This PDF file contains the front matter associated with SPIE Proceedings Volume 7532, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.

The bilateral filter and the non-local means (NL-means) filter are known as very powerful nonlinear filters. The first contribution of this paper is to give a general framework which involves the bilateral filter and the NL-means filter. The general framework is derived based on Bayesian inference. Our analysis reveals that the range weight in the bilateral filter and the similarity measure in the NL-means filter are associated with a noise model or a likelihood distribution. The second contribution is to extend the bilateral filter and the NL-means filter for a general noise model. We also provide a filter classification. The filter classification framework clarifies the differences among existing filters and helps us to develop new filters. As example of future directions, we extend the bilateral filter and the NL-means filter for a general noise model. Both extended filters are theoretically and experimentally justified.

In this paper, we propose a new edge detection method that combines the concepts of Logarithmic Image Processing (LIP) and gray level ratio operations. The presented method provides a new approach in edge detection where edges are detected at different ratios of gray levels independently. The proposed method detects edge details based on the band of gray levels ratio of interest that these details lie within, while current edge detection algorithms control such edge details by threshold variations. In the proposed algorithm, variations of some introduced constant value introduce different edge details and not necessarily more or less details. Extensive simulations demonstrate that this method produces competitive results by suppressing impulsive noise, the ability to trace the development of certain object structures, and segmenting objects that belong to a certain area of interest.

A collaborative image processing method is proposed to enhance the shape and details of a scene through synthesizing a set of multi-light images that capture a scene with fixed view-point but various lighting positions. A very challenging problem is to remove the artifacts due to shadow edges from the synthesized image. To address this problem, a simple Sobel filter based method is provided by utilizing the feature of multi-light images in which the shadow edges are usually not overlapped. A detail layer that contains the details of all images is firstly constructed by using a gradient domain method and a quadratic filter. Then a base layer is produced by using only one input image. The detail layer is finally added to the base layer to produce the desired detail enhanced image. Using this method, the details lost in the shadow of original input image can be reproduced by the details of other images and the sense of depth is preserved well in the synthesized image. Interactivities are also provided for users to adjust the appearance of the detail enhanced image according to their preferences.

An original blurriness assessment method for video frames is presented. In the first place two measurements are performed uniformly on the whole frame. The first quantifies the perceptive contrast experienced by the human eye, the second uses anisotropic diffusion to estimate a possible detail loss due to blurriness. Secondarily two further indices are devised, each one suitable to a particular sort of image content. The first is dedicated to main, uniform objects, where smoothed borders are deemed to be the main source of blurriness perception. First, a technique is devised to extract all edges, including the smoothest ones. Then the width of such edges is measured, and more weight is given to long than to short ones. The second index measures the activity of textured areas, trying to detect blurriness inside the base texture elements. The four devised indices enable automatic quantification of the strength of blurriness and some hints at its origin. In particular, some new results have been achieved in the ability to automatically distinguish natural blurriness, present in the image content, from undesired one, introduced during encoding and processing.

In design of many image processing methods and algorithms, it is assumed that noise is i.i.d. However, noise in real life images is often spatially correlated and ignoring this fact can lead to certain problems such as reduction of filter efficiency, misdetection of edges, etc. Thus, noise characteristics, namely, variance and spatial spectrum are to be estimated. This should be often done in a blind manner, i.e., for an image at hand and in non-interactive manner. This task is especially complicated if an image is textural. Thus, the goal of this paper is to design a practical approach to blind estimation of noise characteristics and to analyze its performance. The proposed method is based on analysis of data in blocks of fixed size in discrete cosine transform (DCT) domain. This allows further use of the obtained DCT spectrum for denoising and other purposes. This can be especially helpful for multichannel remote sensing (RS) data where interactive processing is problematic and sometimes even impossible.

