| Internet as representatives of fashion and the high-end technology has become the majorform of living work and family entertainment. With the fast and the (zhiguan),it has becomethe main method of communication and the information access. When internet got itsrecognition, we had to face the bad side of it. Because of its fast and simple way, many sadthings would accompany the information that we really need, just like when we take our foodwith bacteria. The violence, pornography and the superstition would infect our mind fasterthan the bacteria. We can imaging that when a ten years old boy open a pornography webpage, we shouldn`t proud the reflect the development of science and technology in order toprove the opening up. What we really should do is stopping these unhealthy, and making theinternet clean and clear.This will be through the identification of sensitive photo to completed the unhealthy shield.Main task is to identify the skin parts of images. First we process the photo with chromaticityspace models. The approach colligates the distributed feature in YUV and YIQ color space,and differentiates skin color according to the brightness and angle of UV of images. Then wethrough Texture detection to finish the final photo. We will compare the different skin-colordetection algorithm to find out the most power one to finish the sensitive photo detection. Weuse this algorithm to support shielding pornographic websites.In color detection model research process, mainly on color statistical model, mixed Gaussmodel and chroma space model three kinds of skin detection model is compared and analyzed,through comparison of experimental data the final adoption of chroma space model as a skincolor model.In the skin texture extraction process, mainly on the Gabor filtering, gray statistical andco-occurrence matrix and gray variance based on region four algorithm comparison andanalysis, also through comparison of experimental data revealed first-order gray statisticmethod both in effect and in time than other methods. Finally, the system using this method.Experiments and analysis show that the system has good recognizing effect, most sensitivephoto can be shielded. We choose the Chroma space model to complete our algorithm, then itproved effective. For the development of large anti-pornography software in the network, eliminate the undesirable content support, can be used to construct network image classifierand embedded webpage browser porn filter.This paper realizes the content also has shortcomings, in image filtering process, thepictures produced not ideal error. The main reason is the sensitive image filter only throughthe skin color detection and texture detection as main discriminant strategy may produce largeerror. In order to avoid the above shortcomings, in future work mainly based on identificationof the organ, to decrease the error, improve the detection performance of filter. |
