A deformable wavelet template (DWT) is proposed for object shape description in this research. Wavelet templates offer not only the global information at lower scales but also local features at higher differential scales. It provides a natural tool for multiresolution representation and can be used conveniently in a hierarchical matching procedure. We first address three main processing steps in the DWT-based ATR system for feature extraction. They are: (1) image preprocessing and target shape extraction, (2) shape feature normalization and (3) wavelet decomposition. Then, a multiscale matching procedure is discussed. The performance of the proposed algorithm is demonstrated with extensive experimental results.

This paper introduces a new algorithm for target recognition in FLIR sensor image data. Recognition of this type is difficult because FLIR targets often have dramatically different temperature profiles; they may have very low contrast; and their local statistics can often be very similar to those of the background. The approach presented in this paper is based on a cascade of stages involving successive applications of morphological processing, enhancement, and extraction operations. The operators and order of their application for the detection was developed to accentuate and localize the general features of FLIR image targets. Experimental evaluations indicate that the algorithm performs well, even for targets with mixed temperature distributions and those with low contrast.

Grenander's pattern theory offers a unified approach to characterizing variability in complex systems. Automatic target recognition systems for forward-looking infrared sensors must be robust to three kinds of variability: (1) Geometric variability--Target appearances vary with their orientations and positions; (2) Image variability--Target appearances vary with their thermodynamic state, and natural backgrounds consist of widely varying textures; (3) Complexity/scene variability--The number of targets encountered will not be known in advance, and targets may enter or leave the scene at random times. Pattern theoretic algorithms based on jump-diffusion processes which accommodate variabilities (1) and (3) have been proposed. The diffusions account for (1) by estimating positions and orientations, and the jumps account for (3) by adding and removing hypothesized targets and changing target types. Here we extend the work to better accommodate (2) by summarizing the thermodynamic state of targets with a parsimonious set of variables which become nuisance parameters in the Grenander/Bayesian formulation.

Low-contrast FLIR tank object detection is a difficulty. This paper presents a new method based on fractal geometry and multiscale analysis for the target detection. A new metric called multiscale fractal character vector which can distinguish man-made objects and natural scenes is defined. And then a segmentation algorithm based on this new metric is given. Finally, experimental results have shown our method can give better segmentation results than the usual segmentation method which is based on H parameter of only one scale image.

Automatic classification of surface ships by means of imaging sensors through the submarine's periscope is of interest to the naval underwater warfare center of the US Navy. In this paper we discuss a testbed designed for periscope video ship classification based on model-based automatic target recognition paradigm, will present the performance results for the application of some of the existing algorithms and will present a sequential tree based technique for ship recognition.

Automatic detection and recognition of targets by means of passive IR sensors suffer from limitations due to lack of sufficient contrast between the targets and their background, and among the facets of a target. In this paper the results of a suite of polarization-sensitive automatic target detection and recognition algorithms on sets of simulated and real polarimetric IR imagery are presented. A custom designed Polarimetric IR imaging sensor is used for collecting real polarimetric target data under a variety of conditions. Then a set of novel algorithms are designed and tested that uses the target and background Stokes parameters for detection, segmentation and classification of targets. The empirical performance results in terms of the probabilities of detection, false alarm rate, segmentation accuracy, and recognition probabilities as functions of number of pixels on target, aspect and depression angles and several background conditions (clutter densities) of applying this ATR algorithms on the polarimetric data and its comparison with a typical IR only ATR are demonstrated that shows that a noticeable improvement can be achieved.

A new synthetic ladar image generation tool has been developed for use in ATR algorithm development and evaluation. The image generation tool, called SYLVER, was designed originally for air-to-ground targeting scenarios and simulates the effects of arbitrary platform and target motion as well as sensor scanning. Particular attention has been given to realistic rendering in the vicinity of range discontinuities, since edges are important to many recognition algorithms. This paper presents an overview of the image generation approach and details the mathematical models employed. Example images are shown and characteristics of the simulated imagery are compared with those of field collected data.

A model-based vision (MBV) approach to automatic target cuing/recognition (ATC/R) using real infrared (IR) image data will be presented. The MBV-ATC/R is comprised of three parts: Focus of attention (FOA), indexing, and match/search. The FOA module analyzes the IR image and extracts (segments) regions of interest that may contain targets. The focus of this article will be on the FOA portion of the MBV-ATC/R approach. In particular, three methods of FOA will be optimized, compared, and fused. The first FOA module is a Least Asymmetrical Daubechies wavelet decomposition method. The second FOA module is a physiologically based Difference of Gaussians. The third FOA module is a Morphological hit- and-miss transform. The three FOA algorithms are individually optimized using a genetic algorithm. Then an adaptive pulse coupled neural network is used to fuse the results.

