Block Based Histogram Feature Extraction Method for Bleeding Detection in Wireless Capsule Endoscopy

Wireless capsule endoscopy (WCE) is a non-invasive video technology to detect small intestinal diseases, such as bleeding. It provides straight vision of patient's entire gastrointestinal (GI) tract, however examining large amount of images by physician takes longer time thus computer-aided scheme draws attention. In this paper, an effective features extraction scheme is proposed for automatic bleeding frame detection in WCE video. Instead of using red (R), green (G), and blue (B) plane separately a transform plane consisting of red to green pixel intensity ratio is used. Also rather than considering individual pixel, a surrounding neighborhood block of that individual pixel is considered which mitigates individual pixel randomness problem. A simple but consistent block statistic, namely maximum value of a block, is chosen as representative feature for each block. Histogram is found by utilizing all block maxima and frequency of histogram is considered as feature. For the purpose of classification, k-nearest neighbors (KNN) classifier is employed. Extensive experimentation is carried out upon several WCE videos, which are collected from a publicly available database and satisfactory performance result is obtained in comparison to some of the exiting methods.