The Real Time Endoscopic Image Analysis Algorithm

Abstract

New algorithm for the real time processing of narrow--band endoscopic images with a highly productive distributed intellectual analytic decision--making system for multiscale endoscopic diagnostics is presented. The algorithm uses scale--invariant feature transform detector, computing skeletons of gastric mucosa pit-patterns, "Bag of visual words" ("Bag of features") method, and K--means method for key points. Resulting algorithm is completely automated, performs real time analysis, and does not require preliminary selection of interest area. Image classification accuracy exceeds 78%. The use of neural network recognition improves image classification accuracy up to 93%.