THE SCALING PROBLEM IN THE CLASSIFICATION OF IMAGES BY SPATIAL FILTERING.
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OHIO SCHOOL OF ENGINEERING
Pagination or Media Count:
The purpose of this investigation is to modify an existing model of the human visual system, based on the spatial filtering of the two-dimensional Fourier transform of an input stimulus, to allow for considerable scale changes in the input stimuli. The primary tool compensation, the scaling property of the Fourier transform is discussed and its application to discrete two-dimensional patterns is presented. A method for determining the scale factor of an input pattern, given some a priori information on several input patterns, is described. The scale factors derived from this method are shown to relate to the human estimation of size. A complete scaling algorithm, applicable to many pattern recognition problems, is presented. The results of two computer simulations, using this algorithm, are presented. The first simulation demonstrates the successful classification of handwritten geometric outlines, with moderate scale differences, and the second simulation shows the classification of Russian characters with scale differences of up to 3.5 to 1. These results clearly demonstrate both the need for the scaling algorithm and its effectiveness. Author