A histogram based scheme in YIQ domain for automatic bleeding image detection from wireless capsule endoscopy

Abstract

Wireless capsule endoscopy (WCE) is a recent video technology to investigate bleeding in gastrointestinal (GI) tract. Generally, in a WCE video, a physician has to analyze a large number of image frames, which is very time consuming and often leads to human error. Hence, an automatic scheme for bleeding image detection in WCE video has great demand. In this paper, an automatic scheme based on image histogram is proposed to detect bleeding frames in WCE video. In order to compute the histograms from pixel values, instead of using conventional RGB color plane, YIQ color plane is used, which offers scope of utilizing human color-response characteristics. Based on extensive experimentation, it is found that histogram patterns, obtained from bleeding and non-bleeding images in Y, I, and Q planes, exhibit significant differences. Frequency values of three histograms, constructed from the preprocessed WCE image, are used in cascade to obtain desired feature vector. For the purpose of classification, K nearest neighbor (KNN) supervised classifier is used. The performance of the proposed method is tested on several WCE images taken from publicly available WCE video database and it is found that, the proposed method offers superior classification performance, in comparison to that obtained by some existing methods, in terms of accuracy, specificity, and sensitivity.