The handover of this journal from Brian Thompson re-sembles an in-flight change of pilots. I have to learn and modify the procedures while the flight continues. Brian and I would hope that the passengers would never notice that there has been any change made in the cockpit. It turns out, however, the craft is also being modified in-flight. In fact, a second craft is being launched and you will get your choice of which one you want to fly on. Then there is the weather forecast...!

Over the last few years, tremendous advances have been made for both adaptive and nonadaptive recognition techniques, especially for image processing and pattern recognition applications.

A widely used approach to hyperspectral image classification is to model a mixed-pixel vector as a linear superposition of substances resident in a pixel with additive Gaussian noise. Using this linear mixture model many image processing techniques can be applied, such as linear unmixing or orthogonal subspace projection. However, a third source not considered in this model, called interference (clutter or structured noise), may sometimes give rise to more serious signal deterioration than the additive noise. We address this issue by introducing the interference into the linear mixture model. Including interference in the model enables us to treat the interference as another undesired source, like a passive jammer, so that it can be eliminated prior to detection and classification. This is particularly useful for hyperspectral images, which tend to have a high SNR but a low signal-to-interference ratio with the interference difficult to identify. To find and reject interference, we propose an unsupervised vector quantization-based interference rejection (UIR) approach in conjunction with either an orthogonal subspace projection (OSP) or an oblique subspace projection (OBSP) to simultaneously project a pixel into signature space as well as to null out interference. Since there is no prior knowledge about the interference, the UIR is implemented in an unsupervised manner to generate the desired interference clusters so that they can be annihilated by the OSP or OBSP. The proposed approach is shown by evaluation with Hyperspectral Digital Imagery Collection Experiment (HYDICE) data to exhibit considerable improvement in comparison to linear unmixing or the OSP where interference is not considered.

A series of psychophysical trials is conducted to study the human perception of images of various shapes that are degraded by various combinations of Gaussian blurring, sampling, and fixed pattern noise. These images were created using computer programs designed to simulate the types of image degradation associated with thermal imagers that employ focal plane arrays (FPAs) of detectors. A total of 46 observers participated in these trials, during which over 130,000 observations were recorded. The responses collected during the trials are used to develop an empirical model of human vision applicable to 2-D images having a gray scale, but not including other colors. This model can be used to predict the probability of a human observer correctly distinguishing between images of two different objects.

A statistical spectral band selection procedure and classifiers for an active multispectral laser radar (LADAR) sensor are described. The sensor will operate in the 1 to 5 xm wavelength region. The algorithms proposed are tested using library reflectance spectra for some representative background materials. The material classes considered include both natural (vegetation and soil) and man-made (camouflage cloth and tar-asphalt). The analysis includes noise statistics due to Gaussian receiver noise and target induced speckle variations in the LADAR return signal intensity. The results of this analysis are then directly applied to an artificially generated spatial template of a scene consisting of these four material classes. The performance of four different classifier algorithms, which include a minimum distance classifier, a log- domain minimum distance classifier, a Bayes speckle-only classifier, and a Bayes speckle-Gaussian classifier, are evaluated. We show that the Bayesian classifier designed for speckle and Gaussian noise statistics outperforms the other classifiers. Our results also indicate that even when exact knowledge of the observation model is available, the classifier performance for speckled images can be poor unless the number of integrated speckle cells is large.

Three-dimensional object classification from 2-D IR images is shown. The wavelet transform is used for edge detection. Edge tracking is used for removing noise effectively in the wavelet transform. The invariant Fourier descriptor is used to describe the contour curves. Invariance under out-of-plane rotation is achieved by the feature space trajectory neural network working as a classifier.

