Research On Pornographic Images Detection Method With Skin-like Color Backgrounds

With the rapid development of Internet technology and its application, the pornographicinformation spreads via intricate medias at an unprecedented speed, which is harmful topeople especially teenagers. To improve the supervision of network information, andensure the information on the Internet is healthy, a fast and accurate pornographicimages detection method is urgently needed.This paper focuses on the study of sensitive image detection technology. The mainwork and results are as follows.As we all know, skin-color detection is the key to sensitive image recognition. Weturn to study on the properties of skin-color, color space and color model, and severalcommonly used color model is also analyzed in the paper. To solve the skin-color errorsproblems with skin-like backgrounds, a second skin-color detection algorithm isproposed in the paper which is based on the consistency of human body skin-colordistribution and the study of how to feedback the face information to the skin-colordetection. The first skin-color detection is achieved by using YCgCr space skin-colorinformation and Gaussian Distribution Model. Then significant level is adopted todescribe the similarity between face skin-color and skin-color, and the secondskin-color detection is realized. The ellipse model that describes the areas human bodymay occur which based on three reasonable assumptions is used to achieve significantlevel dynamic selection, thereby achieving the second skin-color detection algorithm.By using the model whether there is error detection in the first skin-color detection canbe defined. Experimental results show that the second-skin detection algorithm whichhas highly accuracy can effectively reduce the error detection of skin-like colorbackground, and can be used in sensitive image detection.Face detection is not only used to improve the performance of skin-color detection,but also to calculate the skin-color areas for sensitive image detection. After analyzingseveral common methods of face detection, a combination of YCgCr space colorinformation and AdaBoost face detection algorithm is adopted, which is verified highlyaccurate and fast by plenty of experiments.To achieve fast and accurate detection of the sensitive images, a novel approachwhich combines template matching with the second skin-color detection and faceinformation is proposed. The use of template matching can improve the speed of sensitive image detection, and the second skin-color detection can guarantee the rate ofsensitive image detection rate further. Experiments show that the method is particularlysuitable for the detection of pornographic images with face and skin-like colorbackgrounds.