We study by way of simulations the creation and propagation of noise in a dispersive medium. The noise is generated as the sum of elementary signals and we study the noise like behavior in regards to the properties of the elementary signals and the dispersion relation.

A robust target tracking algorithm is proposed to overcome a number of challenges often associated with the FLIR imagery. Several disjoint intermediate background models are used to form an accurate and dynamic representation of the current background. The signatures of moving targets are captured and enhanced through a set of image filters, while the next movements of these targets are reasonably estimated using a set of kinematic predictors. By integrating the effective target detection method with the robust background modeling process, an excellent target tracking performance can be achieved.

Critical to a large portion of mission scenarios within the intelligence, surveillance, and reconnaissance (ISR) sensor community is the challenge to ensure designated targets of interest are reliably tracked in dynamic environments. Current generation trackers frequently loose track when targets become temporarily obscured, shadowed, or is in close proximity to other objects. In this paper we propose and demonstrate a generic confirmation of identity module that is based on the Distance Classifier Correlation Filter (DCCF) and is applicable to a variety of tracking technologies. The prevailing idea of this technique is that during a trackers valid track phase, learning exemplars are provided to a filter building process and templates of the tracked targets are created real-time online. Differences in orientation are handled through the creation of synthetic views using real target views and image warping techniques. After obscuration and/or during periods of track ambiguity, each new candidate track is matched against the prior valid track(s) using DCCF matching to resolve uncertainty.

This paper presents an automatic approach for camera/image based detection, recognition and tracking of flying objects (planes, missiles, etc.). The method detects appearing objects, and recognizes re-appearing targets. It uses a feature-based statistical modeling approach (e.g. HMM) for motion-based recognition, and an image feature (e.g. shape) based indexed database of pre-trained object classes, suitable for recognition on known and alerting on unknown objects. The method can be used for detection of flying objects, recognition of the same object category through multiple views/cameras and signal on unusual motions and shape appearances.

Learning is one of the most crucial components, which increases generality, flexibility, and robustness of computer vision systems. At present, image analysis algorithms adopt particular machine learning methods resulting in rather superficial learning. We present a new paradigm for constructing essentially learnable image analysis algorithms. Learning is interpreted as optimization of image representations. Notion of representation is formalized within information-theoretic framework. Optimization criterion is derived from well-known minimum description length (MDL) principle. Adaptation of the MDL principle in computer vision has been receiving increasing attention. However, this principle has been applied in heuristic way. We deduced representational MDL (RMDL) principle that fills the gap between theoretical MDL principle and its practical applications. The RMDL principle gives criteria both for optimal model selection of a single image within given representation, and for optimal representation selection for an image sample. Thus, it can be used for optimization of computer vision systems functioning within specific environment. Adequacy of the RMDL principle was validated on segmentation-based representations applied to different object domains. A method for learning local features as representation optimization was also developed. This method outperformed some popular methods with predefined representations such as SURF. Thus, the paradigm can be admitted as promising.

This paper presents a wideband jamming emitter localization method based on the fusion of multiple direction-of-arrivals (DOAs) and time-difference-of-arrival(s) (TDOA(s)) obtained from multiple unmanned-aerial-vehicles (UAVs). In this technique, we assume that multiple trajectory controllable UAVs are available and they are equipped with smart antennas to estimate the emitter's angle (DOA) with respect to themselves. In addition, UAVs communicate with each other to support the estimation of TDOA between pair(s) of UAVs and the emitter. The obtained DOA and TDOA information is fused at one UAV using an extended Kalman filter (EKF) to localize and track the mobile/static jamming emitter. In this method, we use DOA fusion to provide a good initialization of the emitter position to guarantee the EKF's convergence speed. The emitter localization accuracy is provided by the fusion of TDOAs if three or more UAVs are available. The simulation results show that the localizing-tracking error is less than 15% of the range difference estimation error standard deviation with three UAVs that is accurate enough for jamming emitter destruction.

This paper discusses a novel image noise reduction strategy based on the use of adaptive image filter kernels. Three adaptive filtering techniques are discussed and a case study based on a novel Adaptive Gaussian Filter is presented. The proposed filter allows the noise content of the imagery to be reduced whilst preserving edge definition around important salient image features. Conventional adaptive filtering approaches are typically based on the adaptation of one or two basic filter kernel properties and use a single image content measure. In contrast, the technique presented in this paper is able to adapt multiple aspects of the kernel size and shape automatically according to multiple local image content measures which identify pertinent features across the scene. Example results which demonstrate the potential of the technique for improving image quality are presented. It is demonstrated that the proposed approach provides superior noise reduction capabilities over conventional filtering approaches on a local and global scale according to performance measures such as Root Mean Square Error, Mutual Information and Structural Similarity. The proposed technique has also been implemented on a Commercial Off-the-Shelf Graphical Processing Unit platform and demonstrates excellent performance in terms of image quality and speed, with real-time frame rates exceeding 100Hz. A novel method which is employed to help leverage the gains of the processing architecture without compromising performance is discussed.

In this paper, a new approach of multi-class target recognition is proposed for remote sensing image analysis. A multiclass feature model is built, which is based on sharing features among classes. In order to make the recognition process efficient, we adopted the idea of adaptive feature selection. In each layer of the integrated feature model, the most salient and stable feature are selected first, and then the less ones. Experiments demonstrated the approach proposed is efficient in computation and is adaptive to scene variation.

The reliability of the data analysis from sensors is the main factor for making the right decisions for target recognition. Obviously, the reliability depends on the quality of the sensors and processing electronics. However, the cases where the target is clearly determined are not numerous. More often, we receive partial and distorted images from the sensors employed. Thus, we have a task of determining the correct initial image that was distorted by various factors before and after it was detected by the sensors. The proposed approach is an adaptive intelligent system that uses algorithms, renewable data base as well as the possibility of changing the detection system's parameters and modes of operation depending on the signal received from the identified objects.

