Texture analysis for ulcer detection in capsule endoscopy images

Capsule endoscopy (CE) has gradually seen its wide application in hospitals in the last few years because it can view the entire small bowel without invasiveness. However, CE produces too many images each time, thus causing a huge burden to physicians, so it is meaningful to help clinicians if we can employ computerized methods to diagnose. This paper presents a new texture extraction scheme for ulcer region discrimination in CE images. A new idea of curvelet based local binary pattern is proposed as textural features to distinguish ulcer regions from normal regions, which makes full use of curvelet transformation and local binary pattern. The proposed new textural features can capture multi-directional features and show robustness to illumination changes. Extensive classification experiments using multilayer perceptron neural network and support vector machines on our image data validate that it is promising to employ the proposed texture features to recognize ulcer regions in CE images.