Publication of the journal

The section is currently being updated

One of the commonly used algorithms in computer vision is the human skin segmentation in the image. The most common approach to skin segmentation is training a binary classifier to separate human skin (foreground) pixels from the background ones. In the paper the problem of training such binary classifiers has been discussed. A smoothed ROC curve has been constructed. The recommendations for a test set obtaining for correction comparison of binary classifiers have been given. Using the proposed method the training of single Gaussian and Bayesian classifiers by the proposed method has been carried out. The results of testing and comparison of the obtained classifiers have been given. It has been shown that this method can improve the quality of skin segmentation up to 4 % of TRP metric and up to 10% of FRP. The Bayesian classifier, trained according to the proposed method provides the best segmentation results among all considered classifiers.

124498, Moscow, Zelenograd, Bld. 1, Shokin Square, MIET, editorial office of the Journal "Proceedings of Universities. Electronics", room 7231

+7 (499) 734-62-05
magazine@miee.ru