Correlation filters (CFs) can detect multiple targets in one scene making them well-suited for automatic target recognition (ATR) applications. Quadratic CFs (QCFs) can improve performance over linear CFs. QCFs are able to detect one class of targets and reject clutter. We present a method to increase the QCF capabilities to detect two classes of targets and reject clutter. We integrate the ATR tasks of detection, recognition, and tracking algorithms using the Multi-Frame Correlation Filter (MFCF) framework. Our simulation results demonstrate the algorithm's ability to detect multiple targets from two classes while rejecting clutter.

A research area based on the application of information theory to machine learning has attracted considerable interest in the last few years. This research area has been coined information-theoretic learning within the community. In this paper we apply elements of information-theoretic learning to the problem of automatic target recognition (ATR). A number of researchers have previously shown the benefits of designing classifiers based on maximizing the mutual information between the class data and the class labels. Following prior research in information-theoretic learning, in the current results we show that quadratic mutual information, derived using a special case of the more general Renyi's entropy, can be used for classifier design. In this implementation, a simple subspace projection classifier is formulated to find the optimal projection weights such that the quadratic mutual information between the class data and the class labels is maximized. This subspace projection accomplishes a dimensionality reduction of the raw data set wherein information about the class membership is retained while irrelevant information is discarded. A subspace projection based on this criterion preserves as much class discriminability as possible within the subspace. For this paper, laser radar images are used to demonstrate the results. Classification performance against this data set is compared for a gradient descent MLP classifier and a quadratic mutual information MLP classifier.

There are many applications for which it is important to resolve the location and motion of a target position. For the static situation in which a target transmitter and several receivers are not in motion, the target may be completely resolved by triangulation using relative time delays estimated by several receivers at known locations. These delays are normally estimated from the location of peaks in the magnitude of the cross-correlation function. For active radars, a transmitted signal is reflected by the target, and range and radial velocity are estimated from the delay and Doppler effects on the received signal. In this process, Doppler effects are conventionally modeled as a shift in frequency, and delay and Doppler are estimated from a cross-ambiguity function (CAF) in which delay and Doppler frequency shift are assumed to be independent and approximately constant. Delay and Doppler are jointly estimated as the location of the peak magnitude of the CAF plane. We present methods for accurately estimating delay for the static case and delay and the time-varying Doppler effects for non-static models, such as the radar model.

Some marine mammals as well as bats are known to emit sophisticated waveforms while searching for objects or hunting prey. Some dolphins have been observed to change their sonar pulse depending on the environment. Incorporating these strategies into sonar waveform and receiver design has become an active area of research. In this paper, we explore the application of an optimal waveform design scheme recently given by Kay, to the detection of elastic objects. We examine the benefits of optimal waveform design versus transmitting a linear FM waveform, as well as performance loss suffered by assuming a point target. The optimization approach designs the magnitude spectrum of the transmit waveform and, accordingly, there is an unlimited number of "optimal" transmit waveforms with the same magnitude spectrum. We propose a time domain optimization criterion to obtain the transmit waveform with the optimal magnitude spectrum and the smallest possible duration, as well as the waveform with the optimal magnitude spectrum and the longest possible duration. The former waveform allows for higher ping rates, but necessarily has higher time domain peak power, while the latter waveform has lower time domain peak power and lower ping rates. A method to obtain waveforms that are a blend of these two extremes is also presented, allowing a smooth trade-off between ping rate and peak power.

For some time, applying the theory of pattern recognition and classification to radar signal processing has been a topic of interest in the field of remote sensing. Efficient operation and target indication is often hindered by the signal background, which can have similar properties with the interesting signal. Because noise and clutter may constitute most part of the response of surveillance radar, aircraft and other interesting targets can be seen as anomalies in the data. We propose an algorithm for detecting these anomalies on a heterogeneous clutter background in each range-Doppler cell, the basic unit in the radar data defined by the resolution in range, angle and Doppler. The analysis is based on the time history of the response in a cell and its correlation to the spatial surroundings. If the newest time window of response in a resolution cell differs statistically from the time history of the cell, the cell is determined anomalous. Normal cells are classified as noise or different type of clutter based on their strength on each Doppler band. Anomalous cells are analyzed using a longer time window, which emulates a longer coherent illumination. Based on the decorrelation behavior of the response in the long time window, the anomalous cells are classified as clutter, an airplane or a helicopter. The algorithm is tested with both experimental and simulated radar data. The experimental radar data has been recorded in a forested landscape.

Previously we have given explicit expressions for the moments of a pules propagating in a dispersive medium. Liu and Yeh have given the moments for a pulse in a random medium with no dispersion. In this paper we derive the time moments of a pulse propagating in a random medium with dispersion.

Modeling of the electromagnetic (EM) scattering mechanisms from two-dimensional (2-D) time-evolving sea surfaces is a particularly complicated problem. The intricate structure of surface waves and the scattering models noticeably influence the simulated radar signatures. Scattering calculation and Doppler spectra from the sea surfaces have been intensively studied, experimentally as well as theoretically in the past decades. However, to author's knowledge, very few results can be found in literatures for Doppler spectra from two-dimensional time-evolving nonlinear sea surfaces. In this work we focus on the Doppler spectral characteristics from 2-D time-evolving nonlinear sea surfaces. Based on Creamer's sea surface model, the first-order small slope approximation (SSA) method is applied to solve the 3-D scattering problem. The Doppler spectra of the backscattered signals from 2-D time-evolving sea surfaces are studied for different incident angles (from normal to grazing) as well as wind directions (from upwind to crosswind). The impacts of the nonlinearity on the Doppler shifts and spectral widths of backscattered signals are analyzed.

Future target acquisition missions of military aircraft require a robust classification and tracking system of ground targets. In combination with onboard ATD systems, a fast "Find, Fix, Track" cycle for airborne platforms using EO, IR and SAR imaging sensors should be achieved. For EO/IR image sequences, a 3D matching and pose estimation method was developed by EADS internal research. The approach determines the resemblance between rendered 3D CAD models and sensor images to identify the best-matching object pose by optimizing different similarity measures. In order to assess the suitability of this method for real-world military aircraft missions, the present paper introduces a number of robustness requirements w.r.t. sensors, scenarios, object classes and environmental conditions and systematically evaluates the proposed method on a set of image sequences ranging from purely synthetic over laboratory conditions to real-world recordings in a rapid prototyping environment using graphics cards acceleration techniques. The outlook shows possible extensions of the system e.g. tracking and hypothesis management modules as well as the necessary steps to implement and integrate the selected method into a real-time embedded onboard mission system.

