A fully automated ulcer detection system for wireless capsule endoscopy images

Abstract

This paper deals with ulcer region detection of the small bowel from wireless capsule endoscopy images (WCE). Indeed, joining both texture and color show a great usability, to correctly predict abnormal tissues in WCE images with a higher accuracy and reproducibility. In texture analysis, scale is an important information, we can visualize the same texture as being different textures in many scales. The local binary pattern (LBP) shows it's efficiency as texture operator in many studies. A multi-scale approach based on LBP and Laplacian pyramid transform, is proposed here. This rotation-and-scale invariant method aims to distinguish ulcerous regions from a normal ones, in an efficient way. In addition, the proposed approach was applied on the components of four color spaces : RGB, Lab, HSV and CMY. Ulcer detection, was performed using the support vector machine (SVM) [1]. The results obtained validate the efficacity of our proposed system with average accuracy of 95.61%, an average sensitivity of 97.68% and an average specificity of 94.40%.