This paper describes a method of using elliptical features in model matching that forms the basis of a system for vehicle detection and classification. The novelty of the system is the employment of an algorithm that utilizes both line and ellipse features simultaneously. The baseline algorithm has been successfully used in images, which contain straight line features, to find a discrete correspondence between an object model and image features and to determine the pose of the object in the image relative to the camera. This research enhances the baseline algorithm by using elliptical image features to recognize circular objects in a model. Elliptical features show many desired properties in recognition. Utilizing these features not only increases the confidence level of detection and classification, but also provides the system with a good initial pose for a more robust performance.

The automatic recognition and interpretation of complex structures from imagery, e.g. radar or IR images, needs a structure-oriented analysis method. This paper describes a syntactic approach for object classification in radar and IR images, by which the geometric structures of targets are analyzed using a knowledge-based production system. Those objects which are describable by two- or three-dimensional models have to be acquired independently of changes of illumination, weather conditions or status of vegetation. Perspective distortions and model description errors also have to be tolerated. The aim of the implemented method is to classify the interesting targets in uncertain data with uncertain models.

While the laser radar systems have high performance at short ranges and low altitudes, the atmospheric effects have been the major constraints of detection and parameter estimation of laser pulses at long ranges and high altitudes. The turbulence which depends on different atmospheric states is hard to quantify due to the wavelength dependent effects of various conditions at different layers of the atmosphere. The turbulence may also be caused by interaction of the atmosphere with other objects, such as the vortex flow due to the aerodynamics of the air targets, or the nonlinear propagation characteristic of the high energy laser pulses. These adverse effects of the atmosphere have been limiting the usefulness of the laser radar systems for a wide range of applications. If the atmosphere is considered as a nonlinear media with nonuniform index of refraction, then it can be thought of as a nonlinear distributed lens under diffraction limited conditions. In this paper, a neural network modeling of the ionosphere layer is presented and the laser pulse is characterized by a set of input features. The transient CO₂ laser pulses is simulated to transmit through the atmosphere to a satellite-borne receiver. The satellite receiver model is composed of three stages, i.e., the filtering and processing of the ionospheric propagated waveform, the envelope extraction and channel simulation, and the detection and parameter estimation. The received signal is then evaluated against the background noise through Monte Carlo simulations.

In this paper the problem of estimation and tracking of geometric position of an object in a 3D space is considered using a network of sensors positioned at known points. Observations from each sensor include bearing and slant range of each object. Cramer-Rao error bound for estimation of target cartesian coordinates is derived and analyzed, in order to study the effects of measurement noise and sensor distribution in geometrical space. A basic structure for an adaptive algorithm is proposed for data acquisition and tracking.

In this paper, an effective method is proposed to compute the data association probabilities in the joint probabilistic data association (JPDA) method. In this method, the a posteriori probability of the origin of each measurement required in JPDA, is computed directly, based on an approximate formula. The Computational and memory requirements of this method decrease greatly compared to the original JPDA. Simulation results show that despite great reduction in computational complexity, the algorithm approximates an exact JPDA more accurately than previously proposed simplifying methods such as SPDA.

An adaptive scheme is introduced to detect and track moving targets under multiscenario environment including target in complex background, target is partial obstructed and multi- target or multi-bait. The performance modeling is important to do with the fact that it can enhance the ability to predict the target tracking system performance in scenarios where data is not available. A critical use of performance models is selecting proper algorithms and adaptation of algorithm parameters to meet different scenarios' changes. The performance model is built at algorithm level instead of system level. For each algorithm, its performance model was built based upon experimental design. Some image metrics are employed to describe scene variation, and integrated into performance model. In our experiments, with the controls of adaptive controller of our system based on algorithm performance model and expert knowledge, system can select proper algorithms, tune algorithm parameters, and target was tracked perfectly in total 400 frames.

We study the scattering interaction of electromagnetic pulses of short duration with a few targets. The targets are two spheres and a short cylinder made of metal, and they are buried at selected depths in dry sand contained in an indoor sandbox. The backscattered echoes are extracted by an impulse radar system playing the role of a ground penetrating radar (GPR). In general, multiple scattering between a buried target and the ground surface and scattering from discontinuities in the sand distort the returned echoes to the extent that target recognition by means of frequency signature is nearly impossible. These obstacles for successful target recognition can be counteracted by analyzing returned echoes by means of time-frequency distributions of the Wigner-type, or Cohen class, by which it is possible to study how each one of the target's signature features evolves in time. Numerous members of the Cohen class of time-frequency distributions have been proposed over the years, each with its own property of concentrating the features in time-frequency and ability of suppressing undesirable cross-terms interference. We examine how, and how well a few members of the Cohen class reveal the time-progression of each target's features. The time-frequency distributions we compare in this survey are the pseudo-Wigner distribution, the Choi-Williams distribution, the adaptive spectrogram, the cone-shaped distribution, the Gabor spectrogram, and the spectrogram. We discuss the ability of each method of analysis to extract and concentrate features of the signature of each target in time-frequency and to suppress undesirable interference from cross- terms or multiple scattering. The results serve to assess the possibility of identifying subsurface targets using a GPR.

