Detection of Pornographic Digital Images Using Support Vector Machine and Neural Network

In this paper a new approach for detecting explicit content or pornographic images is proposed. The suggested system contains three phases. In the first phase, the image is segmented into skin region and non-skin region by an ANN in a pixel-wise manner. After image segmentation, a set of discriminative features including the ratio of skin pixels to total image pixels, the number of faces in the image, the area of largest skin area are extracted. Finally, support vector machines are used to classify the image in two normal image and explicit image using the extracted feature vector. A set of 400 images is used to learn and test the system. The detection rate for classifying pixels in skin and non-skin is 91.8 and as the result, the proposed system could obtain 89.9 accuracy in classifying images based on their content. Furthermore, the true positive rate and false positive rate for the proposed system are 92% and 0.125%, respectively.