Neural network associative memory has found wide applications in pattern recognition techniques. We propose an associative memory model for binary character recognition. The interconnection strengths of the memory are binary valued. The concept of sparse coding is used to enhance the storage efficiency of the model. The question of imposed preconditioning of pattern vectors, which is inherent in a sparsely coded conventional memory, is eliminated by using a multistep correlation technique and the ability of correct association is enhanced in a real-time application. A potential optoelectronic implementation of the proposed associative memory is also described. The learning and recall is possible by using digital optical matrix-vector multiplication, where full use of parallelism and connectivity of optics is made. A hologram is used in the experiment as a longterm memory (LTM) for storing all input information. The short-term memory or the interconnection weight matrix required during the recall process is configured by retrieving the necessary information from the holographic LTM.

The architecture and mathematical analysis of a new multichannel multistage holographic optical random access memory (HORAM) architecture and an experimental demonstration of its feasibility are presented. The new HORAM can be used for ultra-high-capacity storage and high-speed random retrieval of information. A two-stage HORAM, using a Dammann grating and a multifocus holographic lens, clearly shows the capability of storing 2000 holographic matched filters. The functional requirements for key optical elements including laser sources, spatial light modulators, and electro-optical shutters for making a desired practical and compact HORAM are described. The potential extension of the HORAM system for multiple-channel optical pattern recognition, classification, and image restoration are described.

The recognition of objects using target acquisition systems is modeled by a sensor's minimum resolvable temperature (MRT), the Johnson criteria, atmospherics, and object specifics. Collectively, these three characteristics provide an acquisition model for estimating the probability of object recognition (and detection, identification) as a function of sensor-to-object range. This technique is called the probabilities of discrimination. When quantifying the performance of intelligence- surveillance-reconnaissance (ISR) systems, object recognition is assessed using the National Imagery Interpretability Scale (NIIRS). Each NIIRS level corresponds to a different capacity for object recognition and is defined by a set of recognition criteria. The general image quality equation (GIQE) is the ISR sensor model that determines the expected NIIRS level of a sensor for a given set of sensor parameters. It is important that electro-optical sensor engineers understand both of these recognition models. The segregation between the target acquisition and ISR sensor communities is becoming less sharp as ISR sensors are beginning to be used for target acquisition purposes and visa versa. Network and wireless communication advances are providing the means for dual exploitation of these systems. Descriptions of these two recognition models, probabilities of discrimination, and the GIQE are provided. The two models are applied to example systems. Finally, the two models are compared and contrasted.

We describe a model-based image analysis system that automatically estimates the 3-D orientation vectors of satellites and their subcomponents by analyzing images obtained from a ground-based optical surveillance system. We adopt a two-step approach: pose estimates are derived from comparisons with a model database and pose refinements are derived from photogrammetric information. The model database is formed by representing each available training image by a set of derived geometric primitives. To obtain fast access to the model database and to increase the probability of early successful matching, a novel index-hashing method is introduced. An affine point-matching method is also introduced for improving system performance on a wide variety of satellite shapes. We present recent results, which include our efforts at isolating and estimating orientation vectors from degraded imagery on a significant database of satellites.

A technique based on information theory principles for the assessment of the degree of dissimilarity between binary images is proposed. The obtained results are then used to compute the optimal resolution needed for distinction between two images. We first describe how the image overall information content can be defined, identified, assessed, and analyzed. This concept is then extended to the measurement of the overall information content of more than one image. This information measure is then shown to be related to recognition information, which is a measure of the degree of dissimilarity between images. Recognition information provides a reliable way of evaluating the optimal scan resolution required for unambiguous recognition of an image. To demonstrate the feasibility of this technique, a series of tests are carried out on typed characters. The results illustrate the power of information assessment techniques in analyzing the recognition capabilities of a pattern recognition system.

When radiologists search a medical x-ray image for an abnormality, their eyes often fixate and refixate the true target, dwelling on it for prolonged times, often without recognizing that they have discovered the object of search. Monitoring the eye position of the radiologist provides the x and y coordinates of the dwelling location. This location can be superimposed on the image and dynamically fed back to the radiologist for reevaluation. When this is done, the probability of recognizing and reporting an abnormality is shown to be enhanced significantly. An increase of 20% in observer performance is observed for radiologists searching chest images for tumors after receiving perceptual feedback compared to a second look without perceptual feedback. The truepositive rate increased and the false-positive rate decreased. Perceptual feedback represents a potentially significant technique for enhancing lesion recognition in radiology.]

