Assessment of Crohn's disease lesions in Wireless Capsule Endoscopy images using SVM based classification

Wireless Capsule Endoscopy (WCE) allows a physician to examine the entire small intestine without any surgical operation. The aim of this paper is to perform this assessment of lesions with a level of consistency and accuracy comparable to human observers. This work is focused on analysis of CE images for Crohn's disease lesion. The normal and the lesion images are classified based upon the size, and the severity of the lesion and any surrounding inflammation. For this classification investigation of long range features are necessary. In this paper color, edge and texture features were extracted for the CE images. The color and the edge features are extracted by the MPEG-7 Dominant Color Descriptor and MPEG-7 Edge histogram Descriptor. The texture features are based on the MPEG-7 texture descriptor. The MPEG-7 homogeneous texture descriptor uses Gabor filters of different scales and orientations. The texture feature alone was not sufficiently accurate for the analysis of Crohn's disease. A new method is introduced in this paper that uses Local Binary Pattern (LBP) to extract the textural features. LBP is a local texture descriptor to describe the intensity distribution. Texture contents of an image region are characterized by the distribution of LBP. These features are trained, tested and classified using SVM classifier in order to have supervised learning model. In this proposed work, the accuracy is improved with better performance over other previous methods.