A novel color textural feature towards capsule endoscopy video summary

Capsule endoscopy (CE) has found its wide application in examination of the small intestine due to its advantage of reviewing this part that is unreachable for traditional endoscopy. However, a computer aided detection system is in needed since it costs an experienced physician much time to go through all the video data. As an attempt to overcome this shortcoming, CE video summary seems to be a feasible solution for this challenging problem. In this paper, we study the application of a novel textural feature aiming for CE video summary. To overcome the possible illumination and camera motion in the imaging process, spatial pyramid based color histogram is first proposed to describe the color content in a CE frame. Then, Gaussian based local binary pattern is further utilized to describe the textural features in a CE frame. Affinity propagation method is further utilized to recognize the representative frame in a video segment. Preliminary experiments demonstrate an encouraging video abstract performance in terms of fidelity and recall.