Capsule endoscopy images classification by color texture and support vector machine

Capsule endoscopy (CE) is considered to be a state-of-the-art imaging modality for digest tract diseases detection, especially for small intestine, which is unreachable by traditional endoscopy techniques. However, the large number of images produced by the procedure, about 60,000 images for each examination, cause a time consuming and attention intensive task for physicians, necessitating the development of computer aided detection system. In this paper, we propose a computerized scheme to discriminate among normal CE images and CE images with three common diseases in GI tract, i.e., bleeding, ulcer and tumor. To achieve this goal, features related to color texture characteristics of CE images are extracted. Based on the features, multiclass support vector machine (SVM) built from binary SVM is further applied to separate different CE images. Preliminary experiments on our present image data demonstrate that it is promising to employ the proposed scheme built upon one-against-one binary SVM to differentiate different CE images.