Content-based Image Filtering

Along with the wide band unceasing popularization and the increased speed, multimedia technologies' unceasing development, multimedia information in the network assume the detonation-like growth. Nowadays the network is flooded with massive pornographic content, that has brought a very tremendous influence and harm for people's daily life. Therefore, the filtration through technological means on information in the Internet especially from the overseas is already imminent.Skin color detection is the first step to filter pornographic images. This paper proposes a new method of skin detection. It first estimated the color pixel density then obtained the maximum number density of points using the mean shift algorithm, and then made skin color region growing. The method is robust of illumination, and it can effectively avoid the interference of the background color. For color images, this paper compares the three types of skin texture detection algorithm: Gabor filtering, gray co-occurrence matrix method and the simple gray level statistics. Gabor filtering and gray level co-occurrence matrix methods need a volume of time to be calculated. They are not suitable for real-time applications; but simple gray-scale statistical computing both simple and yet effective presentation of the features of pornographic images. This paper presents a calculation of both beneficial and can effectively express the Grayscale color image feature vector. At the same time, the Hu invariant moments, and moment invariants on the edge of the connectivity features are presented. Then the connected domain computational method is given. After that a comparative analysis of BP neural network and support vector machine's classification performance is proposed, with the result of that SVM is better than BP neural network to classify pornographic images. For SVM, in this paper, a kernel function selection and parameter adjustment algorithm are presented. It can get optimal parameter adjustment in a given training set. At last, we make a conclusion to our recovery research, including the achievement and the defeat in our work.