Fast and robust image processing algorithms based on approximations to order statistics

A test system with four cameras in the infrared and visual spectra is under development at FFI (The Norwegian Defence Research Establishment). The system may be mounted on a jet aircraft or may be used in a land-based version. It can be used for image acquisition or for testing of automatic target recognition (ATR) algorithms. The sensors on board generate large amounts of data, and the scene may be rather cluttered or include anomalies (e.g. sun glare). This means we need algorithms which are robust, fast, able to handle complex scenes, and data from up to four sensors simultaneously. Typically, estimates of mean and covariance are needed for the processing. However, the common maximum likelihood (ML) estimates are in general too sensitive towards outliers. Algorithms based on order statistics are known to be robust and reliable. However, they are computationally very heavy. But approximations to order statistics do exist. Median of medians is one example. This is a technique where an approximation of the median of a sequence is found by first dividing the sequence in subsequences, and then calculating median (of medians) recursively. This technique can be applied for estimating the mean as well as the standard deviation. In this paper we extend this method for estimating the covariance matrix and the mean vector, and discuss the strategy with respect to robustness and computational efficiency. Applications for use in image processing and pattern recognition are given.

A research platform with four cameras in the infrared and visible spectral domains is under development at the Norwegian Defence Research Establishment (FFI). The platform will be mounted on a high-speed jet aircraft and will primarily be used for image acquisition and for development and test of automatic target recognition (ATR) algorithms. The sensors on board produce large amounts of data, the algorithms can be computationally intensive and the data processing is complex. This puts great demands on the system architecture; it has to run in real-time and at the same time be suitable for algorithm development. In this paper we present an architecture for ATR systems that is designed to be exible, generic and efficient. The architecture is module based so that certain parts, e.g. specific ATR algorithms, can be exchanged without affecting the rest of the system. The modules are generic and can be used in various ATR system configurations. A software framework in C++ that handles large data ows in non-linear pipelines is used for implementation. The framework exploits several levels of parallelism and lets the hardware processing capacity be fully utilised. The ATR system is under development and has reached a first level that can be used for segmentation algorithm development and testing. The implemented system consists of several modules, and although their content is still limited, the segmentation module includes two different segmentation algorithms that can be easily exchanged. We demonstrate the system by applying the two segmentation algorithms to infrared images from sea trial recordings.

In this paper, probabilistic fusion of multi-sensor data is applied to mine detection. Probabilistic fusion combines information in the form of scores from automatic target recognition (ATR) algorithms for each sensor. This fusion method has previously demonstrated improved mine detection performance when used with multi-sensor data from the Mine Hunter/Killer system. The sensor suite includes a ground- penetrating radar, metal detectors, and an IR camera; data were collected at a prepared test site. Results of applying the probabilistic fusion method to recent MH/K multi-sensor data using various new ATR algorithms are presented and analyzed in detail. Changes in detection performance are quantified for different combinations of the various ATR algorithms and sensors. It is shown that fusion improves mine detection performance even when the individual sensor and ATR algorithms have very different performance levels. This implies that multi-sensor approaches to mien detection should continue to be pursued.

For the Automatic Target Recognition (ATR) algorithm, the quality of the input image sequence can be a major determining factor as to the ATR algorithm's ability to recognize an object. Based on quality, an image can be easy to recognize, barely recognizable or even mangled beyond recognition. If a determination of the image quality can be made prior to entering the ATR algorithm, then a confidence factor can be applied to the probability of recognition. This confidence factor can be used to rate sensors; to improve quality through selectively preprocessing image sequences prior to applying ATR; or to limit the problem space by determining which image sequences need not be processed by the ATR algorithm. It could even determine when human intervention is needed. To get a flavor for the scope of the image quality problem, this paper reviews analog and digital forms of image degradation. It looks at traditional quality metric approaches such as peak signal-to-noise ratio. It examines a newer metric based on human vision data, a metric introduced by the Institute for Telecommunication Sciences. These objective quality metrics can be used as confidence factors primarily in ATR systems that use image sequences degraded due to transmission systems. However, to determine the quality metric, a transmission system needs the original input image sequence and the degraded output image sequence. This paper suggests a more general approach to determining quality using analysis of spatial and temporal vectors where the original input sequence is not explicitly given. This novel approach would be useful where there is no transmission system but where the ATR system is part of the sensor, on-board a mobile platform. The results of this work are demonstrated on a few standard image sequences.