We describe a general approach to the representation and recognition of 3-D objects as it applies to automatic target recognition (ATR) tasks. The method is based on locally adaptive target segmentation, neural network classifier design, and a novel view selection mechanism that develops ‘‘visual filters’’ responsive to specific target classes that encode the complete viewing sphere with a small number of prototypical examples. The optimal set of visual filters is found via a crossvalidation-like data reduction algorithm used to train banks of backpropagation (BP) neural networks. To improve recognition accuracy under noisy or occluded conditions, as well as to eliminate false alarms, the proposed recognition system employs a temporal evidence integration technique that enables tracking and lock-on even when both targets and camera move. Experimental results on synthetic and real-world imagery demonstrate the feasibility of our approach.

Robust pattern recognition within the Bayesian framework for scene segmentation/boundary detection is often hampered by the presence of textures within natural images. To improve segmentation/ boundary detection on natural images, it is necessary to combine multiple features effectively. Two algorithms for combining both color and texture features to assist boundary detection processes are introduced. One combines features through the surface processes and the other through the line processes. The algorithms can be generalized for combining any number of feature sets.

Target tracking research has been of interest to several different groups of researchers from different perspectives. An event of perhaps greatest importance in the history and development of target tracking research is the new trend in the architectural revolution in current algorithms and techniques that are used for target tracking, i.e., the advent of neural networks and their applications to nonlinear dynamical systems. It is established in the literature that the mathematical complexity of the state-of-the-art tracking algorithms has gone far beyond the computational power of conventional digital processors. Since the introduction of Kalman filtering, several powerful mathematical tools have been added to target tracking techniques, e.g., probabilistic data association, correlation and gating, evidential reasoning, etc. All these methods have one thing in common: they track targets rather differently from the way natural systems do. It is rather difficult to come up with a sound mathematical proof and verification of the concept for different parallel distributed architectures that seem appropriate for a general class of target tracking applications. However, the volume of contributions within the last decade in the application of various neural network architectures to different classes of target tracking scenarios can not simply be ignored. Therefore, the various neural-network-based tracking algorithms that have been introduced since 1986 are classified and addressed and their common views as well as their differences in results and in architectures are discussed. The role of mathematics in each of these algorithms and the extent that conventional methods are used in conjunction with the neural-network-based techniques are also addressed.

The possibility of using certain kinds of neural networks, shared weight neural networks, to recognize vehicles in IR images is investigated. We use the neural network as a preprocessor to find tank wheels in the images, followed by a more specific traditional search in the region(s) of interest found by the network; i.e., the neural network is used as a feature detector. In this way, although we train the networks on individual samples, we use the trained networks as nonlinear filters on entire images. The resulting system is tested on scale dependency and sensitivity to clutter in the background of images and performs adequately.

Composite classifiers that are constructed by combining a number of component classifiers are designed and evaluated on the problem of automatic target recognition (ATR) using forward-looking iR (FLIR) imagery. Two existing classifiers, one based on learning vector quantization and the other on modular neural networks, are used as the building blocks for our composite classifiers. A number of classifier fusion algorithms, which combine the outputs of all the component classifiers, and classifier selection algorithms, which use a cascade architecture that relies on a subset of the component classifiers, are analyzed. Each composite classifier is implemented and tested on a large data set of real FLIR images. The performance of the proposed composite classifiers are compared based on their classification ability and computational complexity. it is demonstrated that the composite classifier based on a cascade architecture greatly reduces computational complexity with a statistically insignificant decrease in performance in comparison to standard classifier fusion algorithms.

