An algorithm for automated segmentation for bleeding detection in endoscopic images

Wireless capsule endoscopy is an important advanced diagnostics method. It produces huge amount of images during travel through patient's digestive tract and that usually requires automated analysis. One of the most important abnormalities is bleeding and automated segmentation for bleeding detection is an active research topic. In this paper we propose an algorithm for automated segmentation for bleeding detection in capsule endoscopy images. The algorithm uses block based segmentation where average saturation from the HSI model and skewness and kurtosis of uniform local binary patterns histogram are used as features for the support vector machine classifier. Support vector machine parameters are tuned using grid search. The proposed method was tested using standard benchmark images and compared with other approaches from literature using Dice similarity coefficient and misclassification error as metrics, where it obtained better results using simpler features.