Human Body Region Detection Based On The Features Of Contour And HOG In Pornographic Images

The progress of science and the development of the technology to made the concept of "personal computer" already includes desktops, laptops, smartphones, tablets and other personal computing devices. At the same time, the rapid progress and popularization of the Internet and mobile Internet, made the information out of the limitation of physical distance. This has brought great convenience to people’s life. When we feel this from computer and network brings, at the same time, we can also see there are too many bad network which are contains violence, pornography and other bad cultural information in the Internet. Although these information are virtual, but people cannot ignore them for the children and minors’ physical and mental health brings negative effects in the real world. Detect and cut off these from the Internet, and build an open, healthy and positive network environment for children and young Internet users, that is the problem of the world’s governments of the countries, regions and organizations are focusing on. Therefore, study and research a high efficiency, automation, accurate information detection and blocking system, to purify the children and minors online environment, protect the health of their physical and mental development has important significance.In this thesis, we started from the general algorithm for human detection in static images, to achieved outstanding results in recent years in the human body detection algorithm based on HOG feature descriptor carried on deeply study and analysis thoroughly. Concrete study of the idea, process and performance of detection algorithm, and through concrete experiment requirements affecting the final test result of testing framework are analyzed in details. In the pornographic image, human body detection is proposed for human detection algorithm based on HOG feature testing framework improvement ideas, and from theory and experiment demonstrates the feasibility of the improvements.We compared the similarities and differences between the pornographic images and normal images, according to the characteristics of the specific pornographic image database, improved the detection of human body based on HOG feature. Choose to build out by contour detection are interested area as an area to be detected. The algorithm framework of based on HOG feature descriptor detection by using the testing in multiple locations and dimensions of the image for sliding window to improvement is interested in local areas to sliding frame to be detected. By improve the framework, greatly reduce eventually produce image detection window, and shorten the testing time. At the same time, according to the characteristics of the pornographic image library, adapted the window level classifier to redesign and training to the specific pornographic image database. Through the above improvements, this part which is the human region detection components can meet the requirements of the overall system for detecting time, and in general, it can correct body area monitoring and identification of pornographic images.