A new generalized algorithm is proposed to automatically extract a mouth boundary model from human face images. Such an algorithm can contribute to human face recognition and lip-reading-assisted speech recognition systems, in particular, and multimodal human computer interaction systems, in general. The new model is an iterative algorithm based on a hierarchical model adaptation scheme using deformable templates, as a generalization of some of the previous works. The role of prior knowledge is essential for perceptual organization in the algorithm. The prior knowledge about the mouth shape is used to define and initialize a primary deformable model. Each primary boundary curve of a mouth is formed on three control points, including two mouth corners, whose locations are optimized using a primary energy functional. This energy functional essentially captures the knowledge of the mouth shape to perceptually organize image information. The primary model is finely tuned in the second stage of optimization algorithm using a generalized secondary energy functional. Basically each boundary curve is finely tuned using more control points. The primary model is replaced by an adapted model if there is an increase in the secondary energy functional. The results indicate that the new model adaptation technique satisfactorily generalizes the mouth boundary model extraction in an automated fashion.

We propose an enabling algorithm for automatic target detection wherein the targets are almost similar but differ in fine details. Multiple translated and scaled target images are processed in the Mellin transform domain and subsequently detected using a learning vector quantization (LVQ) neural network.

An optical diffraction method is described for inspecting periodic structures such as combs or semiconductor leads. Coherent light passing between the prongs of the structure self-interfere at the fractional Talbot plane to provide a simple method of inspection. Computer simulation and laboratory experiments show the viability of this approach. The theory assumes infinite structures. In practice, large end effect signals arise due to the finiteness of the periodic structure. A neural network is demonstrated that learns to distinguish end effect signals from prong damage signals. The variability of the measuring process in a production environment makes neural networks an appropriate approach for this task.

The approximation capabilities of two different four-layered neural networks are studied. First, a network with the backpropagation algorithm is analyzed, and its error surface, convergence properties, and network design are considered. An alternative to the backpropagation approach is presented, namely, we construct a network that uses a one- pass algorithm. We show that the proposed network can correctly classify N different patterns with 4 ?N?3 hidden units. We also show that an arbitrarily small approximation error can be obtained for this network by adjusting the appropriate parameters.

Increasing emphasis is being placed on detecting damage in civil structures in order to estimate remaining life, forestall catastrophic failure, and make more effective use of limited maintenance resources. Image analysis represents a very promising tool for the rapid inspection of large structures that have significant areas unavailable for direct inspection. We describe a straightforward biologically inspired artificial life implementation methodology whereby cellular automata (CAs) can be used to analyze multispectral images of large civil structures to identify structural damage. The size of the CA population can be correlated with structural lifetime, required time to detailed inspection, and other engineering parameters. A simple application of the methodology is described and the results of its operation are presented.

We address a new general method for linear and nonlinear feature extraction for simultaneous representation and classification. We call this approach the maximum representation and discrimination feature (MRDF) method. We develop a novel nonlinear eigenfeature (NLEF) extraction technique to represent data with closed-form solutions and use it to derive a nonlinear MRDF algorithm. Results of the MRDF method on synthetic databases are shown and compared with results from standard Fukunaga-Koontz transform and Fisher discriminant function methods. The method is also applied to an automated product inspection problem (discrimination) and for classification and pose estimation of two similar objects under 3-D aspect angle variations (representation and discrimination).

The feature space trajectory (FST) neural net is used for classification and pose estimation of the contents of regions of interest. The FST provides an attractive representation of distorted objects that overcomes problems present in other classifiers. We discuss its use in rejecting clutter inputs, selecting the number and identity of the aspect views most necessary to represent an object, and to distinguish between two objects, temporal image processing, automatic target recognition, and active vision.

we demonstrate an optical implementation of real-time preprocessing using an Epson liquid crystal television (LCTV) working as a spatial matched filter in an optical correlation setup. Two filters for smoothing and edge enhancement using both the hybrid modulations of the programmable LCTV are studied and compared.