3D imagery has a well-known potential for improving situational awareness and battlespace visualization by providing enhanced knowledge of uncooperative targets. This potential arises from the numerous advantages that 3D imagery has to offer over traditional 2D imagery, thereby increasing the accuracy of automatic target detection (ATD) and recognition (ATR). Despite advancements in both 3D sensing and 3D data exploitation, 3D imagery has yet to demonstrate a true operational gain, partly due to the processing burden of the massive dataloads generated by modern sensors. In this context, this paper describes the current status of a workbench designed for the study of 3D ATD/ATR. Among the project goals is the comparative assessment of algorithms and 3D sensing technologies given various scenarios. The workbench is comprised of three components: a database, a toolbox, and a simulation environment. The database stores, manages, and edits input data of various types such as point clouds, video, still imagery frames, CAD models and metadata. The toolbox features data processing modules, including range data manipulation, surface mesh generation, texture mapping, and a shape-from-motion module to extract a 3D target representation from video frames or from a sequence of still imagery. The simulation environment includes synthetic point cloud generation, 3D ATD/ATR algorithm prototyping environment and performance metrics for comparative assessment. In this paper, the workbench components are described and preliminary results are presented. Ladar, video and still imagery datasets collected during airborne trials are also detailed.

Laser-based 3D sensors measure range with high accuracy and allow for detection of objects behind various type of occlusion, e.g., tree canopies. Range information is valuable for detection of small objects that are typically represented by 5-10 pixels in the data set. Range information is also valuable in tracking problems when the tracked object is occluded under parts of its movement and when there are several objects in the scene. In this paper, on-going work on detection and tracking are presented. Detection of partly occluded vehicles is discussed. To detect partly occluded objects we take advantage of the range information for removing foreground clutter. The target detection approach is based on geometric features, for example local surface detection, shadow analysis and height-based detection. Initial results on tracking of humans are also presented. The benefits with range information are discussed. Results are illustrated using outdoor measurements with a 3D FLASH LADAR sensor and a 3D scanning LADAR.

Automatic target recognition (ATR) based on the emerging technology of Compressed Sensing (CS) can considerably improve accuracy, speed and cost associated with these types of systems. An image based ATR algorithm has been built upon this new theory, which can perform target detection and recognition in a low dimensional space. Compressed dictionaries (A) are formed to include rotational information for a scale of interest. The algorithm seeks to identify y(test sample) as a linear combination of the dictionary elements : y=Ax, where A ∈ R^nxm(n<<m) and x is a sparse vector whose non-zero entries identify the input y. The signal x will be sparse with respect to the dictionary A as long as y is a valid target. The algorithm can reject clutter and background, which are part of the input image. The detection and recognition problems are solved by finding the sparse-solution to the undetermined system y=Ax via Orthogonal Matching Pursuit (OMP) and l₁ minimization techniques. Visible and MWIR imagery collected by the Army Night Vision and Electronic Sensors Directorate (NVESD) was utilized to test the algorithm. Results show an average detection and recognition rates above 95% for targets at ranges up to 3Km for both image modalities.

Subspace projection is an effective and established way to form classes in the Automatic Target Acquisition (ATA) problem. Class subspace formation is viewed in this paper as an over specified F h = u problem. Recent advances in compressive imaging show that this problem can be solved for sparse matrices via iterative techniques. Convergence of these techniques is aided by a metric induced by an appropriately selected norm. In this paper we will use infrared data to show this rapid class formation and to compare convergence for two norms. Based on this class formulation a new method for ATA solution will also be demonstrated.

In this paper, we develop a framework for using only the needed data for automatic target recognition (ATR) algorithms using the recently developed theory of sparse representations and compressive sensing (CS). We show how sparsity can be helpful for efficient utilization of data, with the possibility of developing real-time, robust target classification. We verify the efficacy of the proposed algorithm in terms of the recognition rate on the well known Comanche forward-looking infrared (FLIR) data set consisting of ten different military targets at different orientations.

The development of a more unified theory of automatic target recognition (ATR) has received considerable attention over the last several years from individual researchers, working groups, and workshops. One of the major benefits expected to accrue from such a theory is an ability to analytically derive performance metrics that accurately predict real-world behavior. Numerous sources of uncertainty affect the actual performance of an ATR system, so direct calculation has been limited in practice to a few special cases because of the practical difficulties of manipulating arbitrary probability distributions over high dimensional spaces. This paper introduces an alternative approach for evaluating ATR performance based on a generalization of NorbertWiener's polynomial chaos theory. Through this theory, random quantities are expressed not in terms of joint distribution functions but as convergent orthogonal series over a shared random basis. This form can be used to represent any finite-variance distribution and can greatly simplify the propagation of uncertainties through complex systems and algorithms. The paper presents an overview of the relevant theory and, as an example application, a discussion of how it can be applied to model the distribution of position errors from target tracking algorithms.

This paper describes a new wavelet-based anomaly detection technique for Forward Looking Infrared (FLIR) sensor consisting a Long-wave (LW) and a Mid-wave (MW) sensor. The proposed approach called wavelet-RX algorithm consists of a combination of a two-dimensional (2-D) wavelet transform and the well-known multivariate anomaly detector called the RX algorithm. In our wavelet-RX algorithm, a 2-D wavelet transform is first applied to decompose the input image into uniform subbands. A number of significant subbands (high energy subbands) are concatenated together to form a subband-image cube. The RX algorithm is then applied to each subbandimage cube obtained from wavelet decomposition of LW and MW sensor data separately. Experimental results are presented for the proposed wavelet-RX and the classical CFAR algorithm for detecting anomalies (targets) in a single broadband FLIR (LW or MW) sensors. The results show that the proposed wavelet-RX algorithm outperforms the classical CFAR detector for both LW and for MW FLIR sensors data.