In this paper, we present an algorithm for image stitching that avoids performance hindrance and memory issues in diverse image processing applications/ environments. High-resolution images could be cut into smaller pieces by various applications for ease of processing, especially if they are sent over a computer network. Image pieces (from several highresolution images) could be stored as a single image-set with no information about the original images. We propose a robust stitching methodology to reconstruct the original high-resolution image(s) from a target image-set that contains components of various sizes and resolutions. The proposed algorithm consists of three major modules. The first step sorts image pieces into different planes according to their spatial position, size, and resolution. It avoids sorting overlapped pieces of the same resolution in the same plane. The second module sorts the pieces from different planes according to their content by minimizing a cost function based on Mean Absolute Difference (MAD). The third module relates neighboring pieces and determines output images. The proposed algorithm could be used at a pre-processing stage in applications such as rendering, enhancement, retrieval etc, as these cannot be carried out without access to original images as individual whole components.

An image processing algorithm is presented that adaptively converts color images to grayscale. The intent of the conversion is to preserve color information that is traditionally lost by the conversion process. The conversion produces high-contrast grayscale representations with enhanced color discriminability. A web-based psychometric study confirms that the algorithm is mostly preferred over traditional algorithms. The algorithm employs a multi-stepped approach that includes color clustering, 3-dimensional partitioning, and simulated annealing.

A new motion registration method based on specific level-line grouping from two images taken at different times is proposed in this paper. Our approach is an appropriate method for matching outdoor scene image sequences taken from an on-board camera, since it is robust toward contrast changes. Moreover, it does not require any estimate of the unknown transformation between images. The image registration is performed through an efficient level-line cumulative multi-stage election procedure. Each stage provides an improvement in estimating the unknown transformation. We consider in this study that the transformation between successive images is affine, which is a valid hypothesis under small time intervals between images. Our matching process is based upon adapted primitive invariants. Particular emphasis is placed on the primitive grouping process which depends on the nature of the chosen transformation.

In this work a novel technique for selecting key points is proposed. Key points are used in many image processing applications and should be robust with respect to noise, rotation, blurring, and so on. The selection is based on the amount of local Fisher's information about location, orientation and scale. Based on the relationship between Taylor polynomials in Cartesian coordinates and Zernike polynomials in polar coordinates, the Fisher's information matrix can be written in terms of the image Zernike's expansion coefficients, which can be easily computed by means of a bank of filters. To evaluate the performances of the proposed method we consider four different distortions at three levels. Experimental results show that the performances, in terms of repeatability rate, are better that the performances obtained by the conventional Harris detector.

It has been shown through various research efforts over the past few decades that there are numerous advantages to sampling images hexagonally rather than rectangularly. Despite the advantages, hexagonal imaging has not been generally accepted as being advantageous due to the lack of sensors that sample hexagonally, lack of displays for hexagonal images, and lack of an elegant addressing scheme. The advantages gained by sampling hexagonally are offset by the additional processing required to deal with the problems that are inherent with the previously proposed addressing schemes. Hence, there is insufficient motivation to develop sensors and displays that operate in the hexagonal domain. This paper introduces an addressing scheme, array set addressing, that solves the problems exhibited by other approaches. This new approach represents the hexagonal grid with a pair of rectangular arrays, supporting efficient linear algebra and image processing manipulation. Early results have shown that image processing techniques such as convolution, downsampling, calculating Euclidean distances, and vector arithmetic can be done with no more complexity than is required for processing rectangularly sampled images.

The sharpness of an image is a function of its spectral density. Wider spectrum implies sharper image. Thus the image sharpness can be measured by measuring the shape of its spectrum. Bivariate kurtosis can be used to measure the shape and shoulder of a two dimensional probability distribution. It is known that the low frequencies correspond to the slowly changing components of an image and high frequencies correspond to faster gray level changes in the image, which gives information about the finer details such as edges. When an image is in focus, the high frequency components are maximized to define the edges sharply. Thus kurtosis, which measures the width of the shoulder of the probability distribution, corresponding to the high frequencies, can be used to measure the sharpness. This work presents efficient low complexity architecture of kurtosis based image sharpness no reference metric. The calculation of higher order moments is a computational intensive task that involves a large number of additions and multiplications. A recursive IIR filter based implementation of the moments is proposed using a cascade of single pole filters. The conducted simulation results show clearly the reduction in computation while maintaining the same accuracy.

