Research Design And Implement Of The Objectionable Video Detection System

Because of the rapid development of Internet and multi-media technology, the texts,images, audio files on networks are grows exponentially, a variety of pornographyinformation is also flooded with Internet, In allusion to detection and filter thesepornographic information, various software which has its own technology merits.According to the pornography videos on networks, this paper founded by the NationalScience and technology major project “Research in mobile Internet network and thesecurity of information technology(2011ZX03003-005-02)” of NanJing next generationnetwork technology application engineering research center of IACAS(Institute ofAcoustics, Chinese Academy of Sciences), proposed an method and the correspondingalgorithm of detecting pornography video, designed the corresponding detection system,and has carried on construction and testing.The author concerns the detecting skills in two different aspects and does the deepresearch, on the one hand, detect by identifying the name of vide files, because the filename largely reflects the details of the videos. Through participle of The video filesname and other pre-processing, use the fuzzy genetic algorithm to optimize the sensitivewords in the samples, updated and feedback them, then analyzes the weights in thesensitive word-stock of the sensitive words that contain in the file name. finally,recognition videos by calculate the total weights of the file name; on the other hand,from the perspective of analyzing contents of the videos, extraction the key frames fromvideos and identifies them. In the dissertation, the author stand on the existing matureconditions of sensitive image recognition based on the skin colors detection andidentifying the key part of human bodies, applies the new theory of compressed sensing,which is developing in the few decades, in the identification of the key frames. First ofall, sparse representation of the key frames by extracted, Then figures out the sparsecoefficient solution by l1norm algorithm. next, solving the residual of the obtainedcoefficient. Then it can easy to classifying images through residual. Thus, achieved theidentification purpose. Finally, this paper combined two aspects discussed above todetect and identify porn videos. The experimental results show that the accuracy of theproposed method reaches86.6%, meanwhile, the false detecting rate and missing rate are decrease enormously.