Segmentation of Clinical Endoscopic Images Based on the Classification of Topological Vector Features

In this work, we describe a prototype of an automatic segmentation system and annotation of endoscopy images. The used algorithm is based on the classification of vectors of the topological features of the original image.
We use the image processing scheme which includes image preprocessing, calculation of vector descriptors defined for every point of the source image and the subsequent classification of descriptors. Image preprocessing includes finding and selecting artifacts and equalizating the image brightness. In this work, we give the detailed algorithm of the construction of topological descriptors and the classifier creating procedure based on mutual sharing the AdaBoost scheme and a naive Bayes classifier. In the final section, we show the results of the classification of real endoscopic images.