This research paper investigates the use of the wavelet transform to extract spatially-invariant wavelet-based shape signatures for automatic target recognition (ATR). Target signatures based on shape information can be generally categorized as either contour-based or region-based. The wavelet-based shape signatures facilitate detection and localization of important edge and texture information aiding in discrimination between targets. To demonstrate the advantages of both edge and region information, we present an approach that combines region-based shape methods and the wavelet transform for generating target signatures. Our approach generates a rotationally invariant class of wavelet signatures based on the spatial ground pixel coverage of the target is determined from the region-of-interest (ROI) in the wavelet domain. This process results in a multiresolution representation of the target, and provides a hierarchical approach to target signature-matching. We demonstrate this methodology using signatures from aircraft targets utilizing the Angular Radial Transform (ART) as the region-based shape signature. Region-based signatures are shown more robust than contourbased signatures in the presence of noise and disconnected target regions providing greater confidence in target identification. Our research results show the value of combining the rotational invariance of the ART signatures with the localization and edge discrimination properties of the wavelet transform.

Automated target cueing (ATC) can assist analysts with searching large volumes of imagery. Performance of most automated systems is less than perfect, requiring an analyst to review the results to dismiss false alarms or confirm correct detections. This paper explores methods for improving the presentation and visualization of the ATC output, enabling more efficient and effective review of the detections flagged by the ATC. The techniques presented in this paper are applicable to a wide range of search problems using data from different sensors modalities. The information available to the computer increases as ATC detections are either accepted or rejected by the analyst. It is often easy to confirm obviously correct detections and dismiss obvious false alarms, which provides the starting point for the automated updating of the visualization. In machine learning algorithms, this information can be used to retrain or refine the classifier. However, this retraining process is appropriate only when future sensor data is expected to closely resemble the current set. For many applications, the sensor data characteristics (viewing geometry, resolution, clutter complexity, prevalence and types of confusers) are likely to change from one data collection to the next. For this reason, updating the visualization for the current data set, rather than updating the classifier for future processing, may prove more effective. This paper presents an adaptive visualization technique and illustrates the technique with applications.

The global war on terror has plunged US and coalition forces into a battle space requiring the continuous adaptation of tactics and technologies to cope with an elusive enemy. As a result, technologies that enhance the intelligence, surveillance, and reconnaissance (ISR) mission making the warfighter more effective are experiencing increased interest. In this paper we show how a new generation of smart cameras built around foveated sensing makes possible a powerful ISR technique termed Cascaded ATR. Foveated sensing is an innovative optical concept in which a single aperture captures two distinct fields of view. In Cascaded ATR, foveated sensing is used to provide a coarse resolution, persistent surveillance, wide field of view (WFOV) detector to accomplish detection level perception. At the same time, within the foveated sensor, these detection locations are passed as a cue to a steerable, high fidelity, narrow field of view (NFOV) detector to perform recognition level perception. Two new ISR mission scenarios, utilizing Cascaded ATR, are proposed.

Spectral variability remains a challenging problem for target detection in hyperspectral (HS) imagery. In this paper, we have applied the kernel-based support vector data description (SVDD) to perform full-pixel target detection. In target detection scenarios, we do not have a collection of samples characterizing the target class; we are typically given a pure target signature that is obtained from a spectral library. In our work, we use the pure target signature and first-order Markov theory to generate N samples to model the spectral variability of the target class. We vary the value of N and observe its effect to determine a value of N that provides acceptable detection performance. We have inserted target signatures into an urban HS scene with varying levels of spectral variability to explore the performance of the proposed SVDD target detection scheme in these scenarios. The proposed approach makes no assumptions regarding the underlying distribution of the scene data as do traditional stochastic detectors such as the adaptive matched filter (AMF). Detection results in the form of confusion matrices and receiver-operating-characteristic (ROC) curves demonstrate that the proposed SVDD-based scheme is highly accurate and yields higher true positive rates (TPR) and lower false positive rates (FPR) than the AMF.

Spectral variability remains a challenging problem for target detection in hyperspectral (HS) imagery. In previous work, we developed a target detection scheme using the kernel-based support vector data description (SVDD). We constructed a first-order Markov-based Gaussian model to generate samples to describe the spectral variability of the target class. However, the Gaussian-generated samples also require selection of the variance parameter σ 2 that dictates the level of variability in the generated target class signatures. In this work, we have investigated the use of decision-level fusion techniques for alleviating the problem of choosing a proper value of σ 2 . We have trained a collection of SVDDs with unique variance parameters σ 2 for each of the target training sets and have investigated their combination using the traditional AND, OR, and majority vote (MV) decision-level rules. We have inserted target signatures into an urban HS scene with differing levels of spectral variability to explore the performance of the proposed scheme in these scenarios. Experiments show that the MV fusion rule is the best choice, providing relatively low false positive rates (FPR) while yielding high true positive rates (TPR). Detection results show that the proposed SVDD-based decision-level scheme using the MV fusion rule is highly accurate and yields higher true positive rates (TPR) and lower false positive rates (FPR) than the adaptive matched filter (AMF).

An essential component within most approaches used to evaluate ATR algorithm performance is an image database from which are chosen a training set of images. Several fundamental questions arise regarding the adequacy of the database to represent the desired domain of effectiveness, the sufficiency of the training set, potentiality of enhancing the constituents of the training set, suitability to determine signal-to-clutter performance, and realism of fairly comparing performance of ATR algorithms to one another. These questions have been addressed through an investigation into a unified approach for database analysis and how it can be applied to evaluating ATR performance metrics.

