This pdf file contains the front matter associated with SPIE Proceedings Volume 7799, including Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

In this paper we focus on the recovery of 3D shape (in a well defined sense) from matched features in 2D views. Our approach to shape reconstruction is invariant to pose and not dependent on knowing any of the camera parameters or the camera's location. The methods are also independent of any choice of coordinates used to represent the 2D or 3D features. We begin by examining two fundamental problems. The first problem involves determining global equations for the geometric constraints (object/image equations) that must hold between a set of object feature points and any image of those points, and using these equations to effectively reconstruct the 3D shape. The second involves describing the shape spaces for point features.

A number of hyperspectral (x, y, λ) imaging systems work on the principle of limited angle tomography. In such systems there exists a region of spatial and spectral frequencies called the "missing cone" that the imaging system cannot recover from data using any direct reconstruction algorithms. Wavelets are useful for imaging objects that are spatially and in many cases also spectrally compact. However wavelet expansion functions have three-dimensional frequency content intersecting the missing cone region; this means the wavelets themselves are altered thus compromising the corresponding datacube reconstructions. As the missing cone of frequencies is fixed for a given imaging system, it is reasonable to adjust parameters in the wavelets themselves in order to reduce the intersection between the wavelets' frequency content and the missing cone. One wavelet system is better than another when the frequency content of the former has a smaller amount of overlap with the missing cone. We will do this analysis with a couple of classic wavelet families, the Morlet and the Difference of Gaussian (DOG) for an existing hyperspectral tomographic imaging system to show the feasibility of this procedure.

This paper proposes a comprehensive overview of state of the art 2D/3D, view-based indexing methods. The principle of 2D/3D indexing methods consists of describing 3D models by means of a set of 2D shape descriptors, associated with a set of corresponding 2D views (under the assumption of a given projection model). Notably, such an approach makes it possible to identify 3D objects of interest from 2D images/videos. An experimental evaluation is also proposed, in order to examine the influence of the number of views and of the associated viewing angle selection strategies on the retrieval results. Experiments concern both 3D model retrieval and image recognition from a single view. Results obtained show promising performances, with recognition rates from a single view higher then 66%, which opens interesting perspectives in terms of semantic metadata extraction from still images/videos.

Image algebra is a rigorous, concise notation that unifies linear and nonlinear mathematics in the image domain. Image algebra was developed under DARPA and US Air Force sponsorship at University of Florida for over 15 years beginning in 1984. Image algebra has been implemented in a variety of programming languages designed specifically to support the development of image processing and computer vision algorithms and software. The University of Florida has been associated with development of the languages FORTRAN, Ada, Lisp, and C++. The latter implementation involved a class library, iac++, that supported image algebra programming in C++. Since image processing and computer vision are generally performed with operands that are array-based, the Matlab™ programming language is ideal for implementing the common subset of image algebra. Objects include sets and set operations, images and operations on images, as well as templates and image-template convolution operations. This implementation, called Image Algebra Matlab (IAM), has been found to be useful for research in data, image, and video compression, as described herein. Due to the widespread acceptance of the Matlab programming language in the computing community, IAM offers exciting possibilities for supporting a large group of users. The control over an object's computational resources provided to the algorithm designer by Matlab means that IAM programs can employ versatile representations for the operands and operations of the algebra, which are supported by the underlying libraries written in Matlab. In a previous publication, we showed how the functionality of IAC++ could be carried forth into a Matlab implementation, and provided practical details of a prototype implementation called IAM Version 1. In this paper, we further elaborate the purpose and structure of image algebra, then present a maturing implementation of Image Algebra Matlab called IAM Version 2.3, which extends the previous implementation of IAM to include polymorphic operations over different point sets, as well as recursive convolution operations and functional composition. We also show how image algebra and IAM can be employed in image processing and compression research, as well as algorithm development and analysis.