This paper smmnarizes a preliminary investigation into a space-based multi-class system-oriented High Range Resolution (HRR) Automatic Target Recognition (ATR) algorithm. Recent conflicts have highlighted the need to detect, track and recognize time critical moving ground targets in all-weather standoff (airborne and space-based) sensor platforms. Current sensor technologies based on Moving Target Indication (MTI) and/or Synthetic Aperture Radar (SAR) imaging are inadequate. While MT1 provides the ability to detect and track ground moving targets, it lacks target discrimination capability due to limited bandwidth. SAR has target discrimination capability for stationary targets, but target motion will blur the image. As a result, ATR of ground moving targets using HRR radar has recently received increased interest within the Air Force. The baseline algorithms presented in this paper are template-based mean squared error classifiers. The paper includes a brief introduction into recent HRR ATR efforts at the Air Force Research Laboratory Sensors Directorate's ATR Assessment Branch (AFRL/SNAA). Motivations and limitations for space- based sensors are discussed along with potential sensor fusion algorithms. The paper concludes by discussing the advantages of HRR ATR for feature aided

For new military aircraft a specification of sensor characteristics and performance is required at an early stage in the design cycle, well before testing of a prototype. In the early days of military aviation the Johnson criteria [1] were used to determine the sensor resolution required for target recognition by a human. In the present day sensor data are processed by computer using various Automatic Target Recognition (ATR) algorithms. However there is no accepted method for predicting the sensor resolution and SNR required for reliable ATR and hence there is risk that any chosen sensor may fail to support the required ATR performance. This paper reports a study into the use of publicly-available CAD models for aircraft to address this requirement. The study considers the worst-case confusion between two views of 15 different aircraft types. For simplicity only rotations by an angle θ about the Z (vertical) axis are considered. Firstly the sensor resolution and noise level is fixed. Then for each aircraft type and view angle an ensemble of synthetic silhouettes are generated. Using these ensembles, a-posteriori distributions of 5 standard scale-invariant shape features (eccentricity, orientation, solidity, circularity and bounding box aspect ratio) are estimated for each view angle θ. The performance of ATR at the given resolution and noise level is predicted by estimating the Bayes Error Rate [2] when deciding between each aircraft type and the 14 non-matching types using these features. The worst-case confusion in terms of erroneous aircraft type and view angle is identified. The sensor resolution is then changed and the above process repeated to investigate the effect of varying sensor resolution on performance. As expected, high sensor resolution leads to low probability of misclassification, even in the worst-case. Reduction in resolution and increasing noise level causes the Bayes Error Rate to rise quickly. The Bayes Error Rate gives a fundamental limit to the reliability of classification, irrespective of the actual type of classification algorithm used. The predictions from the model are confirmed by testing against a standard classifier for specific discrimination examples. Further development of the approach presented is expected to yield a method for specifying sensor resolution requirements for specific ATR problems.

Following the tendency of increasingly using imaging sensors in military aircraft, future combat airplane pilots will need onboard artificial intelligence for aiding them in image interpretation and target designation. This document presents a system which is able to simulate high-resolution artificial SAR imagery and thereby facilitates automatic target recognition (ATR) algorithm development. The system provides a comprehensive interface that allows dynamically requesting imagery depending on the location and heading of a simulated carrier platform. Landscapes, structures and target signatures are generated based on digital terrain data and target models. An assessment of dissimilar database preparations for sensor simulation was done with respect to the different properties of SAR imaging compared to optical imaging. The document presents selected results for specific landscape elements. Post-processing algorithms for overcoming weaknesses of digital terrain databases and improving image realism are presented. Simulated sensor imagery is useful in a wide range of applications, two of which are training of ATR algorithms and sensor simulation in flight simulation environments. Using an existing ATR method as an example, the applicability and the influences of synthetic imagery on ATR training are shown and first approaches on how to validate the correctness of the imagery are explained. The integration of the system into a flight simulator in the context of interfacing and control topics serves as a concluding example.

The paper presents the results of developing an algorithm for automatic target recognition in broadband (0.1-10) terahertz images. Due to the physical properties of terahertz radiation and associated hardware, such images have low contrast, low signal-to-noise ratio and low resolution – i.e. all the characteristics of a low-count images. Therefore, standard recognition algorithms designed for conventional images work poorly or are not suitable at all for the problem considered. We have developed a fundamentally different approach based on clustering 2D point clouds in accordance with a set of predefined patterns. As a result, we reduce the problem of target recognition to the problem of maximizing the image data likelihood with respect to the classes of model objects up to the size and position. The resulting recognition algorithm has a structure close to that of the well-known EM algorithm; its formal scheme is at the end of the paper.