Correlation filters (CF) have been widely used for detecting and recognizing patterns in 2-D images. These filters are designed to yield sharp correlation peaks for desired objects while exhibiting low response to clutter and background. CFs are designed using training images that resemble the object of interest. However it is not clear what should be the background of these training images. Some methods use a white background while others use the mean value of the target region. It is important to determine an appropriate background since a mismatched background may cause the filter to discriminate based on the background rather than the target pattern. In this paper we discuss a method to choose training images, and we compare the effects of different backgrounds on the filter performance in different scenarios using both synthetic (pixels in the background chosen from a Gaussian distribution) and real backgrounds (photographs of different sceneries) for testing. In our comparisons we do not restrict ourselves to using a background with constant pixel intensity for training but also include in the training images backgrounds with varying pixel intensity with mean and standard deviation equal to the mean and standard deviation of the target region. Experiments show that without a prior knowledge of the background in the testing images, training the filters using a background with the mean and variance of all the desired objects tends to give better results.

The Spectral and Polarimetric Imagery Collection Experiment (SPICE) is a collaborative effort between the US Army ARDEC and ARL that is focused on the collection of mid-wave and long-wave infrared imagery using hyperspectral, polarimetric, and broadband sensors. The objective of the program is to collect a comprehensive database of the different modalities over the course of 1 to 2 years to capture sensor performance over a wide variety of weather conditions, diurnal, and seasonal changes inherent to Picatinny's northern New Jersey location. Using the Precision Armament Laboratory (PAL) tower at Picatinny Arsenal, the sensors will autonomously collect the desired data around the clock at different ranges where surrogate 2S3 Self-Propelled Howitzer targets are positioned at different viewing perspectives in an open field. The database will allow for: 1) Understanding of signature variability under adverse weather conditions; 2) Development of robust algorithms; 3) Development of new sensors; 4) Evaluation of polarimetric technology; and 5) Evaluation of fusing the different sensor modalities. In this paper, we will present the SPICE data collection objectives, the ongoing effort, the sensors that are currently deployed, and how this work will assist researches on the development and evaluation of sensors, algorithms, and fusion applications.

Stochastic resonance has received significant attention recently in the signal processing community with emphasis on signal detection. The basic notion is that the performance of some suboptimal detectors can be improved by adding independent noise to the measured (and already noise contaminated) observation. The notion of adding noise makes sense if the observation is the result of nonlinear processing, and there exist proven scenarios where the signal-to-noise ratio improves by adding independent noise. This paper reviews a set of parametric and nonparametric sub-optimal radar target classification systems and explores (via computer simulation) the impact of adding independent noise to the observation on the performance of such sub-optimal systems. Although noise is not added in an optimal fashion, it does have an impact on the probability of classification error. Real radar scattering data of commercial aircraft models is used in this study. The focus is on exploring scenarios where added noise may improve radar target classification.

This paper describes a method and system of human-like attention and object segmentation in visual scenes that (1) attends to regions in a scene in their rank of saliency in the image, (2) extracts the boundary of an attended proto-object based on feature contours, and (3) can be biased to boost the attention paid to specific features in a scene, such as those of a desired target object in static and video imagery. The purpose of the system is to identify regions of a scene of potential importance and extract the region data for processing by an object recognition and classification algorithm. The attention process can be performed in a default, bottom-up manner or a directed, top-down manner which will assign a preference to certain features over others. One can apply this system to any static scene, whether that is a still photograph or imagery captured from video. We employ algorithms that are motivated by findings in neuroscience, psychology, and cognitive science to construct a system that is novel in its modular and stepwise approach to the problems of attention and region extraction, its application of a flooding algorithm to break apart an image into smaller proto-objects based on feature density, and its ability to join smaller regions of similar features into larger proto-objects. This approach allows many complicated operations to be carried out by the system in a very short time, approaching real-time. A researcher can use this system as a robust front-end to a larger system that includes object recognition and scene understanding modules; it is engineered to function over a broad range of situations and can be applied to any scene with minimal tuning from the user.

This paper describes an algorithm and system for rapidly generating a saliency map and finding interesting regions and in large-sized (i.e., extremely high-resolution) imagery and video. Previous methods of finding salient or interesting regions have a fundamental shortcoming: they need to process the entire image before the saliency map can be outputted and are therefore very slow for large images. Any prior attempts at parallelizing this operation involve computing feature maps on separate processors, but these methods cannot provide a result until the entire image has been processed. Rather than employing a single-step process, our system uses a recursive approach to estimate the saliency, processing parts of the image in sequence and providing an approximate saliency map for these regions immediately. With each new part of the image, a series of normalization factors is updated that connects all image parts analyzed so far. As more of the image parts are analyzed, the saliency map of the previously analyzed parts as well as newly analyzed parts becomes more exact. In the end, an exact global saliency map of the entire image is available. This algorithm can be viewed as (1) a fast, parallelizable version of prior art, and/or (2) a new paradigm for computing saliency in large imagery/video. This is critical, as the analysis of large, high-resolution imagery becomes more commonplace. This system can be employed in a default, bottom-up manner or a directed, top-down manner which will assign a preference to certain features over others. One can apply this system to any static scene, whether that is a still photograph or an image captured from video.

This paper illustrates an approach to sequential hypothesis testing designed not to minimize the amount of data collected but to reduce the overall amount of processing required, while still guaranteeing pre-specified conditional probabilities of error. The approach is potentially useful when sensor data are plentiful but time and processing capability are constrained. The approach gradually reduces the number of target hypotheses under consideration as more sensor data are processed, proportionally allocating time and processing resources to the most likely target classes. The approach is demonstrated on a multi-class ladar-based target recognition problem and compared with uniform-computation tests.