A simple acousto-optic system is described that is capable of demodulating frequency- and phase-modulated signals using a bicell photodetector. A summary of an analytical study of the system is provided, describing its main performance characteristics. Results from a Mathcad model of the system are presented showing that, by taking the sum and difference of the outputs from the two bicell elements, it is possible to identify and demodulate a wide range of different types of modulation without any a priori information. Thus the technique is well suited for surveillance receiver and electronic support measures (ESM) applications. In particular, it is shown that biphase shift keying (BPSK), quaternary phase shift keying (QPSK), offset-keyed QPSK, differential QPSK, continuous phase frequency shift keying (FSK) [including minimum shift keying (MSK)], and ordinary FSK all give unique and easily identifiable signatures. Experimental results are presented, which are in good agreement with both theoretical and modeled results.

A spatial light modulator is often used as the basic element in real-time recognition architectures. However, many of the devices suffer from poor uniformity and/or nonlinearities that affect the overall performance of any real-time recognition scheme. We propose using optical feedback as a method of improving device performance. We demonstrate both positive and negative feedback separately through a mixture of on-board electronic processors and optical polarization intensity encoding. Placing the spatial light modulator in a positive feedback system causes the system to act like a memory. Placing the spatial light modulator in a negative feedback system makes the spatial light modulator response more analog, which in turn leads to improved uniformity.

The sensitivity across a solid state detector array varies as a result of differences in transmission, diffusion and scattering properties over the sensor. This variation will occur over a range of scale lengths and its knowledge is of importance for improved device design and in a variety of applications, for example, event centroiding in photon counting systems. A measurement of the sensitivity variation on a subpixel scale for a two-phase front-illuminated CCD is reported. The measurement is made using a scanning reflection microscope. A variation in sensitivity between the phases within a pixel is clearly observed, as well as variations on a much smaller spatial scale.

An acousto-optic correlator (AOC) associated with pseudorandom sequences (PRS), as used in spread spectrum receivers, is considered. Simple expressions are derived for the output signal-to-noise ratio of an AOC in two cases: when the frequency response H(f) of the correlator is equal to rect(f/?f) or equal to sinc(?f/?f), where rect(x) is the rectangle function, sinc(x) = sinx/x and ?f is the correlator bandwidth. These simple expressions are given in terms of the ratio a of the code bit rate Fc of the PRS to the bandwidth Af of the AOC. It will be shown that when H(f) = rect(f/?f) and a is greater than 2, a loss results in the output signal-to-noise ratio approximately equal to 1/? with respect to an ideal (infinite bandwidth) AOC. When H(f) = sinc(?f/?f) the loss equals 1.5/?.

Performance analysis of M-ary orthogonal signaling in multimode graded index fiber optic channels is considered. An APD is used for detection and {0,1%}-valued optical orthogonal codes with auto- and cross-correlation coefficients of unity are used as addressing sequences in the analysis. Signal-dependent shot noise, modal noise in the fiber, receiver thermal noise, and multiple-user interference noise are included in evaluating the BER probability. The BER probability is evaluated by assuming the receiver output statistics are Gaussian. The results of the numerical analysis reveal the following: (1) At fixed bit rate and transmitted power, the BER probability achieved byM-ary orthogonal signaling is significantly lower than that of the binary signaling scheme, (2) the system can withstand a higher level of modal noise without significantly degrading the system for a large number of active users, (3) the effect of modal noise on the BER probability can be further reduced by using higher values of M compared to the baseband binary case, and (4) an optimum value exists for the mean gain of APD at which the BER probability is minimum.

A polarization-insensitive fiber optic acoustic sensor is constructed with a 3x3 coupler and Faraday rotator mirrors. This system deals with the polarization-induced signal fading by a passive method. The method in eliminating polarization-induced signal fading and the output intensities of this system are described and demonstrated. The homodyne symmetric demodulation system used to recover the sensing signal is also analyzed in detail with experimental verification. The sensing head of this sensor is made of single mode optical fiber wound into a cylindrical shell with cast polyurethane rubber as the acoustic window. The sensitivity of this sensor is measured in anechoic water tank. The results show that this sensor has a stable receiving sensitivity of - 145.5 dB re 1V/?Pa and the bandwidth extends from 2 to 50 kHz. This sensor configuration can be used as a high sensitivity, wide band underwater acoustic sensor.

