric field generating the region is altered, and the size of the region is reduced. We present a model that describes how this dynamic depletion region, along with the location of impurities, will result in pixels that produce less dark current after being exposed to light and additionally show nonlinear production rates with respect to exposure time. These types of effects have been observed in digital imagers, allowing us to compare empirical data with the modeled data.

Interpolation is an essential and broadly employed function of signal processing. Accordingly, considerable development has focused on advancing interpolation algorithms toward optimal accuracy. Such development has motivated a clear shift in the state-of-the art from classical interpolation to more intelligent and resourceful approaches, registration-based interpolation for example. As a natural result, many of the most accurate current algorithms are highly complex, specific, and computationally demanding. However, the diverse hardware destinations for interpolation algorithms present unique constraints that often preclude use of the most accurate available options. For example, while computationally demanding interpolators may be suitable for highly equipped image processing platforms (e.g., computer workstations and clusters), only more efficient interpolators may be practical for less well equipped platforms (e.g., smartphones and tablet computers). The latter examples of consumer electronics present a design tradeoff in this regard: high accuracy interpolation benefits the consumer experience but computing capabilities are limited. It follows that interpolators with favorable combinations of accuracy and efficiency are of great practical value to the consumer electronics industry. We address multidimensional interpolation-based image processing problems that are common to consumer electronic devices through a decomposition approach. The multidimensional problems are first broken down into multiple, independent, one-dimensional (1-D) interpolation steps that are then executed with a newly modified registration-based one-dimensional control grid interpolator. The proposed approach, decomposed multidimensional control grid interpolation (DMCGI), combines the accuracy of registration-based interpolation with the simplicity, flexibility, and computational efficiency of a 1-D interpolation framework. Results demonstrate that DMCGI provides improved interpolation accuracy (and other benefits) in image resizing, color sample demosaicing, and video deinterlacing applications, at a computational cost that is manageable or reduced in comparison to popular alternatives.

To develop a no-reference video quality model, it is important to know how the perceived strengths of artifacts are related to their physical strengths and to the perceived annoyance. When more than one artifact is present, it is important to know whether and how its corresponding perceived strength depends on the presence of other artifacts and how perceived strengths combine to determine the overall annoyance. We study the characteristics of different types of artifacts commonly found in compressed videos. We create artifact signals predominantly perceived as blocky, blurry, ringing, and noisy and combine them in various proportions. Then, we perform two psychophysical experiments to independently measure the strength and overall annoyance of these artifact signals when presented alone or in combination. We analyze the data from these experiments and propose models for the overall annoyance based on combinations of the perceptual strengths of the individual artifact signals. We also test the interactions among different types of artifact signals. The results provide interesting information that may help the development of video quality models based on artifact measurements.

A novel method for local image descriptor called a Haar-like compact local binary pattern (HC-LBP) is presented, which is robust to Gaussian noise and illumination changes and reduces the dimension of features by modifying the local binary pattern (LBP) method in two ways. First, the binary Haar-like feature method is applied to calculate region-based rank order. The binary Haar-like feature method maintains only the ordinal relationship and not the difference of intensity. Using a region-based binary operation, the proposed local image descriptor becomes more robust to Gaussian noise and illumination changes. Second, a HC-LBP feature is generated by applying the binary Haar-like feature according to the edge class in order to reduce feature dimensions and increase the discriminating power of each feature. In the feature matching experiment, the proposed method outperforms the popular scale invariant feature transform, center-symmetric local binary pattern and center-symmetric local ternary pattern in the presence of illumination changes, Gaussian noise, image blurring, and viewpoint change. Compared with popular local feature descriptors, the proposed method could get more than 0.1 increases in recall. Also, the HC-LBP descriptor could be used for many computer vision applications such as image retrieval, face recognition, and texture recognition.

We investigated a parametric kernel-driven active contour (PKAC) model, which implicitly transfers kernel mapping and piecewise constant to modeling the image data via kernel function. The proposed model consists of curve evolution functional with three terms: global kernel-driven and local kernel-driven terms, which evaluate the deviation of the mapped image data within each region from the piecewise constant model, and a regularization term expressed as the length of the evolution curves. In the local kernel-driven term, the proposed model can effectively segment images with intensity inhomogeneity by incorporating the local image information. By balancing the weight between the global kernel-driven term and the local kernel-driven term, the proposed model can segment the images with either intensity homogeneity or intensity inhomogeneity. To ensure the smoothness of the level set function and reduce the computational cost, the distance regularizing term is applied to penalize the deviation of the level set function and eliminate the requirement of re-initialization. Compared with the local image fitting model and local binary fitting model, experimental results show the advantages of the proposed method in terms of computational efficiency and accuracy.

Conventional threshold decomposition-based multilevel halftoning algorithms decompose an input image into layers, halftone each one of them with a binary halftoning algorithm, and combine their binary halftones to produce the final multilevel halftone. As layers are handled one by one subject to a stacking constraint, darker pixels in the final multilevel output are positioned first, which makes it difficult for the algorithms to preserve bright spatial features. This issue is addressed and a solution to eliminate this bias is proposed. Simulation result shows that the proposed method can provide an output of better quality as compared with conventional threshold decomposition-based algorithms both subjectively and objectively.