Space Time Adaptive Processing (STAP) is a multi-dimensional adaptive signal processing technique, which processes the signal in spatial and Doppler domains for which a target detection hypothesis is to be formed. It is a sample based technique and based on the assumption of adequate number of Independent and Identically Distributed (i.i.d.) training data set in the surrounding environment. The principal challenge of the radar processing lies when it violates these underlying assumptions due to severe dynamic heterogeneous clutter (hot clutter) and jammer effects. This in turn degrades the Signal to Interference-plus-Noise Ratio (SINR), hence signal detection performance. Classical Wiener filtering theory is inadequate to deal with nonlinear and nonstationary interferences, however Wiener filtering approach is optimal for stationary and linear systems. But, these challenges can be overcome by Adaptive Sequential State Estimation (ASSE) filtering technique.

Generally, the designs of digital image processing algorithms and image gathering devices remain separate. Consequently, the performance of digital image processing algorithms is evaluated without taking into account the artifacts introduced into the process by the image gathering process. However, experiments show that the image gathering process profoundly impacts the performance of digital image processing and the quality of the resulting images. Huck et al. proposed one definitive theoretic analysis of visual communication channels, where the different parts, such as image gathering, processing, and display, are assessed in an integrated manner using Shannon's information theory. In this paper, we perform an end-to-end information theory based system analysis to assess edge detection methods. We evaluate the performance of the different algorithms as a function of the characteristics of the scene, and the parameters, such as sampling, additive noise etc., that define the image gathering system. The edge detection algorithm is regarded to have high performance only if the information rate from the scene to the edge approaches the maximum possible. This goal can be achieved only by jointly optimizing all processes. People generally use subjective judgment to compare different edge detection methods. There is not a common tool that can be used to evaluate the performance of the different algorithms, and to give people a guide for selecting the best algorithm for a given system or scene. Our information-theoretic assessment becomes this new tool to which allows us to compare the different edge detection operators in a common environment.

In the iterative CAD design of new materials by digital computers, it is necessary to obtain the differential coefficients, that is, component-sensitivities caused by the small deviation of inner-components in a given electromagnetic field expressed by the Maxwell relations. Further, to determine the step-size of the numerical iterative CAD design that uses the discrete sample values of the wave form at the sample points with the same interval of the step-size, it is required to estimate the error favorably between the original wave form and its numerical approximation. In this paper, firstly, we present conservation operators in micro-electromagnetic field and its macro-expression in the electromagnetic field. Secondly, we present some concrete conservation operators and make clear that the certain quantities, such as the stored energy of a small inner-component in the closed electromagnetic field, are closely correlated to the differential coefficients of the electric field and the magnetic field observed at the outer ports. Secondly, in a single-mode electromagnetic field, we obtain the relation between the stored energy, and the component-sensitivities caused by the small deviation of the inner-component. Thirdly, we present a brief survey of the progress in the development of meta material and show the usefulness of combining the above results with the optimum nonlinear approximation in the iterative design of linear or nonlinear meta material.

The author presents an overview of 3D object recognition with photon-counting integral imaging using Independent Component Analysis (ICA). High resolution elemental images of 3D objects are captured at different angles to allow object recognition at different orientations using synthetic aperture integral imaging (SAII). Generated photon-counting elemental images are used to reconstruct the 3D images at different distances from the camera lens using a maximum a posteriori estimation method. The kurtosis maximization-based algorithm is applied as a non-gaussian maximization method to extract the independent features from the training data set. High dimensional data is pre-processed using Principal Component Analysis (PCA) to reduce the number of dimensions. The author demonstrates how this method can effectively recognize 3D objects despite a small expected number of photons. This may be important for low light applications in medical or other settings.