Wave scattered by an object in a stratified medium is much different from that in a homogeneous medium. In a stratified medium, the scattered energy is not spread out spherically. Therefore, the scattered far-field consists of two components: free-wave far-field and guided-wave far-field. In many cases only free-wave far-field or only guided-wave far- field can be measured. Our research concerns the problems of determining the shape of scatterer from the incomplete far-field data. We have found that the choice of incident waves is very important in determination of the shape of the unknown object from only free-wave far-field or only guided-wave far-field. We obtained analytical relations of the object, incident wave and scattered free-wave far-field and guided-wave far-field. These relations are used in reconstruction of the shape of unknown object numerically.

A hybrid automatic target recognition system is presented that exploits advances in two new fields in detection theory and signal analysis. The first is in the area of Universal Classification that offers asymptotic optimal solutions to non-Gaussian properties of signals and the second is in the field of multi-resolution analysis (MRA) that uses the automatic feature isolating properties of the wavelet transform. The Universal Classifier is used as the first stage of a hybrid ATR system that efficiently shifts through large quantities of imagery locating regions of interest that contain `target-like' features. The target chips of interest are then passed through the MRA to be classified at the final stage. Wavelets are adequate to the study of unpredictable signals with both low frequency components and sharp transitions. As a result, there has been recent interest in applying this new signal processing field to the target recognition problem. But few have combined the natural feature extraction capability of time- frequency methods in the classification stage. In this approach, we utilize the sub-space `crystals' from a specific decomposition and operate a classification strategy against each crystal of the transform. The complete ATR system is presented as well as performance examples using both real synthetic aperture radar data and data generated using the Xpatch signature prediction code.

The recognition and discrimination of underwater targets by an active sonar array depends on its fundamental structural dynamics. The acoustic energy distribution projected from a dense, multi-element array is critically dependent on the effects of structural defects in the array and its interface on the water medium. Knowledge of these effects and defects is crucial in insuring correct array operation. Array dynamics are directly related to the mechanical and physical interfaces between transducers and their supporting and radiating structures. Such complex assemblies require sophisticated bonding and processing of compliant and rigid materials which determine the operational parameters of the sonar system. A nondestructive method for the visualization of anomalous structural characteristics would be of great value in both development and testing of sonar systems. Augmenting this with a method for visualizing and comparing the true near-field acoustic energy distribution of an active sonar array would also greatly increase the understanding of the contributions and effects of structural defects in the array and its interface to the water medium. Knowledge of the mechanisms for acoustic energy transfer through the sonar structure into the water medium can be derived from these visualizations and would enhance the analysis of returned signals for homing and target recognition in operation. Holographic interferometry has presented itself as a viable and useful method for the realization of this type of information.

We have examined a large set of dolphin-emitted acoustic pulses (`echolocation clicks') which were reflected from various elastic shells that were suspended, underwater, 4.5 m in front of the animal in a large test site in Kaneohe Bay. A carefully instrumented analog-to-digital system continuously captured the emitted clicks and also the returned, backscattered, echoes (A/D conversion at 500 kHz). Using standard conditioning techniques and food reinforces, the dolphin is taught to push an underwater paddle when the `correct' target--the one he has been trained to identify--is presented to him. He communicates to us his consistently correct identifying choices in this manner. We have examined many echoes returned by three types of cylindrical shells in both the time and frequency domains. We will show exactly how specific features observable in these displays are directly related to the physical characteristics of the shells. Our processing uses certain fundamental resonance principles to show which echo- features contain information about the size, shape, wall-thickness, and material composition of both the shell and its filler substance. In the same fashion that these resonance features give us the identifying characteristics of each shell, we believe they also give them to the dolphin. These echo features may tell him the target properties without any need for computations. We claim that this may be the fundamental physical explanation of the dolphin's amazing target-ID feats, upon which they base their recognition choices. Our claim may be substantiated by the detailed analysis of many typical echoes returned by various shells, when they are interrogated by several dolphins. Thus far, this analysis of many echoes from many shells has only been carried out for a single dolphin.

