Pornography Detection By AdaBoost Algorithm Based On The New Features

Along with Chinese Internet access, network brings us convenient, but there were also some negative issues; Yellow information are flooding, such as prostitution on the Internet, naked chat, pornographic movies, pornographic pictures, pornographic words, sex videos and so on.Sex videos' harm are unprecedented, Because they contain the pictures and the sound.They poison heart health of young people Severely,affect social stability,even affect our country's future.Therefore,eliminating pornographic information has become a urgent.When giving strict legal means,it needs to add some science technology to eliminatie the pornographic information.If the application of science technology can be successfully carried ,it helps to purify the social,the network environment and reduce many social security problems.So introduce some improved features in the AdaBoost algorithm to train the pornography detector.In the training process of AdaBoost algorithm system, when selecting many features,there will be many problems,for example,calculation speed is slow , the accuracy is low, and even cause non-convergence.If few features are selected ,classifiers can be trained quickly,but the accuracy rate of detector and false detection rate of detector decreased, and it is hard to find a balance between them,so limit some applications in our lives.For these defects in traditional AdaBoost algorithm,we attempt to modify and delete some existing Harr_like Features,also increase some new features.This paper obtains the following results:1, For the detection of pornographic images ,we have introduced some new features. such as trapezoidal features, T-shaped features; breast and eye for the human body is difficult to distinguish,we add new distinctive features.Improve the weak classifiers' weights' updating mean during training progress.2, Improved Snake segmentation.in this paper ,we improve Snake algorithm based on improved AdaBoost algorithm.we take the detected the target s' location as the initial snake segmentation point.Added a new internal energy function.So that it can converge to the concave objects' edge.So that enhance the degree of intelligence and automation,increase some engineering practice Operation.