The recent advance in image intensifier based volume tomographic angiography has stirred a significant interest in the development of accurate and efficient algorithms to correct image intensifier distortions. An algorithm and its calibration protocol are developed to correct the distortion caused by the image intensifier–CCD camera imaging chain in a single step regardless of distortion sources. This algorithm first partitions a distorted image into small quadrilateral regions, then maps each region individually to the corrected image using the corresponding bilinear functions. The general applicable parameters of these bilinear functions can be obtained using a specially designed grid phantom. The algorithm and the calibration protocol are tested using this grid phantom on a newly built image intensifier based volume tomographic angiography imaging system. The maximum distortion error is reduced from 5.96 mm before correction to 0.05 mm after correction. Further validation of the proposed algorithm is performed through three-dimensional tomographic reconstructions of a vascular phantom and animal studies. The reconstruction results indicate that the accuracy of the algorithm and the calibration protocol is acceptable for volume tomographic angiography. This algorithm can also be applied to radiation therapy and stereosurgical treatment planning where image intensifier distortions must be corrected.

An image modification technique for assembling subpictures into a seamless picture is explored. The technique can be applied even when multiple subpictures of any shape and any size are overlapping with each other. To determine image modification functions, we introduce the concept of energy minimization which has been widely used for image processing and understanding. We use two kinds of energy: one represents the smoothness and the other represents the fitness of the image modification function. Image modification functions are defined so that the sum of those energies becomes minimum. Experimental results are shown to demonstrate the feasibility of the proposed method.

A new clustering algorithm derived from the Markovian model of the gravitational clustering concept is proposed that works in the RGB measurement space for color image. To enable the model to be applicable in image segmentation, the new algorithm imposes a clustering constraint at each clustering iteration to control and determine the formation of multiple clusters. Using such constraint to limit the attraction between clusters, a termination condition can be easily defined. The new clustering algorithm is evaluated objectively and subjectively on three different images against the K-means clustering algorithm, the recursive histogram clustering algorithm for color (also known as the multi-spectral thresholding), the Hedley-Yan algorithm, and the widely used seedbased region growing algorithm. From the evaluation, it is observed that the new algorithm exhibits the following characteristics: (1) its objective measurement figures are comparable with the best in this group of segmentation algorithms; (2) it generates smoother region boundaries; (3) the segmented boundaries align closely with the original boundaries; and (4) it forms a meaningful number of segmented regions.

Two parameters for quantifying optical power distribution and lens symmetry are proposed as a tool to aid the lens designer in the search of optimum configurations.

A phase-shifting projection moire´ method particularly intended for high-speed three-dimensional inspection of fine objects is presented. Emphasis is on realization of phase-shifting fringe analysis in projection moire´ topography using a set of line grating pairs designed to provide different phase shifts in sequence. Further, a time-integral fringe capturing scheme is devised to remove undesirable high-frequency original grating patterns in real time without time-consuming software image processing. Finally, the performances of the proposed method are discussed with measurement results.

A novel implementation method based on the principle of optical interference to realize all-optical logic gates without the need of nonlinear optical materials is proposed. A general design rule of implementing the logic gates is obtained from the dimensionless intensity distribution of interference fringes formed by multiple equally spaced slits. Various designs of logic gates such as NAND, XOR, and so on are presented. The feasibility of implementing the logic gates and their exemplary applications to edge detection are demonstrated in several proof-of-principle experiments. System issues are discussed.