The Air Force Institute of Technology's Center for Directed Energy's (AFIT/CDE), under sponsorship of the HEL Joint Technology Office, and as part of a multidisciplinary research initiative on aero optics effects, has designed and fabricated a laser pointing/tracking system. This system will serve as the laser source for a series of in-flight data collection campaigns involving two aircraft. Real-Time tracking systems have a distinct difference from automatic image analysis. Both activities often involve the segmentation of an image and the automatic location of an item of interest. A number of advanced tracking algorithms have been developed for applications involving processing previously captured data. Medical imaging applications frequently use post processing algorithms to segment anomalies in medical imaging. In this paper we discuss an airborne laser pointing and tracking system and its requirements, designed and implemented at AFIT. This application is different because the image processing must be completed during the inter-frame period. AFIT analyzed available tracking algorithms including: centroid tracking, Fitts correlator, Posterior Track, and Active Contour. These algorithms were evaluated on their ability to both accurately track and to be computed in real time using existing hardware. The analysis shows that some of the more accurate tracking algorithms are not easily implementable in real time. Often there are large numbers of correlations that must be computed for each frame. Higher resolution images quickly escalate this problem. Algorithm selection for tracking applications must balance the need for accuracy and computational simplicity. Real time tracking algorithms are limited by the amount of time between frames with which to processes the data. Specialized hardware can improve this situation. We selected centroid tracking for the airborne application and evaluate its performance to show that it meets design requirements.

For long-range imaging or low signal-to-noise ratio environments sightline jitter is a primary source of image degradation. The conventional pointing & stabilization figure of merit is therefore the jitter RMS, with bearing friction often the largest contributor overall. Recent work has shown that pixel smear during camera integration can be reduced if adaptive friction compensation 'shapes' the jitter frequency content in addition to reducing the RMS value. This paper extends this work by automating the tuning process for the sightline control parameters by using a genetic algorithm. The GA fitness metric is the integral of the modulation transfer function due to any residual sightline jitter. It is shown that this fitness function is significantly better than the current root-mean-square figure of merit typically employed in stabilization loop design.

Recently, the author solved a general problem of fusing asynchronous tracks while taking into account communicatrion delays, data latency, and out of sequence tracks. The objective of this paper is to perform preliminary performance analysis of the asynchronous track fusion algorithm. In this study, two snesors providing asynchronous measurements are considered, where the update track fusion rate is fixed. Communication delay between at least of the senors' platform and the fusion center may exit. Monte Carlo simulations are performed using simulated traget tracks. The performance of the individual sensors as well as that of the fused track is provided. The preliminary results show the benefit of track fusion under more realistic assumptions than what is currently the practice.

An adaptive image pre-processor has been developed for a next-generation video tracking system. In a previous paper we presented a wavelet-based enhancement pre-processor (AWEP) which showed good segmentation capability. Here we discuss the impact of structural and implementation constraints placed on the algorithm during targeting on a low power FPGA device. We discuss the underlying issues and outline our approach to compensating the effect and regaining stability. Output results are given illustrating the segmentation performance after applied optimization of the decomposition filter kernels. A set of results from the tracking system are presented to demonstrate the effectiveness of the AWEP implementation on the tracking performance applied to real video.

Structural dynamics is one of the most important elements of a precision gimbal design which often dominates the system line-ofsight stabilization and pointing performance. Structural effects are manifested in these systems in several unrelated ways that the systems engineer, and other members of the design team, must understand in order to insure a successful design. Once the effects are clearly understood, analysis techniques, such as finite elements, can be applied to provide models to accurately predict the various interactions and evaluate potential designs. Measurement techniques such as modal analysis can also be used to obtain models of existing hardware and to verify the design. However, the successful and efficient application of the above process requires that the underlying principles and effects are well understood by all the members of the engineering design team. This usually includes, as a minimum, the control systems engineer, the structural analyst and the mechanical engineer but may involve other members of the design team as well. Appropriate transfer functions for the various interactions, for example, can be defined and provided by the structural analyst to the control system engineer to evaluate and performance predictions can be iterated as necessary until the entire system meets the required performance in the intended dynamic environment. Often, however, one or more members of the team do not have an appreciation for the effects or design process required and the result is a frustrated design effort and lower system performance that might have otherwise been easily achieved. While different systems can have vastly different requirements and configurations, the above effects and techniques are common to most and this paper is an attempt to provide a straightforward outline of the more common of these in order to improve communication among design team members so that they can all contribute at their maximum potential.

The paper describes some design and implementation aspects of a low-cost inertial estimation unit based on comercially available inertial sensors. The primary task for the unit described in this paper is to estimate the attitude (orientation, pose), but the extension to estimating the position and height is planned. The size of the unit is about the size of a handheld device. It includes a commercially available 3-axis rate gyro combined in a single package with a 3-axis accelerometer and a 3-axis magnetometer. In order to include position estimation capabilities a GPS receiver is attached and a barometric pressure sensor can be added. The primary limitation of the implementation described in this paper is that it assumes no long term acceleration of the carrier (neither along a linear nor along a curved path), which makes the result of less value in aerospace industry but may have some appeal to researcher engineers in other fields. The data measured by the three sensors are fused using the Extended Kalman filtering paradigm. No model of the dynamics of the carrier (aircraft, mobile robot or a patient) is relied upon, the only modeled dynamics is that of sensors, such as the bias and noise. The choice of extended Kalman filtering methodology was dictated by strong requirements on computational simplicity. Some experience with implementation of the proposed scheme on a digital hardware (ARM7 based microcontroller) is shared in the paper. Finally, functionality of the presented device is demonstrated in experiments. Besides simple indoor tests, fly experiments were conducted using a small UAV helicopter.

The paper reports on a few control engineering issues related to design and implementation of an image-based pointing and tracking system for an inertially stabilized airborne camera platform. A medium-sized platform has been developed by the authors and a few more team members within a joint governmental project coordinated by Czech Air Force Research Institute. The resulting experimental platform is based on a common double gimbal configuration with two direct drive motors and off-the-shelf MEMS gyros. Automatic vision-based tracking system is built on top of the inertial stabilization. Choice of a suitable control configuration is discussed first, because the decoupled structure for the inner inertial rate controllers does not extend easily to the outer imagebased pointing and tracking loop. It appears that the pointing and tracking controller can benefit much from availability of measurements of an inertial rate of the camera around its optical axis. The proposed pointing and tracking controller relies on feedback linearization well known in image-based visual servoing. Simple compensation of a one sample delay introduced into the (slow) visual pointing and tracking loop by the computer vision system is proposed. It relies on a simple modification of the well-known Smith predictor scheme where the prediction takes advantage of availability of the (fast and undelayed) inertial rate measurements.

