Obscene image recognition based on model matching and BWFNN

Abstract

In this paper the obscene images first primarily recognizes through the human skin color detection and key point model matching. The other images that are not confirmed extract characteristic of obscene images through edge detection, posture estimation and wavelet compression, and then recognized using the optimizing broaden weighted fuzzy neural network, which is called two-phase recognizing method. The experiment indicates the method that this paper present can recognize the obscene images effectively. © Springer-Verlag Berlin Heidelberg 2005.