CALE & ROTATION INVARIANT TARGET RECOGNITION USING LOGARITHMICALLY PRE-P

Abstract

Moving target detection and tracking is an important technique of video processing for its huge potential in military and other applications. The major focus in such scenarios is always a comprehensive target recognition comprising correct classification of target object to facilitate target locking and tracking. This research encompasses development of enhanced technique to recognize the target objects in images irrespective of distortions. Scale and rotation invariance has been the main challenge in the detection of the objects in various military based machine recognition projects. The translation of image in terms of sensor geometry modeled by a mammal retina enables us to resolve the invariance issues. Logarithmic mapping is a mathematical transform that presents the scale or rotation variations in terms of shifts which enables us to classify images correctly. Endeavour has been made to combine logarithmic mapping with correlation pattern recognition methods for improved detection results, which can be swiftly used for military applications like target tracking in videos captured by unmanned aerial vehicles. Having tackled invariance issues, correlation based detection is focused to find the best match of an incoming image with the reference database depending on maximum average correlation height. Correlation pattern recognition has emerged as a comprehensive tool to exploit matching issues linked with target detection. An ongoing research in this field provides us with an opportunity to use the latest versions of correlation filters to develop some enhanced detection capabilities. The three primary issues dealt with in the design of correlation filters include the ability to suppress clutter and noise, easy detection of the correlation peaks and distortion tolerance. Each of the correlation filter is synthesized using representative training images of a respective target. This allows the filter to exhibit distortion tolerance over a limited range of target orientations. 2 _______________________________________________________________ Scale & Rotation Invariant Target Recognition By joining these filters with the strengths of logarithmic translation we get an added advantage which is desired in most of the recognition applications. The detection results are obtained in the form of an output correlation planes. These planes have also been compared for different filter combinations displaying a wide variety of options for handling target detection challenges. Each product has also been tested extensively by using an image database containing all possible distortions in it and results have been analyzed thoroughly to view the suitability of these filters for tackling invariance problems.