Stochastic models for recognition of occluded objects

Abstract

Recognition of occluded objects in synthetic aperture radar (SAR) images is a significant problem for automatic object recognition. Stochastic models provide some attractive features for pattern matching and recognition under partial occlusion and noise. In this paper, we present a hidden Markov modeling (HMM) based approach for recognizing objects in synthetic aperture radar (SAR) images. We identify the peculiar characteristics of a SAR sensor and using these characteristics we develop feature based multiple stochastic models for a given SAR image of an object. The models exploiting the relative geometry of feature locations or the amplitude of scattering centers in SAR radar return are based on sequentialization of scattering centers extracted from SAR images. In order to improve performance under real world situations, we integrate these models synergistically using their probabilistic estimates for recognition of a particular object at a specific azimuth. Experimental results are presented using real SAR images with varying amount of occlusion. © Springer-Verlag Berlin Heidelberg 2000.