Precision tracking applications using two-axis gimbal or antenna actuation systems suffer from a singularity when the inner axis reaches +-90 degrees. This is known by various terms - the keyhole singularity, gimbal lock or the nadir problem. Practically, sightline control is degraded and often lost in a neighborhood of this singularity. In this paper, two nonlinear control algorithms are applied to sightline pointing and stabilization control in the neighborhood of the nadir; the traditional cosecant correction and the nonlinear generalized minimum variance technique. Both controllers were tested against a validated model of an Aeromech TigerEye turret.

A simple analytical model for the signal acquisition range for a laser guided mortar is presented. The signal consists of a repetitively pulsed laser of fixed pulse duration and fixed pulse repetition frequency. The pulses are detected by a seeker consisting of a quadrant photodiode and a trans-impedance amplifier. Noise is introduced from solar irradiance and from the detector/amplifier electronics. The model maximizes the acquisition range by optimizing trans-impedance amplifier circuit components. A comparison of integrating multiple low energy pulses (MPLD) versus detecting each pulse individually (conventional) is made.

Multi-beam technology is one of the key technologies in optical phased array systems for multi-object treatment and multi-task operation. A multi-beam forming and steering method was proposed. This method uses isosceles triangle multilevel phase grating (ITMPG) to form multiple beams simultaneously. Phase profile of the grating is a quantized isosceles triangle with stairs. By changing the phase difference corresponding to the triangle height, multiple beams can be steered symmetrically. It took 34 ms to calculate a set of parameters for one ITMPG, namely one steering. A liquid crystal spatial light modulator was used for the experiment, which formed 6 gratings. The distortion of which had been compensated with the accuracy of 0.0408 λ. Each grating included 16 phase elements with the same period. Steering angle corresponded to the triangle height, which is the phase difference. Relative diffraction efficiency for multiple beams was greater than 81%, intensity nonuniformity was less than 0.134, and the deflection resolution was 2.263 mrad. Experimental results demonstrate that the proposed method can be used to form and steer symmetrical multiple beams simultaneously with the same intensity and high diffraction efficiency in the far field, the deflection resolution is related to the reciprocal of grating period.

Volume holographic correlators offer the ability to encode and compare thousands of templates in one operation. Angle multiplexing of each individual template means the position of the correlation spot in the output plane corresponds to the matching template. To be useful as a correlator the shift invariance must be restored by scanning the input image. This can be achieved by implementing the input signal modulation on a high speed SLM such as a MQW or DLP that is capable speeds in excess of 30kHz. The output correlation peak is read out using a high-speed linear CCD camera. The Bragg angle affects the number of templates that can be held on the hologram. However, this is not the same in both directions and this changes the correlator's shift invariance ability in different scan directions. In this paper we investigate this and how it affects the correlator's performance. This arrangement allows thousands of templates to be searched at video rate. The scanning nature allows space domain correlation to be implemented. The system we describe offers the ability to pre-filter the signal. We report on the results of a MACH filter implemented in a volume holographic correlator. The scanning window allows some interesting pre-filtering to be performed, such normalisation and non-linear optimisation.

Pattern recognition deals with the detection and identification of a specific target in an unknown input scene. Target features such as shape, color, surface dynamics, and material characteristics are common target attributes used for identification and detection purposes. Pattern recognition using multispectral (MS), hyperspectral (HS), and polarization-based spectral (PS) imaging can be effectively exploited to highlight one or more of these attributes for more efficient target identification and detection. In general, pattern recognition involves two steps: gathering target information from sensor data and identifying and detecting the desired target from sensor data in the presence of noise, clutter, and other artifacts. Multispectral and hyperspectral imaging (MSI/HSI) provide both spectral and spatial information about the target. As the reflection or emission spectral signatures depend on the elemental composition of objects residing within the scene, the polarization state of radiation is sensitive to the surface features such as relative smoothness or roughness, surface material, shapes and edges, etc. Therefore, polarization information imparted by surface reflections of the target yields unique and discriminatory signatures which could be used to augment spectral target detection techniques, through the fusion of sensor data. Sensor data fusion is currently being used to effectively recognize and detect one or more of the target attributes. However, variations between sensors and temporal changes within sensors can introduce noise in the measurements, contributing to additional target variability that hinders the detection process. This paper provides a quick overview of target identification and detection using MSI/HSI, highlighting the advantages and disadvantages of each. It then discusses the effectiveness of using polarization-based imaging in highlighting some of the target attributes at single and multiple spectral bands using polarization spectral imaging (PSI), known as spectropolarimetry imaging.

A new security system is proposed using optical joint transform correlation technique which employs multiple phase-shifted reference images. In the proposed technique, the address code is used as the reference image and fed into four channels after performing phase shifting on them by different amount. The output signals from each channel are added with the input image to be encrypted for security purpose. Joint power spectra (JPS) signals can then be derived by applying Fourier transformation, and the resultant signals are phase-shifted and combined to form a modified JPS signal. Inverse Fourier transformation of the modified JPS signal yields the encrypted image which is now secure from any unauthorized access and/or loss of information. For decryption purpose, the received encrypted signal is first Fourier transformed and multiplied by the address code used in encryption, which is then inverse Fourier transformed to generate the output signal. The proposed technique does not involve any complex mathematical operation on the address code otherwise required in other security techniques. The proposed technique requires a simple architecture and operates fast, automatic and is invariant to noise and distortions. Performance of the proposed scheme is investigated using computer simulation using binary as well as gray images in both noise-free and noisy conditions.

In this paper, we exploit the nonlinearity inherent in four-wave mixing in organic photorefractive materials and demonstrate edge enhancement, contrast conversion, and defect enhancement in a periodic structure. With the availability of these materials, which have large space-bandwidth products, edge enhancement, contrast conversion and defect enhancement are possible.