The proposed framework for cognitive analysis of perfusion computed tomography images is a fusion of image processing, pattern recognition, and image analysis procedures. The output data of the algorithm consists of: regions of perfusion abnormalities, anatomy atlas description of brain tissues, measures of perfusion parameters, and prognosis for infracted tissues. That information is superimposed onto volumetric computed tomography data and displayed to radiologists. Our rendering algorithm enables rendering large volumes on off-the-shelf hardware. This portability of rendering solution is very important because our framework can be run without using expensive dedicated hardware. The other important factors are theoretically unlimited size of rendered volume and possibility of trading of image quality for rendering speed. Such rendered, high quality visualizations may be further used for intelligent brain perfusion abnormality identification, and computer aided-diagnosis of selected types of pathologies.

A robust copyright scheme for image protection based on visual secret sharing (VSS) and Bose-Chaudhuri-Hocquenghem (BCH) code techniques is proposed. This scheme not only maintains the quality of a host image without the change of any pixel value but also generates a meaningful ownership share to improve the management of image copyright. In addition, no codebook is required to store, and the watermark size is independent of the host image. The robustness of watermarking can be enhanced by BCH code. The proposed scheme contains ownership share construction and watermark extraction. In the first phase, an encoded watermark is generated by BCH code from a watermark. Next, an image feature is then extracted by the discrete wavelet transform decomposing from the host image. Finally, an ownership share can be generated by VSS technique from the image feature and the encoded watermark. In the second phase, a master share can be produced from a suspect image. By stacking the master and the ownership shares and using BCH code, an extracted watermark can be obtained. The experimental results show that the proposed scheme using the BCH(15,5) has better robust performance and practicability than existing schemes.

The Beltrami flow is an effective tool for dealing with images in many image-processing tasks. However, the important issue of how to set a proper embedding space is not solved in the Beltrami framework. We attempt to find a suitable embedding space by a nonlinear map. First, the image space is mapped into a high-dimensional feature space whose dimensionality may be infinite. It is found that directly dealing with the Beltrami flow in Hilbert space is impractical due to the unknown mapping function. Fortunately, using the well-known kernel methods, one can obtain the Beltrami flow in Hilbert space by performing inner products in the feature space. We refer to this flow as the kernel Beltrami flow. In addition, we also extend the kernel Beltrami flow to deal with vector-valued images. Finally, we show the effectiveness of the proposed method on gray-level and color images.

A classical problem of additive white (spatially uncorrelated) Gaussian noise suppression in grayscale images is considered. The main attention is paid to discrete cosine transform (DCT)-based denoising, in particular, to image processing in blocks of a limited size. The efficiency of DCT-based image filtering with hard thresholding is studied for different sizes of overlapped blocks. A multiscale approach that aggregates the outputs of DCT filters having different overlapped block sizes is proposed. Later, a two-stage denoising procedure that presumes the use of the multiscale DCT-based filtering with hard thresholding at the first stage and a multiscale Wiener DCT-based filtering at the second stage is proposed and tested. The efficiency of the proposed multiscale DCT-based filtering is compared to the state-of-the-art block-matching and three-dimensional filter. Next, the potentially reachable multiscale filtering efficiency in terms of output mean square error (MSE) is studied. The obtained results are of the same order as those obtained by Chatterjee's approach based on nonlocal patch processing. It is shown that the ideal Wiener DCT-based filter potential is usually higher when noise variance is high.

For classifying images with various appearances, graph embedding based subspace learning has difficulty in taking a comprehensive consideration of both local geometrical structure and between-class discriminative information. In addition, when no sufficient training samples exist, using only the simple weight graph corresponding to labeled samples, the embedding subspace may not be accurately modeled. We present a semisupervised graph embedding algorithm by combining graph embedding and sparse representation. This algorithm can effectively learn a compact and semantic subspace by using a locally connected graph, which can model the geometrical structure and essential correlation of subclusters within a class and can fully utilize both labeled and unlabeled samples. Moreover, using L2,1-norm, the proposed algorithm can preserve the sparse representation property of images from the original space in the lower dimensional projected space. Our experiments demonstrate that the proposed algorithm has better performance than the alternatives reported in recent literature.

Image features are regularly used in computer and machine vision applications to solve problems pertaining to image geometry. The effectiveness of techniques used rely on how accurately corresponding features are located in two (or more) images. Current techniques define a feature as a variation in intensity within a pixel neighborhood. The greater the change, the more pronounced a feature becomes, and is thus more accurately located. The remaining uncertainty is characterized via a covariance matrix, which is used to weight the influence of each feature on model estimation. As our first contribution, we propose a novel technique to decrease feature location uncertainty. Considering uncertainty information from all three channels of an RGB image, we employ covariance intersection (CI) to reduce measurement errors leading to a decrease in homography (H) estimation error. Our second contribution is to use an L∞ filter capable of dealing with system uncertainties to estimate H. By feeding the feature location uncertainty information into our filter, we observe significantly improved estimates of the homography. Our filtering approach outperforms covariance weighted optimization techniques proposed in literature shown through a number of examples.

