An automatic ulcer detection scheme using histogram in YIQ domain from wireless capsule endoscopy images

Being one of the most effective video technologies, wireless capsule endoscopy (WCE) offers the physicians to diagnose the gastrointestinal (GI) diseases like ulcer non-invasively. Physicians, while analyzing the WCE videos, find it tedious to detect ulcer because of the huge amount of image frames present in WCE videos. This tedious reviewing process at times leads to inaccuracy in diagnosing ulcer. This paper proposes an automatic technique to detect ulcer frames from WCE videos utilizing the histogram in Y plane of Y I Q color space which utilizes human color-response characteristics. Exhaustive experimentation on publicly available WCE video database validate that significant differences can be obtained between ulcer and non-ulcer images in histogram patterns of Y plane. Cumulative pixel number in Y plane over an optimum threshold is chosen as feature through histogram analysis. Moreover, advantage in computation and implementation is ensured through the proposed 1-D feature for ulcer detection. The supervised support vector machine (SVM) classifier with Gaussian radial basis function (RBF) kernel is used to evaluate the classification performance.