In this paper we demonstrate image restoration via photorefractive two-beam coupling. Our restoration is based on coupling between the joint spectra of the distortion impulse response and the distorted image, and the clean reference beam. The image restoration is used to demonstrate one-way image transmission in an aberrating medium. Our experimental demonstration is supported by theoretical modeling of the restoration process and by computer modeling.

Bidimensional empirical mode decomposition (BEMD) decomposes an image into several bidimensional intrinsic mode components, which is useful for various image enhancement and/or feature extraction applications. However, because of the requirement of scattered data interpolation and associated difficulties, the classical BEMD methods appear unsuitable for many applications. Recently, a fast and adaptive BEMD (FABEMD) method is proposed, which alleviates some of the difficulties, otherwise encountered in classical BEMD approaches. On the other hand, existing BEMD methods are proposed for gray scale images only. This paper first presents a novel BEMD approach for color images known as color BEMD (CBEMD), which employs FABEMD principle and decomposes a color image into color bidimensional intrinsic mode components based on hierarchical local spatial variation of image intensity and color. In fact, FABEMD facilitates the extension of the BEMD process for color images in a convenient and useful way, whereas the other interpolation based BEMD techniques appear unsuitable for this purpose. In FABEMD, order statistics filters are employed to estimate the envelope surfaces from the data instead of surface interpolation, which enables fast decomposition and well characterized bidimensional intrinsic mode components. Second, the CBEMD is utilized in this paper for adjusting and/or modifying the trend of color images. In this process, the image is reconstructed by adding the color bidimensional intrinsic mode components after applying suitably selected weights. Test results with real images demonstrate the potential of the proposed CBEMD method for color image processing, which include color trend adjustment.

A moving space domain window is used to implement a Maximum Average Correlation Height (MACH) filter which can be locally modified depending upon its position in the input frame. This enables adaptation of the filter dependant on locally variant background clutter conditions and also enables the normalization of the filter energy levels at each step. Thus the spatial domain implementation of the MACH filter offers an advantage over its frequency domain implementation as shift invariance is not imposed upon it. The only drawback of the spatial domain implementation of the MACH filter is the amount of computational resource required for a fast implementation. Recently an optical correlator using a scanning holographic memory has been proposed by Birch et al [1] for the real-time implementation of space variant filters of this type. In this paper we describe the discrimination abilities against background clutter and tolerance to in-plane rotation, out of plane rotation and changes in scale of a MACH correlation filter implemented in the spatial domain.

A practical challenge that designers of hyperspectral (HS) target detection algorithms must confront is the variety of spectral sampling properties exhibited by various HS imaging sensors. Examples of these variations include different spectral resolutions and the possibility of regular or irregular sampling. To confront this problem, we propose construction of a spectral synthetic discriminant signature (SSDS). The SSDS is constructed from q spectral training signatures which are obtained by sampling the original target signature. Since the SSDS is formulated offline, it does not impose any burden on the processing speed of the recognition process. Results on our HS scenery show that use of the SSDS in conjunction with the spectral fringe-adjusted joint transform correlation (SFJTC) algorithm provides spectrallyinvariant target detection, yielding area under ROC curve (AUROC) values above 0.993.

Efficient recognition and clearance of subsurface land mine patterns has been one of the challenging humanitarian and military tasks. Among the several subsurface land mine patterns recognition techniques available, passive imaging techniques are more convenient, safer with good probability of recognition. There exist extensive applications where the joint-transform correlation algorithms have been used for efficient pattern recognition. However, among the several pattern recognition algorithms exist for subsurface land mines, the joint-transform correlation ones has been underrepresented. This paper presents the application of an efficient wavelet-filter joint transform correlation (WFJTC) algorithm for the recognition of passive imagery of subsurface land mines in highly cluttered scenarios, using intensity and polarization-based imagery. We further improve the recognition efficiency of the WFJTC proposing a combined optical-digital enhancement approach. Improvements will be justified using correlation performance metrics.

Surveillance and its security applications have been critical subjects recently with various studies placing a high demand on robust computer vision solutions that can work effectively and efficiently in complex environments without human intervention. In this paper, an efficient illumination invariant template generation and tracking method to identify and track abandoned objects (bags) in public areas is described. Intensity and chromaticity distortion parameters are initially used to generate a binary mask containing all the moving objects in the scene. The binary blobs in the mask are tracked, and those found static through the use of a 'centroid-range' method are segregated. A Laplacian of Gaussian (LoG) filter is then applied to the parts of the current frame and the average background frame, encompassed by the static blobs, to pick up the high frequency components. The total energy is calculated for both the frames, current and background, covered by the detected edge map to ensure that illumination change has not resulted in false segmentation. Finally, the resultant edge-map is registered and tracked through the use of a correlation based matching process. The algorithm has been successfully tested on the iLIDs dataset, results being presented in this paper.

An Automated Target Recognition system (ATR) was developed to locate and target small object in images and videos. The data is preprocessed and sent to a grayscale optical correlator (GOC) filter to identify possible regionsof- interest (ROIs). Next, features are extracted from ROIs based on Principal Component Analysis (PCA) and sent to neural network (NN) to be classified. The features are analyzed by the NN classifier indicating if each ROI contains the desired target or not. The ATR system was found useful in identifying small boats in open sea. However, due to "noisy background," such as weather conditions, background buildings, or water wakes, some false targets are mis-classified. Feedforward backpropagation and Radial Basis neural networks are optimized for generalization of representative features to reduce false-alarm rate. The neural networks are compared for their performance in classification accuracy, classifying time, and training time.

With excellent physical properties the photorefractive crystals, such as BSO (Bi₁₂SiO₂₀), BaTiO₃ and GaAs materials, have, can be widely used in optical correlator to implement auto pattern recognition. As the basic devices in optical correlator, the properties of optically-addressed spatial light modulator are very important. By analyzing the dynamic process of the BSO spatial light modulator, especially the changes of the read-out light while in writing under various operation modes, the distinctness between various operation modes is summarize. Furthermore, considered with the photo-induced current pulses, the method to optimize the BSO spatial light modulator is proposed. The BSO spatial light modulator working in optimum operation mode is used to design a optical correlator to implement auto pattern recognition.