Since the performance of a character recognition system is mainly determined by the classifier, we introduce one that is especially tailored to our application. Working with 100 different classes, the most important properties of a reliable classifier are a high generalization capability, robustness to noise and classification speed. For this reason, we designed a classifier that is a combination of two types of classifiers, in which the advantages of both are united. The fundamental structure is given by a decision tree that has in its nodes either a support vector machine or an artificial neural network. The performance of this classifier is experimentally proven and the results are compared with both individual classifier types.

This paper introduces a novel on-line video browsing and retrieval platform, so-called OVIDIUS (On-line VIDeo Indexing Universal System). In contrast with traditional and commercial video retrieval platforms, where video content is treated in a more or less monolithic manner (i.e. with global descriptions associated with the whole document), the proposed approach makes it possible to browse and access video content in a finer, per-segment basis. The hierarchical metadata structure exploits the ISO/MPEG-7 approach for structural description of video content, which provides a multi-granular, hierarchical framework for heterogeneous metadata fusion. The issues of content interaction and visualization, which are of highest relevance in both annotation and metadata exploitation stages are also addressed. Our innovative approach makes it possible to quickly provide a comprehensive overview of complex video documents with a minimal time and interaction effort. The developed approach shows all its pertinence within a multi-terminal context and in particular for video access from mobile devices.

The Complex Ambiguity Function (CAF) used in emitter location measurement is a 2-dimensional complex-valued function of time-difference-of-arrival (TDOA) and frequency-difference-of-arrival (FDOA). In classical TDOA/FDOA systems, pairs of sensors share data (using compression) to compute the CAF, which is then used to estimate the TDOA/FDOA for each pair; the sets of TDOA/FDOA measurements are then transmitted to a common site where they are fused into an emitter location. However, in some recently published methods for improved emitter location methods, it has been proposed that after each pair of sensors computes the CAF it is the entire CAFs that should be shared rather than the extracted TDOA/FDOA estimates. This leads to a need for methods to compress the CAFs. Because a CAF is a 2-D functions it can be thought of as a form of image - albeit, a complex-valued image. We apply and appropriately modify the Embedded Zerotree Wavelet (EZW) to compress the Ambiguity Function. Several techniques are analyzed to exploit the correlation between the imaginary part and real part of Ambiguity Function and comparisons are made between the approaches. The impact of such compression on the overall location accuracy is assessed via simulations.

A discrete Fourier transform (DFT) or the closely related discrete cosine transform (DCT) is often employed as part of a data compression scheme. This paper presents a fast partial Fourier transform (FPFT) algorithm that is useful for calculating a subset of M Fourier transform coefficients for a data set comprised of N points (M < N). This algorithm reduces to the standard DFT when M = 1 and it reduces to the radix-2, decimation-in-time FFT when M = N and N is a power of 2. The DFT requires on the order of MN complex floating point multiplications to calculate M coefficients for N data points, a complete FFT requires on the order of (N/2)log₂N multiplications independent of M, and the new FPFT algorithm requires on the order of (N/2)log₂M + N multiplications. The FPFT algorithm introduced in this paper could be readily adapted to parallel processing. In addition to data compression, the FPFT algorithm described in this paper might be useful for very narrow band filter operations that pass only a small number of non-zero frequency coefficients such that M << N.

Algorithms for compression of hyperspectral data are commonly evaluated on a readily available collection of Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) images. The calibrated images collected in 1997 show sample value distributions which contain artificial regularities introduced by the conversion of raw data values to radiance units. Being optimal on images having flat histograms, classical DPCM methods do not work in their best conditions. Conversely, the lower lossless bit rates are achieved by algorithms based on lookup-table (LUT) that significantly exploit these artifacts. This singular behavior has not been widely reported and may not be widely recognized. The main consequence is these performances can be misleading if they are extrapolated to images that lack such artifacts. In fact, LUT-based algorithms are not able to achieve the best compression performances on a set of more recent (2006) AVIRIS images that do not contain appreciable calibration-induced artifacts. This is due to a different scaling factor in the calibration procedure. Goal of this paper is to provide a thorough comparison of advanced classical DPCM and LUT-based methods both on the 1997 and the 2006 AVIRIS datasets. In the 2006 data set, both calibrated data (radiances) and raw data (digital counts) have been compressed. Results strengthen the conclusion that even the most developed LUT-based methods do not show improvements over the state of the art when calibration induced artifacts are missing. Concerning classical DPCMs, the methods based on a classified spectral prediction, whose idea was originally developed by the authors in 2001, provide the best compression results.

