Implementation and performance comparison of gmm super pixels for automated bleeding detection in wce images

Abstract

GastroIntestinal (GI) Bleeding is the most often experienced anomaly in the gastrointestinal tract and may also show many other intestinal illness, such as vascular tumor, ulcers, polyps, and Crohn s disease. The typical diagnostic procedure involves physical examination of the entire GI tract by means of a professional clinician in order to identify bleeding as one of the most common anomalies of the disease. The first stage aims at offering pre-processing mechanism for WCE image enhancement. WCE is employed for examining the human digestive tract to recognize the areas that are abnormal. Moreover, this is regarded as a challenging task for the recognition of abnormal areas like bleeding because of dark images and poor quality of WCE. In this approach, pre-processing is employed to ease the feature extraction process. Initially, distribution linearization and linear filtering process is employed. The image contrast is enhanced followed by image subtraction and vignette correction. At last, the de-correlation stretching is presented to attain de-noised and enhanced image. The performance analysis shows that the proposed system is better in offering enhanced and sharpened WCE image. newlineThe second stage has an automated system based on Gaussian Mixture Model Superpixels for segmentation and detection of bleeding in the candidate regions. The presented system is then realized with the seven features that covers color and texture attributes that were extracted from GMM super pixels of WCE images. By the bleeding image detection, the regions of bleeding were segmented from them thereby grouping superpixels incrementally depending on the Delta E color variations. The key objective of the third stage is to carry out performance comparison on various existing techniques like color-based feature extraction, 2 histogram-based approach, Discrete Wavelet Transform (DWT), K-Nearest Neighbor (KNN), K-means, and Support Vector Machine (SVM) techniques employed in the detection of bleeding to that of the proposed Gaussian Mixture