An advanced WCE video summary using relation matrix rank

Jia Sen Huo, Yuexian Zou, Lei Li

Abstract: Wireless Capsule Endoscopy (WCE) is a promising new solution for gastrointestinal disease detection, which is able to view the entire gastrointestinal tract without pain. Since there are more than 50,000 frames in one WCE video from each test, the automatic computer-aided technique is highly demanded to remove the redundant frames while remaining the medical information. This paper presents a systematic method to achieve the effective WCE video summary using whole WCE video frames. A cascade color and texture feature based most representative frame extraction algorithm (CCTS-MRFE) has been developed. The color feature extraction method is developed in HSV space. Meanwhile, the texture feature extraction is determined by the proposed block edge directivity descriptor (BEDD) method in gray space. A relation matrix rank (RMR) based most representative frame (MRF) selection approach further improving the performance of the proposed CCTS-MRFE algorithm has been proposed. The preliminary experimental results show that the proposed methods are able to achieve the effective WCE video summary.