WCE video clips segmentation based on abnormality

Wireless Capsule Endoscopy (WCE) is a state-of-the-art technology to examine the entire gastrointestinal tract. Its main disadvantage is long review time for physicians to diagnose diseases, as it will produce over 55,000 frames per patient for one examination. In this paper we propose a novel strategy to segment WCE video clips based on abnormality. The new scheme is based on a non-parametric corner detection method. The k-means clustering is adopted to extract the most representative frames (MRFs) to summarize the video clip. The experiments were performed on the real patient video clips and the results demonstrate that the MRFs consist of frames of interest and abnormalities, such as bleeding, ulcer and tumor.