This paper presents an innovative watermarking scheme which allows the insertion of information in the Discrete Cosine Transform (DCT) domain increasing the perceptual quality of the watermarked images by exploiting the masking effect of the DCT coefficients. Indeed, we propose to make the strength of the embedded data adaptive by following the characteristics of the Human Visual System (HVS) with respect to image fruition. Improvements in the perceived quality of modified data are evaluated by means of various perceptual quality metrics as demonstrated by experimental results.

In this paper we propose a multispectral image compression based on lossy to lossless coding, suitable for both spectral and color reproduction. The proposed method divides a multispectral image data into two groups, RGB and residual. The RGB component is extracted from the multispectral image, for example, by using the XYZ Color Matching Functions, a color conversion matrix, and a gamma curve. The original multispectral image is estimated from RGB data encoder, and the difference between the original and the estimated multispectral images, referred as a residual component in this paper, is calculated in the encoder. Then the RGB and the residual components are encoded by JPEG2000, respectively a progressive decoding is possible from the losslessly encoded code-stream. Experimental results show that, although the proposed method is slightly inferior to JPEG2000 with a multicomponent transform in rate-distortion plot of the spectrum domain at low bit rate, a decoded RGB image shows high quality at low bit rate with primary encoding of the RGB component. Its lossless compression ratio is close to that of JPEG2000 with the integer KLT.

Distributed Video Coding (DVC) is an emerging video coding paradigm for the systems that require low complexity encoders supported by high complexity decoders. A typical real world application for a DVC system is mobile phones with video capture hardware that have a limited encoding capability supported by base-stations with a high decoding capability. Generally speaking, a DVC system operates by dividing a source image sequence into two streams, key frames and Wyner-Ziv (W) frames, with the key frames being used to represent the source plus an approximation to the W frames called S frames (where S stands for side information), while the W frames are used to correct the bit errors in the S frames. This paper presents an effective algorithm to reduce the bit errors in the side information of a DVC system. The algorithm is based on the maximum likelihood estimation to help predict future bits to be decoded. The reduction in bit errors in turn reduces the number of parity bits needed for error correction. Thus, a higher coding efficiency is achieved since fewer parity bits need to be transmitted from the encoder to the decoder. The algorithm is called inter-bit prediction because it predicts the bit-plane to be decoded from previously decoded bit-planes, one bitplane at a time, starting from the most significant bit-plane. Results provided from experiments using real-world image sequences show that the inter-bit prediction algorithm does indeed reduce the bit rate by up to 13% for our test sequences. This bit rate reduction corresponds to a PSNR gain of about 1.6 dB for the W frames.

Video processing algorithms tend to improve over time in terms of image quality while increasing in its inter and intra dependency. Frames inter prediction exploits temporal similarities across a sequence of consecutive frames, while intra prediction exploits the macroblock's spatial similarities in the same frame. They both work together to efficiently compress the video stream (maximize the signal to noise ratio (SNR) while minimizing the used bandwidth (BW)). Thus, different parts of the video stream (blocks and/or frames) have different semantic importance, and thus require different degrees of protection against network losses to maintain a constant quality of service (QoS). This becomes even more important in layered codec (e.g., scalable video codec SVC/H.264), where the stream is compromised of more than one video layer. Based on the expected video experience, available bandwidth and compute resources, we could use one or more layers to achieve a certain level of experience. This becomes challenging in lossy networks, where losses could harm not only the immediate group of pictures (GOP), but will propagate across multi video layers. In this paper, we present a method to adequately distributed forward error correction (FEC) packets across multi layers to preserve the video experience under lossy conditions.