In this paper we empirically examine the probability of detection (PD) and false alarm rate (FAR) for crash site detectionusing polarimetry to discriminate between aircraft target signatures within natural clutter. To date, the search and rescueprogram has tried several automatic target recognition (ATR) algorithms from the literature. While PD seems reasonablewith these algorithms, the FAR is too high (—iO's per square kilometer). The objective of our analysis is to determine if thisis a limitation of the ATR algorithms tried, or if this is the best that can be hoped for given the polarimetric statistics of thetarget and clutter.Keywords: Search and Rescue, Polarimetry, Cameron Transformation, Synthetic Aperture Radar 1. INTRODUCTION Beaconless search and rescue utilizes a synthetic aperture radar (SAR) system to search for downed aircraft in remoteregions. Low frequency SAR is capable of quickly imaging large areas and penetrating bad weather and foliage cover.Coupled to a computer processing system, that includes automatic crash site detection algorithms, SAR has the potential tobecome a significantly useful search tool. We are currently determining the fundamental limits on the probability ofdetection versus the false alarm rate for crash site detection using radar polarimetry. In this paper we present our analysis ofdata from Half Moon Bay (HMB), California, Teton National Park, Wyoming, and from scale-model tests done by SystemPlanning Corporation (SPC), Virginia.

Radar range-profile classification of targets in clutter (automatic target recognition) is particularly challenging because of the multimodal nature of the data classes. Typically, classical techniques are not adequate, or must be customized to provide good classification performance. The cut and try method of model synthesis is time consuming and may never yield the insight necessary for good classifier performance on unseen data. Inductive classification algorithms are appropriate for these challenging problems because they not only synthesize classifiers, but provide critical information about the data itself which can then be used for further refining (or redefining) of classifier inputs and synthesis strategies. This paper describes one problem solution using radar turntable data for classifier training and testing. Classical classifiers (nearest mean and Fischer pairwise) are compared to polynomial neural network (PNN) and multilayer perceptron (MLP) classifiers. >

Reliance on Automated Target Recognition (ATR) technology is essential to the future success of Intelligence, Surveillance, and Reconnaissance (ISR) missions. Although benefits may be realized through ATR processing of a single data source, fusion of information across multiple images and multiple sensors promises significant performance gains. A major challenge, as ATR fusion technologies mature, is the establishment of sound methods for evaluating ATR performance in the context of data fusion. The Deputy Under Secretary of Defense for Science and Technology (DUSD/S&T), as part of their ongoing ATR Program, has sponsored an effort to develop and demonstrate methods for evaluating ATR algorithms that utilize multiple data source, i.e., fusion-based ATR. This paper presents results from this program, focusing on the target detection and cueing aspect of the problem. The first step in assessing target detection performance is to relate the ground truth to the ATR decisions. Once the ATR decisions have been mapped to ground truth, the second step in the evaluation is to characterize ATR performance. A common approach is to vary the confidence threshold of the ATR and compute the Probability of Detection (PD) and the False Alarm Rate (FAR) associated with each threshold. Varying the threshold, therefore, produces an empirical performance curve relating detection performance to false alarms. Various statistical methods have been developed, largely in the medical imaging literature, to model this curve so that statistical inferences are possible. One approach, based on signal detection theory, generalizes the Receiver Operator Characteristic (ROC) curve. Under this approach, the Free Response Operating Characteristic (FROC) curve models performance for search problems. The FROC model is appropriate when multiple detections are possible and the number of false alarms is unconstrained. The parameterization of the FROC model provides a natural method for characterizing both the operational environment and the ability of the ATR algorithm to detect targets. One parameter of the FROC model indicates the complexity of the clutter by characterizing the propensity for false alarms. The second parameter quantifies the separability between clutter and targets. Thus, the FROC model provides a framework for modeling and predicting ATR performance in multiple environments. This paper presents the FROC model for single sensor data and generalizes the model to handle the fusion case.

