Backpropagation Algorithm for Multiresolution Image Classification

CLASSIFICATION Hossam Osman and Steven D. Blostein Department of Electrical and Computer Engineering, Queen's University, Kingston, Ontario, Canada, K7L 3N6 ABSTRACT This paper proposes a variation of the standard backpropagation (BP) training algorithm for the particular application of multiresolution image classi cation. The proposed variation is that, during the backward phase of training, hidden units whose inputs are extracted from shared image space inhibit one another. The paper shows that BP training that incorporates this inhibition not only solves the image classi cation problem, but also removes redundant multiresolution network inputs aiming at attaining only those necessary for classi cation. Experimental results that demonstrate the viability of the proposed BP variant are presented.