The Research And Application Of Network Video Pornographic Information Detection Technology

The generation of Internet and multimedia technology has brought great progress and development to the society.The network has become a hotbed of the spread of unhealthy contents,especially the study of blocking pornographic information is the current hot and difficult.Presently,the two ways to network purification are that blocking pornographic websites and sensitive vocabulary,the operating efficiency is low,the application is more limited.This text set about from porn video itself,using the pattern recognition technology to automatically identify of pornographic content,finally design and implement a high efficiency and low error rate detection rate of Internet porn video detection system,around this target,the main content of this paper and the research results are as follows:(1)This article first proposed the idea to cascade test,the video decoding into a static image,step by step to detect the faces and skin color area,and then combined with texture feature with color characteristics of female sensitive parts sensitive parts of the trained classifier for porn video detection.(2)Aiming at certainly exist faces and large area of bare color of skin in pornographic image frames,first of all to extract key frames for face and skin color detection.In the face detection algorithm,the AdaBoost model performs best on the comprehensive performance of speed and accuracy,and its accuracy and speed are in accordance with the requirements of this system.A large number of experiments prove that human skin color in YCbCr space has a better clustering,have the characteristics of luminance and chrominance separation at the same time,and skin color pixels in CbCr plane is relatively concentrated distribution in an approximate ellipse area,so this article select YCbCr color space and elliptical skin model for skin detection.(3)Aiming at the characters in many video wearing are exposed but not a pornographic video,such as lingerie catwalk shows,in order to eliminate this part of the video,In this paper,we study and design a connected domain extraction method based on skin color.First to find the face center coordinate points,connected domain of color tag,by whether a human face in front of the largest connected domain and the relationship between the two of the largest connected domain to determine whether to need to be tested to sensitive areas.In this paper,the connected domain extraction method is effective and feasible.(4)In view of the present a pornographic video detection of the problem of low efficiency,I study and design the sensitive parts detection method based on AdaBoost learning.Against women sensitive areas:the bare chest,using improved LBP operator according to the texture characteristics of the sensitive parts of the female,using the sensitive parts of the trained classifier AdaBoost learning algorithm for detection of sex scenes.Experiments show that using the sensitive parts of the classifier for porn video detection is effective and feasible.(5)Designing and implementing Internet porn video detection system.The system mainly includes face detection,skin color detection,connected domain detection,sensitive area detection,database management in five important database management modules,finally completed the system.Actual operation results show that the design of network pornographic video detection system running effect is good,compared with the traditional artificial greatly save labor and time and have higher accuracy.