In this paper, we consider facial expression recognition using an unsupervised learning framework. Specifically, given a data set composed of a number of facial images of the same subject with different facial expressions, the algorithm segments the data set into groups corresponding to different facial expressions. Each facial image can be regarded as a point in a high-dimensional space, and the collection of images of the same subject resides on a manifold within this space. We show that different facial expressions reside on distinct subspaces if the manifold is unfolded. In particular, semi-definite embedding is used to reduce the dimensionality and unfold the manifold of facial images. Next, generalized principal component analysis is used to fit a series of subspaces to the data points and associate each data point to a subspace. Data points that belong to the same subspace are shown to belong to the same facial expression.

The "enhanced spectrum" of an image g[.] is a function h[.] of wave-number u obtained by a sequence of operations on the power spectral density of g[.]. The main properties and the available theorems on the correspondence between spectrum enhancement and spatial differentiation, of either integer or fractional order, are stated. In order to apply the enhanced spectrum to image classification, one has to go, by interpolation, from h[.] to a polynomial q[.]. The graph of q[.] provides the set of morphological descriptors of the original image, suitable for submission to a multivariate statistical classifier. Since q[.] depends on an n-tuple, Ψ, of parameters which control image pre-processing, spectrum enhancement and interpolation, then one can train the classifier by tuning Ψ. In fact, classifier training is more articulated and relies on a "design", whereby different training sets are processed. The best performing n-tuple, Ψ*, is selected by maximizing a "design-wide" figure of merit. Next one can apply the trained classifier to recognize new images. A recent application to materials science is summarized.

In this paper, we introduce a novel Gabor based Spacial Domain Class-Dependence Feature Analysis(GSD-CFA) method that increases the Face Recognition Grand Challenge (FRGC)2.0 performance. In short, we integrate Gabor image representation and spacial domain Class-Dependence Feature Analysis(CFA) method to perform fast and robust face recognition. In this paper, we mainly concentrate on the performances of subspace-based methods using Gabor feature. As all the experiments in this study is based on large scale face recognition problems, such as FRGC, we do not compare the algorithms addressing small sample number problem. We study the generalization ability of GSD-CFA on THFaceID data set. As FRGC2.0 Experiment #4 is a benchmark test for face recognition algorithms, we compare the performance of GSD-CFA with other famous subspace-based algorithms in this test.

This paper describes an improved framework to generate seamless image mosaic automatically. The input images are classified into different image connectivity sets by using speeded features. For each set, the proposed method includes two parts: initial alignment and global optimization. The initial geometrical registration is obtained using the classical parameterization of homography. The photometric parameters among images are computed to balance the overall gain using the least square method. A good global optimization can be achieved with these initial information and multiple constraints between images. Besides, we also take the radial distortion into consideration. Mulitresolution blending combined with an optimal seam selection gives the final seamless mosaic. Various image sets illustrate the robustness of the proposed method in experimental results.

Contrast enhancement main goal consists on improving the image visual appearance but also it is used for providing a transformed image in order to segment it. In mathematical morphology several works have been derived from the framework theory for contrast enhancement proposed by Meyer and Serra. However, when working with images with a wide range of scene brightness, as for example when strong highlights and deep shadows appear in the same image, the proposed morphological methods do not allow the enhancement. In this work, a rational multi-scale method, which uses a class of morphological connected filters called filters by reconstruction, is proposed. Granulometry is used by finding the more accurate scales for filters and with the aim of avoiding the use of other little significant scales. The CIE-u'v'Y' space was used to introduce our results since it takes into account the Weber's Law and by avoiding the creation of new colors it permits to modify the luminance values without affecting the hue. The luminance component ('Y) is enhanced separately using the proposed method, next it is used for enhancing the chromatic components (u', v') by means of the center of gravity law of color mixing.

