On optimization of active contour methods in colonoscopy image segmentation

As colonoscopy is the standard screening approach for colorectal polyps, and first step of correct classification efficient automatic diagnostics accurate detection segmentation existing it worth researching systematically, how databases are responding to two most influential variational methods, geodesic Chan–Vese active contour methods. Due quality variation databases, pre-processing steps made. Then, 14 various filtered images evaluated as different inputs methods using Sørensen–Dice Similarity Coefficient a performance measurement metric. The effects initial mask shape its size together with number iterations, contraction bias smoothness factor were studied. necessary iterations depended on image resolution. In general, Chan-Vese method showed more efficiency match actual polyp than one an possibly located within area. Preprocessing such reflection removal, background subtraction mean or median filtering improve coefficient can by 0.3.