In many physical effects and sensors in biomedical or engineering fields, it is often the case that some small non-linear characteristics are contained in the system. The previous paper treats approximation of non-linear filter bank.¹⁰ But, a running approximation is not treated. In this paper, we establish a theory of a favorable interpolation approximation of running filter banks with non-linear analysis filters based on the one-to-one correspondence between errors in a wide but limited volume and a certain small volume in the variable domain. Some additional considerations about the optimum interpolation approximation are presented also.

In this paper, diabetic retinopathy is chosen for a sample target image to demonstrate the effectiveness of image enlargement through pixel duplication in identifying regions of interest. Pixel duplication is presented as a simpler alternative to data interpolation techniques for detecting small structures in the images. A comparative analysis is performed on different image processing schemes applied to both original and pixel-duplicated images. Structures of interest are detected and and classification parameters optimized for minimum false positive detection in the original and enlarged retinal pictures. The error analysis demonstrates the advantages as well as shortcomings of pixel duplication in image enhancement when spatial averaging operations (smoothing filters) are also applied.

Image segmentation is one of the important applications in computer vision applications. In this paper, we present an image registration method that stiches multiple images into one complete view. Also, we demonstrate how image segmentation is used as an error metric to evaluate image registration. This paper explains about the error analysis using pattern recognition algorithm such as watershed algorithm for calculating the error for image registration applications. In this paper, we compare pixel intensity-based error metric with object-based error metric for evaluating the registration results. We explain in which situation pattern recognition algorithm is superior to other conventional algorithm such as mean square error.

We live in such perceptually rich natural and manmade environments that detection and recognition of objects is mediated cerebrally by attentional filtering, in order to separate objects of interest from background clutter. In computer models of the human visual system, attentional filtering is often restricted to early processing, where areas of interest (AOIs) are delineated around anomalies of interest, then the pixels within each AOI's subtense are isolated for later processing. In contrast, the human visual system concurrently detects many targets at multiple levels (e.g., retinal center-surround filters, ganglion layer feature detectors, post-retinal spatial filtering, and cortical detection / filtering of features and objects, to name but a few processes). Intracranial attentional filtering appears to play multiple roles, including clutter filtration at all levels of processing - thus, we process individual retinal cell responses, early filtering response, and so forth, on up to the filtering of objects at high levels of semantic complexity. Computationally, image compression techniques have progressed from emphasizing pixels, to considering regions of pixels as foci of computational interest. In more recent research, object-based compression has been investigated with varying rate-distortion performance and computational efficiency. Codecs have been developed for a wide variety of applications, although the majority of compression and decompression transforms continue to concentrate on region- and pixel-based processing, in part because of computational convenience. It is interesting to note that a growing body of research has emphasized the detection and representation of small features in relationship to their surrounding environment, which has occasionally been called semantic compression. In this paper, we overview different types of semantic compression approaches, with particular interest in high-level compression algorithms. Various algorithms and approaches are considered, ranging from low-level semantic compression for text and database compaction, to high-level semantic analysis of images or video in which objects of interest have been detected, segmented, and represented compactly to facilitate indexing. In particular, we overview previous work in semantic pattern recognition, and how this has been applied to object-based compression. Discussion centers on lossless versus lossy transformations, quality of service in lossy compression, and computational efficiency.

