In recent years, a great deal of research work has been devoted to the problem of ltering and blocking adult content from Web indexes and browsers. By taking advantage of the fact that there is a substantial correlation between images with large patches of skin and adult images, the skin color detection can o er an e ective and ecient way to detect the adult contents in color images. In this thesis, a study of skin detection algorithms based on skin color is presented. Four color spaces, RGB, Normalized RGB, YCrCb, and HSI in addition to RGB channels ratio are used to represent the skin color. Many algorithms based on one of the above color spaces are previously proposed. In this thesis, these color spaces have been used to get a new skin detection algorithm, which gives higher accuracy in recognizing skin regions. The proposed skin detection algorithm is a rst step in an approach for pornography detection in color images, but not limited in any way to this goal.
In this thesis, we introduce a system for automatically detecting adult contents in image. The pivotal idea of the proposed system is that the skin-colored pixels in a given image are rst marked based on our skin detection algorithm. Then, the skin-colored regions are fed to a specialized geometric analyzer, which attempts to nd a human gure using geometrical constraints on human body structure. If the geometric analyzer nds a suciently individual human parts, the system decides the presence of pornography. To improve the performance of the system, some features based on the output of the skin detector are additionally computed. The features added based on informal observations that adult images are often sized to frame a standing or reclining gure. The system is shown to be e ective on a wide range of shades and skin colors. The proposed system can be used as a part of a web search engine or as an add-in of a browser. Experimental results show that the system is capable of detecting adult contents with an accuracy of 72.2% with 1.7% false alarm.