The Application Of Body-feature Based Skin-detecting Algorithm In Erotic Image Filtering

Building up content-based image filtering on the Internet is a crucial issuebecause of the recent explosion in the amount of online images .But the usualfiltering methods are only based on the IP address and the web text content .In thispaper, a method for detecting skin color which is based on human face and bodypositional information have been proposed to solve the problems .Concretely, themain contributes of the paper are:1)We review the development of face detecting technology .In the view ofveracity and velocity, we select Viola's rapid face detecting algorithm which isbased on the Integral Image and the AdaBoost training .The method make useof the Integral Image to calculate the rectangle features which will compose astrong classifier after the AdaBoost training .Because the rapid calculating ofthe integral image ,the simple-to-complex strategy and the cascade structure ofthe classifier ,the classifying speed increase greatly .We implement the facedetecting with the Intel corporation's OpenCV image processing library .Inorder to successfully detect rotated faces ,each input image is passed to facedetector three times .The first time is the original image ,the next two times areits rotated variants with rotation angle of 45 degrees anticlockwise andclockwise respectively .On the base of rotated face detection we calculate theaverage and the variance of quantum in the YCrCb color space as the face skinfeatures .Then we employ the thinking of ellipse fitting to establish a humanbody model with a series of mathematic formulas .As a result, we achieve togetting the body approximate location .Because there is large naked skin area inerotic image ,the body here means the largest part of exposed skin area in thisalgorithm .2)During the summary of classic skin algorithm ,we analyze the character of theskin as well as the color space .Comparison has been made among several skincolor models .The skin color models include Statistical Color Model;GaussianMixture Model;Chroma Space Model .The classic skin algorithm also use thetexture detecting for the filtration .After the analysis and comparison of thetexture usual algorithm ,our system introduce the one-rank-gray statconsidering the result and speed .In the classic skin detection we only use thecolor information and ignore the importance of the pixel positioninformation .To our common sense ,even the same color in differentdistribution maybe present different object .And the skin pixel appear more inthe center area and around the object than other area .Aiming at the deficiencyof the traditional skin color algorithm, we propose the dynamic thresholdmethod based on the pixel positional information using the body approximatelocating algorithm in the Chapter 2 .Firstly we define the absolute distance andrelative distance to measure the pixel position .Here the absolute distancemeans the distance from the pixel to the center of the image and the relativedistance means the minimum of the distance from the pixel to the center of theface ellipse and the body ellipse .Secondly we take advantage of the Bayestheory to establish the relation between the pixel distance and the skinprobability .The skin probability of the pixel will be larger when its relativedistance is more little .Then the skin threshold can be dynamically determinedaccording to the skin probability and the face skin feature. In the comparison ofstatic algorithm, our algorithm' advantage is that different pixel in differentimage has different skin threshold .After having a experiment on 814 images,we find that the method can wipe off the skin-like color background effetely.3)Erotic image filtering is a non-trivial and challenging task .This paper does notconcentrate on the essence of filter, but focus on offering effective classifyingfeatures to the filter. Although there are many sophisticated classificationmethods such as Support Vector Machine (SVM) and Artificial NeuralNetwork (ANN) ,we opt for using decision tree for its simplicity and itsparticular efficiency .For one thing we summary the usual features of eroticimage .For another, the body model features are proposed after the facedetection and body approximate location .We hope these body features andimproved skin features can wide off the skin-like background effectively .Forthe sake of the experiments, we use a library including 1714 images .From theresult, we find that the passive rate decrease while the positive rates varies little.The reasons for the result include two points: one is that face detection canoffer good skin information;the other is that the face and body positionalinformation directly increases the information inputted the algorithm.4)We introduce the technology of dynamic link library .It has the advantage ofsaving memory, easy upgrading and good service .According to the character ofthe program, we pack the correlative algorithms in the DLL documents .Thealgorithm of the face detection and the body location are packed in theModel.dll, and the dynamic skin threshold are packed in the Distance.dll..