In this research, a new combination discrete cosine transform (DCT) and discrete wavelet transform (DWT) based watermarking system is studied. The first embedded watermark is encrypted using a chaos function, specifically the Lorenz function, based key to further conceal the data. A chaos based embedded watermark method in the DCT domain with blind watermark identification is developed and tested. Next the system is modified to utilize a second watermark embedding and identification/detection process. The second uses a pseudo random number generated (PRNG) watermark that is developed with a function of a Lorenz attractor data point as the seed state for the PRNG watermark in the detection process in the DWT domain. The efficacy of the DCT based technique, DWT based method as well as the combined DCT and DWT method is then compared to a previous techniques such as a NN based DWT based watermark embedding and identification. The three studied methods are subjected to a subset of the Checkmark attacks. Results for projection, shearing, warping, linear distortions, and Wiener filtering attacks are shown for the DWT embedded case.

Analysis of footwear under the Harmonized Tariff Schedule of the United States (HTSUS) is partly based on identifying the boundary ("parting line") between the "external surface area upper" (ESAU) and the sample's sole. Often, that boundary is obscured. We establish the parting line as the curved intersection between the sample outer surface and its insole surface. The outer surface is determined by discrete point cloud coordinates obtained using a laser scanner. The insole surface is defined by point cloud data, obtained using a touch probe device-a coordinate measuring machine (CMM). Because these point cloud data sets do not overlap spatially, a polynomial surface is fitted to the insole data and extended to intersect a mesh fitted to the outer surface point cloud. This line of intersection defines the ESAU boundary, permitting further fractional area calculations to proceed. The defined parting line location is sensitive to the polynomial used to fit experimental data. Extrapolation to the intersection with the ESAU can heighten this sensitivity. We discuss a methodology for transforming these data into a common reference frame. Three scenarios are considered: measurement error in point cloud coordinates, from fitting a polynomial surface to a point cloud then extrapolating beyond the data set, and error from reference frame transformation. These error sources can influence calculated surface areas. We describe experiments to assess error magnitude, the sensitivity of calculated results on these errors, and minimizing error impact on calculated quantities. Ultimately, we must ensure that statistical error from these procedures is minimized and within acceptance criteria.

This paper discusses the error analysis and performance estimation between two different mathematical methods for registering a sequence of images taken by an airborne sensor. Here both methods use homography matrices to obtain the panoramic image, but they use different mathematical techniques to obtain the same result. In Method-I, we use Discrete Linear Transform and Singular Value Decomposition to obtain the homographies and in Method-II we use the Levenberg-Marquardt algorithm as iterative technique to re-estimate the homography in order to obtain the same panoramic image. These two methods are analyzed, compared based on reliability and robustness of registration. We also compare their performance using an error metric that compares their registration accuracies with respect to ground truth. Our results demonstrate that Levenberg-Marquardt algorithm clearly outperforms Discrete Linear Transform algorithm.

In this study, we present spectral measurements of corn chlorophyll content in Changchun (eight times in 2003) and Hailun (five time in 2004), both of which lie in the Songnen Plain, China. Corn canopy reflectance and its derivative reflectance were subsequently used in a linear regression analysis against Chl-a concentration on one by one spectral reflectance. It was found that determination coefficient for Chl-a concentration was high in blue, red and near infrared spectral region, and it was low in green and red edge spectral region, however Chl-a concentration obtained its high determination coefficient in blue, green and red edge spectral region, especially in red edge region with derivative reflectance. Regression models were established based upon 6 spectral vegetation indices and wavelet coefficient, reflectance principal components as well. It was found that wavelet transforms is an effective method of hyperspectral reflectance feature extraction for corn Chl-a estimation, and the best multivariable regressions obtain determination coefficient (R²) up to 0.87 for Chl-a concentration. Finally, neural network algorithms with both specific band reflectance and wavelet coefficient as input variables were applied to estimate corn chlorophyll concentration. The results indicate that estimation accuracy improved with nodes number increasing in the hidden layer, and neural network performs better with wavelet coefficient than that with specific band reflectance as input variables, determination coefficient was up to 0.96 for Chl-a concentration. Further studies are still needed to refine the methods for determining and estimating corn bio-physical/chemical parameters or other vegetation as well in the future.

