Bleeding detection in wireless capsule endoscopy images based on color invariants and spatial pyramids using support vector machines

Wireless capsule endoscopy (WCE) is a revolutionary imaging technique that enables detailed inspection of the interior of the whole gastrointestinal tract in a non-invasive way. However, viewing WCE videos is a very time-consuming, and labor intensive task for physicians. In this paper, we propose an automatic method for bleeding detection in WCE images. A novel series of descriptors which combine color and spatial information is designed in a way that local and global features are also incorporated together. And a kernel based classification method using histogram intersection or chi-square is deployed to verify the performance of the proposed descriptors. Experiments demonstrate that the proposed kernel based scheme is very effective in detecting bleeding patterns of WCE images.