In this paper, we present a circularly symmetric point spread function (PSF) estimation technique for a fully digital autofocusing system. The proposed algorithm provides realistic, unsupervised PSF estimation by establishing the relationship between the one-dimensional ideal step response and the corresponding two-dimensional circularly symmetric PSF. Main advantage of the proposed algorithm is the accurate estimation of the PSF combining by using interpolation, feasible step response estimation. The proposed estimation method will serve as (i) a fundamental procedure in designing an in-house extended depth-of-field (EDoF) system, (ii) a detecting method for improperly manufactured imaging sensors in the production line, and (iii) a PSF estimation method for general image restoration algorithms.

Binary images can be represented by their morphological skeleton transform also called as Medial axis transform (MAT). Shock graphs are derived from the skeleton and have emerged as powerful 2-D shape representation method. A skeleton has number of braches. A branch is a connected set of points between an end point and a joint or another end point. Every point also called as shock point on a skeleton can be labeled according to the variation of the radius function. The labeled points in a given branch are to be grouped according to their labels and connectivity, so that each group of same-label connected points will be stored in a graph node. One skeleton branch can give rise to one or more nodes. Finally we add edges between the nodes so as to produce a directed acyclic graph with edges directed according to the time of formation of shock points in each node. We have generated shock graphs using two different approaches. In the first approach the skeleton branches and the nodes have labels 1, 2, 3 or 4 where as the second approach excludes type 2 label making the graph simpler. All the joints are called as the branch points. We have compared the merits and demerits of the two methods.

Successful user discrimination in a vehicle environment may yield a reduction of the number of switches, thus significantly reducing costs while increasing user convenience. The personalization of individual controls permits conditional passenger enable/driver disable and vice versa options which may yield safety improvement. The authors propose a prototypic optical sensing system based on hand movement segmentation in near-infrared image sequences implemented in an Audi A6 Avant. Analyzing the number of movements in special regions, the system recognizes the direction of the forearm and hand motion and decides whether driver or front-seat passenger touch a control. The experimental evaluation is performed independently for uniformly and non-uniformly illuminated video data as well as for the complete video data set which includes both subsets. The general test results in error rates of up to 14.41% FPR / 16.82% FNR and 17.61% FPR / 14.77% FNR for driver and passenger respectively. Finally, the authors discuss the causes of the most frequently occurring errors as well as the prospects and limitations of optical sensing for user discrimination in passenger compartments.

Vehicle seat occupancy detection systems are designed to prevent the deployment of airbags at unoccupied seats, thus avoiding the considerable cost imposed by the replacement of airbags. Occupancy detection can also improve passenger comfort, e.g. by activating air-conditioning systems. The most promising development perspectives are seen in optical sensing systems which have become cheaper and smaller in recent years. The most plausible way to check the seat occupancy by occupants is the detection of presence and location of heads, or more precisely, faces. This paper compares the detection performances of the three most commonly used and widely available face detection algorithms: Viola- Jones, Kienzle et al. and Nilsson et al. The main objective of this work is to identify whether one of these systems is suitable for use in a vehicle environment with variable and mostly non-uniform illumination conditions, and whether any one face detection system can be sufficient for seat occupancy detection. The evaluation of detection performance is based on a large database comprising 53,928 video frames containing proprietary data collected from 39 persons of both sexes and different ages and body height as well as different objects such as bags and rearward/forward facing child restraint systems.

In this paper, we present two novel medical image enhancement algorithms. The first, a global image enhancement algorithm, utilizes an alpha-trimmed mean filter as its backbone to sharpen images. The second algorithm uses a cascaded unsharp masking technique to separate the high frequency components of an image in order for them to be enhanced using a modified adaptive contrast enhancement algorithm. Experimental results from enhancing electron microscopy, radiological, CT scan and MRI scan images, using the MATLAB environment, are then compared to the original images as well as other enhancement methods, such as histogram equalization and two forms of adaptive contrast enhancement. An image processing scheme for electron microscopy images of Purkinje cells will also be implemented and utilized as a comparison tool to evaluate the performance of our algorithm.

