An Erotic Image Detection Method Based On Integration Of Multiple Classifiers

With the continuous development of information technology, the Internet has become one of the most important and most convenient tool for information dissemination.However,when people are enjoying the convenience that network bring to life,the unhealthy and violence information spread in the networks will affect them.In particular the erotic image, bring serious damage to health of body and mind development of the majority of Internet users, especially youth.Thus,in recent years,more and more domestic and foreign research workers began to join the research in this area.Meanwhile,the state also increased investment in this area.In this context,research work of erotic image has made a lot of results.But there are also some shortcomings,in this paper,according to some shortcomings of existed technology,proposed an erotic image detection method based on integration of multiple classifier,and through the experiment,achieved good detection effect.Now the main research work carried out in this paper are listed below:1.skin color is the most typical features of erotic images, This paper studies deeply the existing color space and color model,and according to their respective advantages and applicability,finally selected the YCbCr color space under the single Gaussian skin color model to conduct the research work of complexion.2.In this paper,proposed a face detection method combined with Hough eliptic transformation,this method avoided the multi-scale zoom of traditional Adaboost algorithm,and the operation of search child window to be examined in the entire image area,improved the efficiency and accuracy of face detection.3.Using support vector machine (SVM) algorithm training features that were extracted by skin color detection and face detection. This method adds the face information in erotic image to the classifier,makes the feature of face information using more fully.4.The local key parts of the human body is another typical feature in erotic image,this paper uses the extension Haar feature to describe the key parts of female breasts,and uses the Adaboost algorithm to train the rectangle feature,generated the breast detection classifier.This makes the information of local key parts of human body has been fully utilized.5.Using multi-classifier integration technology to integrate the above SVM classifier and Adaboost classifier.Experiments show that the method makes full use of the color information,face information and local key parts information in erotic image,the recognition of the erotic image is more accurate.