Texture classification of images from endoscopic capsule by using MLP and SVM – a comparative approach

Abstract

This article reports a comparative study of Multilayer Perceptrons (MLP) and Support Vector Machines (SVM) in the classification of endoscopic capsule images. Texture information is coded by second order statistics of color image levels extracted from co-occurrence matrices. The co-occurrence matrices are computed from images rich in texture information. These images are obtained by processing the original images in the wavelet domain in order to select the most important information concerning texture description. Texture descriptors calculated from co-occurrence matrices are then modeled by using third and forth order moments in order to cope with non-Gaussianity, which appears especially in some pathological cases. Several color spaces are used, namely the most simple RGB, the most related to the human perception HSV, and the one that best separates light and color information, which uses luminance and color differences, usually known as YCbCr.