We present an extension of the equivalent optical waveguide method (EOWM) to analyze multimode optical waveguides with arbitrary refractive index profiles. Effective indices, propagation constants, and coupling/switching properties of multimode planar and channel optical waveguides can be easily obtained. This extension involves modeling the original optical waveguide by means of an equivalent optical waveguide for each guided mode. The effective indices are evaluated by applying the asymptotic effective index method (AEIM). Numerical values were obtained and compared with those presented by other authors. In all cases, good convergence and accuracy was noticed.

Starting from the variational form of Fermat's principle, we identify the optical Lagrangian and obtain the standard form of the ray equations in inhomogeneous optical media. Specializing to cylindrically symmetric media, we define the paraxial ray approximation through the exact first integral of the ray equation. This enables us not only to relate the constants of integration to the boundary conditions at the entry point of the ray into the medium, but more importantly furnishes a systematic way of introducing corrections to the paraxial approximation. We also exhibit power-law refractive index profiles, which result in periodic ray trajectories.

The objective is to investigate a process by which micrometer scale topographical changes are produced on thin chromium films using a pulsed Nd:YLF laser. The surface of chromium films is altered through laser-induced solid-liquid phase transformation and fluid flow. Experimental parametric studies are conducted to correlate the laser parameters with the topography of the laser irradiated surfaces. Experimental and analytical work is also performed to study the transport phenomena involved in the process. A numerical finite-element analysis is used to simulate the transient field variables. A nanosecond-time-resolution, fast photography system is constructed to capture the phase change and the fluid flow occurring at the target surface. The experimental and the numerical studies showed that the surface topography change was caused by the laser-induced surface-tension-driven flow, and the recoil pressure due to surface evaporation had negligible effect on the topography variation.

In this paper it is shown that for E-O systems operating in the solar blind ultraviolet (SBUV) band, one has to consider the clutter noise since in most cases it will be the dominant factor in the sensor performance. Such clutter in various rural populated areas were sampled using a SBUV photon counter whose data was analyzed in terms of statistical and temporal characteristics. It is shown that the clutter in the SBUV can reach very high levels of counts with relatively long duration and that the properties of the so-called clutter "sources" are very similar. The temporal autocorrelation function can be approximated by two exponential functions, the first one with a short time constant and the second one with a relatively long time constant. Periodic modulation was observed in some cases. Its source is not explained at the present time.

Details are presented of the proposal and development of a novel approach to loop filtering for hybrid motion compensated/ differential pulse code modulation/discrete cosine transform (MC/DpCM/ DCT) codecs that seeks to challenge the restrictions placed on the minimum achievable level of the prediction error imposed by the quality of the MC prediction reference pictures. A framework for this loop filter is presented based on the theory of projections onto convex sets (POCS), which draws on knowledge of coding distortion characteristics. This framework is intended to be most beneficial where humanly perceived coding distortions are evident in the reconstructed sequence, and therefore would be most suited for low (constant) bit-rate applications or, generally, for situations where bit-allocation mechanisms are under stress.

In this paper, we describe a method for estimating and compensating 3D camera motion in image sequences for applications to video coding systems. To improve the coding efficiency, a two-stage motion compensation (MC) method is used, consisting of global MC for predicting camera motion and local MC for predicting object motion. For global MC, a new linear motion parameter model is presented to describe the camera motion parameters: zoom, pan, tilt, swing, and focal length. To estimate camera motion parameters based on the proposed model, the mixed least squares-total least squares (LS-TLS) method is used. The proposed estimation procedure consists of feature correspondence establishment, finding the parameters by fitting the correspondence data to the proposed model equation, and outliers rejection to reduce feature matching error. Unlike the existing linear techniques, the proposed method accurately estimates even the large rotation angles and the focal length. Experimental results show that the proposed method outperforms the conventional methods under identical conditions, especially for large rotation images.

In this paper, the term 3-D refers to true 3-D images seen with length, width, and height. Stereoscopic viewing systems for true 3-D images on desktop displays actually present two images on the screen.

The PDF contains the communication section which includes "Evidence of injection locking in a non-dispersive slave CO2 laser"