As a preliminary step in the study of long-term global cloud properties variations and their contribution to the Earth radiation budget, a classification procedure to identify various types of clouds is discussed. This classification scheme highlighting the spatial heterogeneity of cloud structural arrangements is used to characterize and differentiate nine cloud types. The study takes advantage of the capacity of edge gradient operators' techniques generally used to calculate the magnitude and direction changes in the intensity function of adjacent pixels of an image, to identify the various cloud types. The specific approach, based on variations of the edge gradient magnitude and orientation, is applied on daytime global cloud physical features (cloud top temperatures derived from the 11-μm brightness temperature imagery) obtained from the National Oceanic and Atmospheric Administration-Advanced Very-High-Resolution Radiometer (NOAA-AVHRR) satellite observations. The results obtained are compared with those of the International Satellite Cloud Climatology Project (ISCCP) cloud classification algorithm which uses cloud optical properties and pressure levels to distinguish cloud types. Results of these two procedures show good agreement but substantial differences are noticed at polar areas.

In this paper, a robust method for hidden information in the wavelet domain is presented. The proposed steganographic method uses the Wavelet domain iterative center weighted median (WDICWM) algorithm to estimate the noise areas of an RGB color image. The noise areas or pixels are proposed as places to hidden information to provide good invisibility and fine detail preservation of processed images. The proposed steganographic method consistently identifies if the wavelets coefficients of an image contains noise pixels or not, this algorithm works with variances and standard deviations of the same wavelet coefficients of image. The experimental results from a known technique are compared with the proposed method to demonstrate its performance in terms of objective and subjective sense.

Through its evolution with time, anisotropic diffusion provides multi-scale edge-sensitive smoothing of noisy images. Depending on the type of equation used, such a procedure may also have the ability to sharpen the edges. This paper characterizes the edge-sharpening abilities of a well-known diffusion equation based on the characteristics of the second eigenvalue of the Hessian of a function related to the diffusivity function. It then proposes a new way of diffusivity function design based on the natural requirement of the degree of edge sharpening monotonically related to the strength of edges. A comparative example based on the Structural Similarity Index (SSIM) is also presented.

In this contribution a Multiple Description Coding scheme for video transmission over unreliable channel is presented. The method is based on an integer wavelet transform and on a data hiding scheme for exploiting the spatial redundancy and for reducing the scheme overhead. Experimental results show the effectiveness of the proposed scheme.

In this contribution a novel reversible data hiding scheme for digital images is presented. The proposed technique allows the exact recovery of the original image upon extraction of the embedded information. Lossless recovery of the original is achieved by adopting the histogram shifting technique in a novel wavelet domain: the Integer Fibonacci-Haar Transform, which is based on a parameterized subband decomposition of the image. In particular, the parametrization depends on a selected Fibonacci sequence. The use of this transform increases the security of the proposed method. Experimental results show the effectiveness of the proposed scheme.

'View plus depth' is a 3D video representation where a single color video channel is augmented with per-pixel depth information in the form of gray-scale video sequence. This representation is a good candidate for 3D video delivery applications, as it is display agnostic and allows for some parallax adjustments. However, the quality of the associated depth is an issue, as the depth channel is usually a result of estimation procedure based on stereo correspondences or comes from a noisy and low-resolution range sensor. Therefore, proper filtering of the depth channel is needed before it is used for compression and/or view rendering. The problem is even more pronounced in video, where temporal consistency of the depth sequence is required. In this paper, we propose a filtering approach to refine the quality of noisy, blocky, and temporally-inconsistent depth maps. We utilize color constraints from the video channel and modify a previous super-resolution approach to tackle the time consistency for video. Our implementation is fast and highly memory efficient. We present filtering results demonstrating the superiority of the developed technique.