Research And Design Of Image Filtering System Based On Content

Erotic image filtering is an important aspect in the information filtering technique which is gradually becoming an important research topic of the network information security, and it's significant in keeping cyberspace clean and protecting teenager's mind. The problem of pornographic images is actually a classification of images. In this paper, we carried on research on content-based image filtering system and adopted statistical classification method to resolve the problem of erotic image filtering, including key filtering technique, key filtering method, the design and realization of the filtering system. The key technique include skin color detection, skin texture analysis, object area segmentation, image features extraction and the design of classifier.The detection of skin color is important in designing the system of image filtering, and is the foundation of the extraction of features, which is used in statistical classification method. An elliptical area based skin color detection method is proposed after we analyzed the human skin color characteristic, color space choosing, compare of skin color models and the processing of high light and shadow area Experiments show that elliptical area based skin model can effectively detect the skin color in image. According to the characteristic of the false skin color area and skin color area, we propose transformation of K-L based feature extraction of statistical texture features and simple statistical texture feature based skin area detection method. Experiments show that texture analysis technique can effective remove the false skin color area and keep the true skin color area. We also describe the application of the skin color detection in erotic image filtering.Feature extraction is the foundation of statistical classification. We apply various features to erotic image filtering, and describe the method the mask-based feature extraction, including skin color percentage, biggest connected area percentage and other features related to connected area. Object area usually contains more information. We propose a method of object area segmentation, which is voting for the interesting area based on skin information and dynamic sub blocks, and extract the feature in the objectarea to advantage in erotic image filtering.The classifier is the key of the designing of the statistical classification method. The erotic filtering problem is a small sample problem. Support Vector Machines (SVMs) have the ability of study and memory, and have advantage in resolving the small sample problem, non-linear, high dimension pattern recognition and classification problem, so we introduce the principles of SVMs, construction of the feature vector, choice of kernel function and procedure of training, and apply it to erotic image filtering system. Results show that the method of SVMs-based erotic image can attain 80% positive rate.Finally, we designed an erotic image scanning system adopting the thought that many connecting weak classifiers can acquire a strong classifier, which is used to test the method of filtering and the key technique, and is an experimental frame system.