Integration of Valley Orientation Distribution for Polyp Region Identification in Colonoscopy

This work presents a region descriptor based on the integration of the information that the depth of valleys image provides. The depth of valleys image is based on the presence of intensity valleys around polyps due to the image acquisition. Our proposed description method consists of defining, for each point, a series of radial sectors around it and then accumulate the maxima of the depth of valleys image only if the orientation of the intensity valley coincides with the orientation of the sector above. We apply our descriptor to a prior segmentation of the images and we present promising results on polyp detection, outperforming another approach that also integrate depth of valleys information.