Pornographic image rejection using eigenporn of simplified LDA of skin ROIs images

Abstract

This paper proposes alternative solution for pornographic rejection using eigenporn extracted by simplified linear discriminant analysis (LDA) of skin region of interests (ROIs) of pornographic images. The skin ROI is extracted by skin segmentation that is performed by threshold rule of YCbCr color space. The skin ROI is employed to handle the large variability of pornographic images due to backgrounds variations. While eigenporn is employed not only to decrease the dimensionality of the input images but also to handle the large variability of pornographic images due to poses variations. The main aim of this research is to obtain good algorithm for rejection of pornographic contents, which can be developed for rejecting of accessing pornographic contents in the Internet from unexpected people like children. The experimental results show that the proposed rejection system is suitable concept for rejecting of accessing pornographic images, which is shown by higher accuracy and less false rejection rate than those of existing methods (increasing the accuracy by about 2.38% and decrease the FPR and computational time by about 13.20% and 0.195 seconds of those of FD based method, respectively).