Research On Content-Based Erotic Image Filtering Technique And Its Application In IE

With the development of network technique, internet enters into our life.Itbecomes the necessary source from which we obtain knowledge and findinformation. Although the contents of the Internet are more and more abundant,because of huge and hard management of its contents, among them eroticism,violence, reaction are flooding. These websites not only affect us using the Internetnormally but also bring the teenagers very bad influence. Many countrys takemeature to purify the internet environment. An activity of beating eroticinformation had been taking science July 2004. Obviously, forbidding theteenagers to use the Internet is not wise.So in addition to the tipstaff striking eroticwebsites, we should filtrate the erotic information by filter software .Only do thiswe can use the information of internet freedly.But the algorithms of the filter software are based on IP address filter or webpage text content judgement. The former shields information by IP addressmatching, it needs to update the IP address database by collecting IP addressestermly, so it lags evidently. The later may neglect some sites and shield some goodsites because of the limitation of text content. Only content-based analysis andcomprehension can radically solve the problem that net-security technology hasshort ability of image information filter and supervision. Currently some domesticand international scholars research the Content-Based Image filtering techniqueand obtain elementary fruit, but a lot of problems have not been resolved.According to the above analysis, we can see that we can not forbid the eroticinformation to transmit if only adopt one filter technique.so the paper takes this asthe background, relies on Science and Technology Planning Projects of Zhuhai in2004 –Research on Content-Based Erotic Image Filtering technique and itsApplication in IE, researchs the key technique of erotic information filtering basedon image detection.The paper detects erotic information by CBIF, URL filteringand words filtering , designs filter model, reasearches skin algorism and realizes thefilter in IE.Erotic images are characteristic of bareness skin, so we use skin detectingmodels, texture models to detect skin-color area and build binary iamge then distillcharacter vector, finally use corresponding classing algorithm to filtrate images.The skin-color detection seems simple but complicated mainly for theinfluence of the factors such as race, illumination, noise and so on. At present thereare three methods of skin-color detection in common use in the research field:Statistical Color Model; Gaussian Mixture Model; Chroma Space Model. The firstand second adopt the RGB color space, and the third uses the combination of colorspace of YUV and YIQ. Tests prove that the chroma space model is of fine stableadaptive, with a better effect than the other algorithms. While the present colordetection has deficiency in three aspects: 1.detect the non-skin-color area for theskin-color area in the image, such as the flat yellow wooden-board, yellow-brownhair and so on; 2.omit the skin-color area in the image; 3. badly influenced by theillumination. The former skin-color detection seldom calculates the brightness, andjust detects the skin color depending on the chroma, overlooking the differentdistribution of the skin color image in color space caused by the different race (thewhite race, the yellow race and the black race) and the different brightness. To theabove-mentioned problems the article puts forward a new method of detecting theskin-color image on the base of the existing chroma space model, that is thechroma space model based on the self-adaptation of the brightness, the approachcolligates the distributed feature in YUV and YIQ color space, and differentiatesskin color according to the brightness and angle of UV of images. It choosesdifferent scopes corresponding to different skin-color. Experiments show obvioussuperiority on yellow, black (Asian, African) and white skins (occidental). Thecorrectness of skin-color detection is improved to 93.4795%, little change of errorrate and detection time, detection result is improved obviously. The erotic images may be detected in error by the skin color model, becauseof the color's similaration (such as Yellow of sofa, yellow of woolen blanket etc.).We need adopt texture model to process the mask images based on the previousstep. Non-skin area will be eliminated, the skin area in erotic images will be exact,the just detecting ratio is increased, and the error ratio is decreased. It can providethe corresponding classing algorithms valid characteristics. There are four texturemodels: Gabor, Gray Statistic, Gray matrix and Gray variance based on area .Wefind thatthe gray stat method is the best method among these. The gray stat methodis basis in image processing. In the implementation, we use the idea ofone-rank-gray stat to judge whether a point and around part are smooth in skin area.First, we get the area stat gray value, generally it is average value or the gray valueof the pixel in the area center, second compare the pixel value in the area with thestat value to get the traits, and then judge whether it is consistent with the stat traitsof the stat skin area. Through analyzing a lot of erotic images, we extract six characters and giveevery feature a coefficient. We compute the value by a formulae and make it to beclassing eigenvectors to train the classify algorithm and filtrate erotic images. The paper makes use of the technique of BHO to carry out the erotic web pagefilter. BHO is a COM object borned by Internet Explorer automatically, when userstart a new example of Internet Explorer. We can control the interview to thewebsite or modify the contents of the web page dynamically by hanging the BHOon IE. This filter is constitute by the following modules: The URL filteringmodule(include new window filtering);The web page text filtering module;Theimage filtering module; The erotic URL database and the erotic phrases databasemaintenance module.At the time of openning the browser, the filter realized byBHO technique starts together. The filtrating steps as follows: When IE is opened, the filter based on BHO technique startups at the sametime. When navigating a new address, the filter gets the infilling URL addressfirstly, judges whether the URL is prohibited (compared with the URLs address inthe file). If it is prohibited, navigates to the blank page, else entrances the web pagetext filtering module. If there are erotic words, it navigates to the blank page andadds the URL to erotic URL database; else IE downloads resources, and processesthe images: gets the pointer of the browser, gets the image information (such asquantity), and processes images one by one. The filter searches the image locationby the image name, and processes the images. If the image is an erotic image, wemake its URL to be blank, the number of erotic image add 1. When the images aredetected completely, IE navigates to the new address. If the number rate of eroticimages reaches a threshold, user can choose whether he add the URL to erotic URLdatabase. Experiments and analysis show that the system has good recognizing effect,by and large realizes to supervise and analyse images online with little infection onthe net normal operation, But many places of system need to be improved andperfected, such as we'll add face and apparatus recognition and detect skin-colorregion in image according to skin-color of face to increase right rate and canprevent from regarding a photograph of a face as erotic images. We should add andperfect modules of filter and increase its realtime capability.