Texture Image Segmentation Using Wavelet Filters and Cellular Neural Network

This paper presents a texture segmentation algorithm based on Discrete Wavelet Frames(DWF) and Cellular Neural Network(CNN). DWF, zero-crossing, texture energy and selective local averaging are used to get a texture feature extraction and t,o form feature images. Each feature image is segmented into parts by several gray range in its gray histogram. Resulting in the number of pixels that conform to each gray range composition of every feature image, segmentation can be got by those compositions that have a big pixel number. We call this method "CompositionArray method". A new CNN called "Multi-objective CNN" is developed using the fundamental theory of the traditional CNN, it is a multi-objective neural processing instead of 1 and -1 only. The noise of Composition-Array processing can be removed using this new CNN, and we can get a perfect final segmentation result,.