The objective of the paper is to present an effective method of combined statistical estimates for target parameter's estimation under Automatic Object Recognition. Introduced in the paper algorithm can be used in any applications dealing with different-accuracy measurements (possibly, optical or radar). Represented results prove that it is possible to use in calculating also the achieved low-accuracy data files. The proposed method of combined estimates gives final expression for calculating the unknown estimates, which are insensitive to considerably deviations in low-accuracy data files. Further the paper discusses the problems of estimate's accuracy controllability. In the case of estimate's accuracy restrictions are given the available win in the total number of needed measurements is analyzed. The experimental results are satisfactory and agree well with the reference solutions.

The application of correlation filters for automatic target recognition has been traditionally limited to the recognition of critical mobile targets. We demonstrate the use of these filters for `scene matching' in millimeter wave and LADAR imagery for recognizing high valued targets.

Invariant target representation is desirable in any target recognition system. The representation that captures the salient attributes of a target independent of the viewing geometry will reduce the target search space in the recognition system. Three dimensional moment invariants are specially useful due to the increased interest in the ATR community on synthetic aperture radar and laser radar that generates 3D snapshot of the real world as well as passive sensors that produce sequential imagery that represents a 3D view of the sensed scene due to the relative motion of the target and sensor. The distributed (imaging) sensor processing part of the increasingly popular distributed surveillance systems also generate or could generate 3D information about the targets and their background scene. In this paper, the application of 3D moment invariants and perspective invariants for modeling automated target recognition is reviewed. Both two and three dimensional algebraic and moment invariants as well as perspective invariants have been derived and used for various type of recognition. Problems with numerical computation and partially occluded objects have been encountered and resolved. The theory has been demonstrated for synthetic target recognition and the method appears promising for implementation on real imagery.

Several array processing architectures have been devised for mitigation of multipath interference. Those algorithms have focused on rejection of terrain scattered interference, sidelobe and mainlobe clutter, and hot clutter. Two adaptive array architectures for rejection of broadband coherent interference are developed in this paper. Both techniques involve uniform subbanding using the Discrete Fourier Transform or non-uniform subbanding using the wavelet transform. In the first approach, adaptation and weight computation is performed independently in each subband, and in the second technique the Frost constrained LMS algorithm is applied to all subbands in the transform domain. Due to the correlated nature of the jamming signal, spatial averaging is utilized in both cases. The mitigation performance of both algorithms is compared for various scenarios of coherent broadband interference. This paper also focuses on evaluating and rejecting multipath interference due to propellers or rotor of the radar receiver aircraft or helicopter. This type of interference is characterized by an induced doppler spread proportional to the angular velocity of the propeller or the rotor. Furthermore the reflected signal represents the near-field component of the propeller scattering signature. Spatial averaging followed by a transform-domain adaptive beamformer based on the Frost algorithm is used to reject this type of propeller generated multipath interference.

This paper addresses the problem of target identification using features extracted from the time-frequency domain. This approach is an attractive alternative to other target recognition schemes because it makes us of the time-dependence of frequency domain signatures or the frequency dependence of time-domain scattering features. The disadvantage, however, is that time-frequency signatures are based on the magnitude squared of the received backscatter signal and are thus characterized by high noise variance. In this paper, we determine the feasibility of radar target identification in the time-frequency domain. The goal is to compare the performance of time-frequency domain target recognition with other schemes that rely on the raw backscatter data. To ensure fairness, we implement the same type of statistical pattern recognition in both domains. We also attempt three types of time-frequency backscatter representations: Wigner-Ville, Bjorn-Jordan, and spectrograms. We use the analytic form of real radar signatures recorded in a compact range environment for different target azimuth positions.

Many problems have a structure with an inherently two (or higher) dimensional nature. Unfortunately, the classical method of representing problems when using Genetic Algorithms (GAs) is of a linear nature. We develop a genome representation with a related crossover mechanism which preserves spatial relationships for 2D problems. We then explore how crossover disruption rates relate to the spatial structure of the problem space. After discussing why a more appropriate representation is needed and exploring the theoretical aspects of our method, we empirically test our method to verify that it will be effective. We develop an easily understood abstracted class of problems with a 2D structure. A Monte Carlo study comparing the GAs using the string and matrix methods on a number of members of this problem class is then conducted. Results are presented which clearly show that for this particular problem, a matrix oriented GA should be used. Given our success in applying the matrix representation to an abstracted problem, we apply our methods to a real world image processing problem. We develop a method for using a GA with a matrix representation to denoise a greyscale image, and we apply this method to a noisy image. Finally, we discuss further ways in which to extend this work. Possible future image processing applications include various problems such as filter design, segmentation and edge detection. Other applications include semi-parametric density estimation, nonlinear multiple regression and solutions of multi-parameter multi-equation systems. We also discuss how problems where higher dimensional structures might be employed to further generalize our work to cases other than 2D problems.