Neptec Design Group has developed a 3D automatic target recognition and pose estimation algorithm technologydemonstrator in partnership with Canadian DND. This paper discusses the development of the algorithm to work withreal sensor data. The recognition approach uses a combination of two algorithms in a multi-step process. The twoalgorithms provide uncorrelated metrics and are therefore using different characteristics of the target. This allows thepotential target dataset to be reduced before the final selection is made. In a pre-processing phase, the object data issegmented from the surroundings and is re-projected onto an orthogonal grid to make the object shape independent ofrange. In the second step, a fast recognition algorithm is used to reduce the list of potential targets by removingunlikely cases. Then a more accurate, but slower and more sensitive, algorithm is applied to the remaining cases toprovide another recognition metric while simultaneously computing a pose estimation. After passing some self-consistency checks, the metrics from both algorithms are then combined to provide relative probabilities for eachdatabase object and a pose estimate. Development of the recognition and pose algorithm relied on processing of real3D data from civilian and military vehicles. The algorithm evolved to be robust to occlusions and characteristics ofreal 3D data, including the use of different 3D sensors for generating database and test objects. Robustness also comesfrom the self-validating abilities and simultaneous pose estimation and recognition, along with the potential forcomputing error bounds on pose. Performance results are shown for pseudo-synthetic data and preliminary tests with acommercial imaging LIDAR.Keywords: 3D, Automatic target recognition, pose estimation

This paper presents a description of a Bayesian neural network (BNN) automatic target recognition (ATR) algorithm specifically developed to efficiently exploit the correlated relationships between the joint dimensional distribution properties of the multiple features. A multidimensional clustering process is used to define a Parzen estimate which approximates the Bayesian a posteriori decision function, but without limiting constraints regarding multimodal or correlated density function forms in multidimensional feature space. The mathematical derivation is placed in the form of a multilayer neural network, including Kernel function variance adaptation to accommodate small training data sets, and Kernel function tail saturation for robust handling of multidimensional feature data not represented in the training data. The BNN is shown to be a hybrid Bayesian/neural net classifier. It has a robust Bayesian statistical classifier as an initial state, and a feedback learning structure which enables learning corrections localized only to the feature space in the vicinity of the error. In addition, a method is described for integrating scene context information, spatial constraints, and a priori probability information into the BNN classifier. Example uses of the BNN algorithm are discussed for recent programs which obtained multiple feature ATR improvements from using the BNN with multiple frequencies, polarizations, aspect angles, and 3D interferometric SAR features.

Most of the research on developing automatic target recognition (ATR) algorithms for acoustic-seismic landmine detection platforms has been focused on using geometric features, such as size and shape, of anomaly to distinguish between mines and clutter. This approach has achieved some success especially in detecting larger anti-tank mines. However, for smaller anti-personnel landmines, the difference in geometric features between mines and clutter can be very small, if any. To improve the detection vs. false alarm rates, it is necessary to incorporate other features into the ATR process. It has been observed from the collected acoustic data that areas with buried mines reveal more complicated surface vibration structures, such as the ring-like pattern, at certain frequencies than what a one-dimensional lumped mass-spring-dashpot model can describe. In this paper, we utilize the distributed mine/soil interaction model developed by the University of Mississippi to describe the surface vibration patterns. We develop a modified Hankel transform to extract features from areas under interrogation. Under such transform, concentration of energy is closely related to an object's physical properties. The frequency at which the energy concentration occurs corresponds to the object's natural frequency, while the corresponding Bessel basis captures its mode shape. After de-noising the transformed data, we use the frequencies, Bessel bases, and magnitudes of the energy concentrations, together with other geometric features, to form the feature vectors. We tested these features on a dataset consisting of anti-tank and anti-personnel mines as well as blank areas and metallic and non-metallic clutter. Classifiers designed based on the combined geometric and model-based features perform significantly better than those based on the geometric features alone.

Synthetic Aperture Radar (SAR) image scene modeling tools are of high interest to Automatic Target Recognition (ATR) algorithm evaluation because they allow the testing of ATR's over a wider range of extended operating conditions (EOCs). Typical EOCs include target aspect, target configuration, target obscuration, and background terrain variations. A baseline terrain image synthesis technique empirically derived probability density functions (pdfs) for various terrain types from measured data to allow the simulation of user defined scenes. Initial full scene simulation experiments that applied this technique to the MSTAR data showed that using measured images as a data source for creating distribution functions in artificial scenes can introduce error, unless the proper spatial autocorrelation is also modeled. Measured SAR scene pixels have non-zero autocorrelation that blurs edges between different terrain types and creates a texture in the clutter regions of the image. Unfortunately, applying simple blurring techniques, such as a moving weighted window, to model autocorrelation mutes the second and higher moments of the pixel amplitude statistics. We propose a technique that models spatial autocorrelation while preserving the desired the amplitude statistics within each defined terrain class group.