TEXTURE DISCRIMINATION USING DECONVOLUTION FILTERS

Abstract

A new experimental technique for texture discrimination is proposed and it involves the use of deconvolution filters in a coherent optical computing system.

A texture is considered as a convolution of a sub pattern with an array of impulse functions. A procedure to synthesize a deconvolution filter, given a proto sample of a texture, is described. The filter is used successfully in a coherent optical computer for recognizing a chosen texture in a mixture of textures. The discrimination caused by the deconvolution filter is identified by the presence of characteristic dot pattern in the output plane.

The results on texture discrimination obtained with conventional matched filter methods are compared with those of deconvolution techniques and it is concluded that the deconvolution techniques are superior. Further it is shown that unambiguous discrimination of parent texture is possible with the proposed technique even in cases where seemingly different textures are generated by conditions such as overlapping and differences in background illumination.