In this paper, we proposed a highly enhanced compression scheme of the integral imaging (InIm) applying to the residual image (RI) array transformed from Sub-Image Array (SIA). In the pickup process, the object through the virtual pinhole array is recorded in a computer as Elemental Image Array (EIA). Then, SIA is generated from EIA. It provides enhanced compression efficiency by improving the similarity between sub-images (SIs). In the proposed scheme, the reference image is the first one of rearranged SIs based on spiral scanning topology and the RIs are generated from difference between reference image and SIs. Finally, we model the reference image and RIs in InIm as consecutive frames in a moving picture. Therefore, the video compression scheme such as MPEG-4 can be applied to data reduction of the consecutive frames. Experimental results are presented to illustrate the image quality of MPEG-4 and JPEG with the same compression rate. Some experiment is carried out and compression efficiency of the proposed scheme has been improved 296.02% in average according to the each quality factor.

In this work we describe an experimental technique to measure the kinetics of heterogeneous chemical reactions in aqueous electrolytes, trough the continuous measurement of changes in the optical properties of the aqueous media using laser reflectometry near the critical angle. Valuable information of the reaction rates that are taking place in the aqueous side of an electrolyte-glass interface can be obtained if accurate relationships between changes in the reflectance, refraction index, and content of chemical species are established. We report some of these relationships, the experimental methodology, and the mathematical models needed to measure continuously chemical reaction rates, and apply them for the leaching of metallic copper and the dissolution of cupric salts in acidic media. The results show that laser reflectometry near the critical angle can be used as a suitable non-invasive accurate technique to measure in situ the rate of dissolution reactions.

With the improvements in range image registration techniques, this paper focuses on error analysis of two registration methods being generally applied in industry metrology including the algorithm comparison, matching error, computing complexity and different application areas. One method is iterative closest points, by which beautiful matching results with little error can be achieved. However some limitations influence its application in automatic and fast metrology. The other method is based on landmarks. We also present a algorithm for registering multiple range-images with non-coding landmarks, including the landmarks' auto-identification and sub-pixel location, 3D rigid motion, point pattern matching, global iterative optimization techniques et al. The registering results by the two methods are illustrated and a thorough error analysis is performed.

Several estimative factors of image quality have been developed for approaching the human perception objectively^1-3. We propose to take systematically distorted videos into the estimative factors and analyze the relationship between them. Several types of noise and noise weight were took into COSME standard video and verified the image quality estimative factors which were MSE (Mean Square Error), SSIM (Structural SIMilarity), CWSSIM (Complex Wavelet SSIM), PQR (Picture Quality Ratings) and DVQ (Digital Video Quality). The noise includes white noise, blur and luminance...etc. In the results, CWSSIM index has higher sensitivity at image structure and it could estimate the distorted videos which have the same noise type at the different levels. PQR is similar to CWSSIM, but the ratings of distribution were banded together; SSIM index divides the noise types into two groups and DVQ has linear relationship with MSE in the logarithmic scale.

An improved algorithm based on Harris Corner Detection is presented. It has good noise resistive and can extract feature point on edge precisely and reduce the effect of noise and isolated pixels. To improve the Harris corner detection performance, several methods are involved. Instead of the Gaussian function, a B-spline smooth function is used. It can preserve more corner information and prevent corner position offsetting. To get more robust results, a modified corner response is also employed. In the corner matching process, the gradient information within a 5×5 window is used to perform a more robust matching. By an improved image matching algorithm, it can match the feature point precisely and reduce the incorrect matches. After that the images are aligned, the image of the overlapping band is smoothed with fusion algorithm to obtain a seamless mosaic image.