Application Of Face Features In Video-Based Skin Detection

With the development of network technique, Internet already becomes one ofthe main sources which we obtain knowledge and information. While the contentsof the Internet are more and more abundant, eroticism reaction among them areflooding. These websites not only affect us through the Internet normally but alsobring the teenagers very bad influence. So detecting information of this kind shouldbe paid more and more attention. And the algorithms should be efficiency withspeed fast in real-time transport. But the algorithms of the filter software are basedon IP address filter or the judge on the web page text content lag obviously. Onlycontent-based analysis and comprehension can radically solve the problem thatnet-security technology has short ability of image information filter andsupervision.Content-based image retrieval is a technical to retrieval media according to itscontent feature. The goal of this research work is to provide algorithms that canautomatically recognize or understand important feature in an image. By now ,content based image retrieval(CBIA) is mainly implemented based on primitiveimage features, such as color ,shape, texture and so on .Considering that there aremore bare skin in eristic image compared normal image, and the skin regions whichare centralized and coherent occupy more proportion in the whole image. So wecan conclude that skin detection is the keystone in Erotic Image Filtering.Because skin-color of different race (yellow race, white race, and black race)is different in color space, it is hard to detect out skin field in a nutshell, especiallythe image content some region which have skin-color but not skin such as desertand yellow flower etc. So mistake is easy to make. The paper takes this as thebackground, relies on Science and Technology Planning Projects of Zhuhai in 2004–Research on Content-Based Erotic Image Filtering technique and itsApplication in IE, researching the key technique of content-based erotic videosequence. In traditional skin detection schemes, very often a static skin color modelis learned and each image pixel is checked whether or not its color value satisfiesthe learned model. Skin color varies greatly with the different human races. Tomake things worse, skin color, as measured by camera, can change whenillumination condition changes. Therefore, skin detection that uses a static skincolor model is certain to fail in unconstrained imaging conditions. Considering thefact that the face and body of a person always share same colors, color distributionof face regions can provide useful cues to detect skin regions of other body parts,so we propose a novel adaptive skin detection method based on the result of facedetection. Once face is detected, color distribution in face region is used as usefulcontext information for on-line skin color model building. The proposed method isrobust to imaging conditions and not biased by human ethnicity. Three parts areincluded in the paper as follows:1).Tracking and orientation moving object, whose aim is to distinguish themoving body from background in the video. The effective division of body'smoving region is important to later disposal such as object labeling, tracking andbehavior comprehension. The paper use inter-frame difference algorithms to realizeit. Firstly, background is distilled, and then , the motive information of video imagebased on the inter-frame difference is used to obtain the binary image, at last, weget rid of the noise by morphological algorithms such as dilate and erode etc.2). Distilling face feature. In this part, we give out the description andcomment standard, and select P.viola algorithms by comparing precision and speedof several face detection algorithms representative. The method make use of theIntegral Image to calculate the rectangle features which will compose a strongclassifier after the AdaBoost training .Because the rapid calculating of the integralimage ,the simple-to-complex strategy and the cascade structure of theclassifier ,the classifying speed increase greatly .We implement the face detectingwith the Intel corporation's OpenCV image processing library . On the base ofrotated face detection we calculate the average and the variance of quantum in theYCrCb color space as the face skin features . This process eliminates thecircumstance there are moving object but not human body. At the same time, wecatch skin feature correlative, and solve the race disturb with skin detection result.3).Detecting skin region. It is divided into two steps: skin-color detection andskin-texture detection. In this part, we labor the feature of skin-color, color spaceand three methods of skin-color detection in common use presently in the researchfield: Statistical Color Model;Gaussian Mixture Model;Chroma Space Model.Tests prove that the chroma space model which the paper uses is of fine stableadaptive, with a better effect than the other algorithms. After the analysis andcomparison of the texture usual algorithms, our system introduce the one-rank-graystat considering the result and speed . The effect of this system experiencespreferably.Because time and ability of the author is limited, there is lots of work to do todetect eristic video in the web. In the end, the paper give the develop direction oflater work and the expectation in future.