A stereo vision algorithm suitable for fast dense stereo matching is proposed. This method extracts areas of uniform disparity or dense disparity features by testing candidate disparity estimates and classifying pixels according to a matching error. By commencing with a rigorous error threshold, pixels of reliable disparity values are identified. As the threshold is relaxed, other pixels are aggregated, forming dense areas of uniform disparity. Neighboring support is acquired with texture filling and other morphological operations in each disparity feature. The algorithm is tested on publicly available and self-generated data sets and is shown to produce promising results in terms of quality and speed. Handling of regions without texture is discussed and a comparison with other methods is also presented.

By reconsidering some two-dimensional video inherited approaches and by adapting them to the stereoscopic video content and to the human visual system peculiarities, a new disparity map is designed. First, the inner relation between the left and the right views is modeled by some weights discriminating between the horizontal and vertical disparities. Second, the block matching operation is achieved by considering a visual related measure (normalized cross correlation) instead of the traditional pixel differences (mean squared error or sum of absolute differences). The advanced three-dimensional (3-D) video-new three step search (3DV-NTSS) disparity map (3-D Video-New Three Step Search) is benchmarked against two state-of-the-art algorithms, namely NTSS and full-search MPEG (FS-MPEG), by successively considering two corpora. The first corpus was organized during the 3DLive French national project and regroups 20 min of stereoscopic video sequences. The second one, with similar size, is provided by the MPEG community. The experimental results demonstrate the effectiveness of 3DV-NTSS in both reconstructed image quality (average gains between 3% and 7% in both PSNR and structural similarity, with a singular exception) and computational cost (search operation number reduced by average factors between 1.3 and 13). The 3DV-NTSS was finally validated by designing a watermarking method for high definition 3-D TV content protection.

We present a method suitable for a time-to-impact sensor. Inspired by the seemingly "low" complexity of small insects, we propose a new approach to optical flow estimation that is the key component in time-to-impact estimation. The approach is based on measuring time instead of the apparent motion of points in the image plane. The specific properties of the motion field in the time-to-impact application are used, such as measuring only along a one-dimensional (1-D) line and using simple feature points, which are tracked from frame to frame. The method lends itself readily to be implemented in a parallel processor with an analog front-end. Such a processing concept [near-sensor image processing (NSIP)] was described for the first time in 1983. In this device, an optical sensor array and a low-level processing unit are tightly integrated into a hybrid analog-digital device. The high dynamic range, which is a key feature of NSIP, is used to extract the feature points. The output from the device consists of a few parameters, which will give the time-to-impact as well as possible transversal speed for off-centered viewing. Performance and complexity aspects of the implementation are discussed, indicating that time-to-impact data can be achieved at a rate of 10 kHz with today's technology.

A personal identification method is proposed which uses face and ear together to overcome mass information loss resulting from pose changes. Several aspects are mainly considered: First, ears are at both sides of the face. Their physiological position is approximately orthogonal and their information is complementary to each other when the head pose changes. Therefore, fusing the face and ear is reasonable. Second, the texture feature is extracted using a uniform local binary pattern (ULBP) descriptor which is more compact. Third, Haar wavelet transform, blocked-based, and multiscale ideas are taken into account to further strengthen the extracted texture information. Finally, texture features of face and ear are fused using serial strategy, parallel strategy, and kernel canonical correlation analysis to further increase the recognition rate. Experimental results show that it is both fast and robust to use ULBP to extract texture features. Haar wavelet transform, block-based, and multiscale methods can effectively enhance texture information of the face or ear ULBP descriptor. Multimodal biometrics fusion about face and ear is feasible and effective. The recognition rates of the proposed approach outperform remarkably those of the classic principal component analysis (PCA), kernel PCA, or Gabor texture feature extraction method especially when sharp pose change

Blind source separation (BSS) is a process in which mixed signals are separated into their original sources. Both the sources as well as the mixing coefficients are unknown but a priori information about statistical behavior and about the mixing model might be available. We here suggest a generalization of our previous research that showed a new BSS algorithm based on general cross correlation linear operators applied on the sources that are to be separated. In that approach in cases of negligible cross-correlation between the source signals, a very good separation could be obtained. Here we propose to use the fractional Fourier transform in order to reduce the correlation between the source signals and to further enhance the obtained separation performance. We present reduced dependence on the cross-correlation between the source images, resulting in better separation of the mixed sources.

This article [J. Electron. Imaging. 21, (4 ), 043014 (2012)] was originally published online on 22 November 2012 with errors in two author affiliations. Conference Presentation Video Visit SPIE.TV Author affiliations of Sanghyun Joo and Kwanghoon Sohn were erroneously listed as Yonsei University, Department of Electrical and Electronic Engineering, 134 Shinchon-dong, Seodaemun-gu, Seoul 120-749, Republic of Korea. Both authors are actually affiliated with Agency for Defense Development, Daejeon 305-152, Republic of Korea, as listed above. Conference Presentation Video Visit SPIE.TV All online versions of the article were corrected on 11 December 2012.