Polyps Segmentation using Fuzzy Thresholding in HSV Color Space

Colorectal cancer begins in the form of Polyp, abnormal tissue growth on the inner line of the colon or at the rectum. Early detection of polyps by colonoscopy might result in successful treatment through proper diagnosis. Because of varieties in the measure and state of polyps, the conclusion of polyps in colonoscopy recordings is a difficult task. The present work is an attempt to segment polyp regions effectively from colonoscope image frames using Fuzzy thresholding on the V channel of HSV color space. Polyp regions are segmented from healthy regions by grouping the V channel data points (pixel value) of HSV colorspace into three clusters, where every data point in the dataset belongs to one of the three clusters to a certain degree. Based on the separation between the cluster center and the data points, the degree of membership is allotted for each cluster. The data points with the highest membership degrees are used for generating an efficient threshold value, which is unique for every different image. The obtained threshold value results in a binary image, followed by a few morphological operations to get significant results. Next, the refined binary image replicated to the original RGB image, representing only polyp regions. Evaluation of the proposed method using the online kvasir dataset shows satisfactory results and achieves significant accuracy in comparison with previous automatic polyp segmentation methods.