摘要
为了快速准确地清理互联网平台上的非法视频,减少人力物力的投入,在研究非法视频特征的基础上,提出一套基于台标识别的网络视频分类系统。在分析不同来源非法视频台标特性的基础上,根据不同来源的非法视频所携带的台标特性不同,系统采用基于模板匹配和特征识别相结合的方法进行台标识别,以提升识别的准确率。同时采用多种图像增强技术降低分辨率、清晰度、背景等不确定因素对识别结果的影响,提升对非法视频的召回率以及召回的准确率。实验结果表明,该系统可通过筛选出确认安全视频和疑似非法视频来有效减少非法视频筛选中的人力投入。另外,采用FFMPEG解码库解码视频抽取关键帧,提高了系统处理效率,平均单个视频处理时间缩短30%。
In order to clean up the illegal videos on the interact rapidly and accurately and reduce the input of manpower and material resources,a video classification system based on TV logo recognition has been proposed. On the basis of analyzing illegal videos TV logo with different producer,the system adopts a method combined pattern matching and feature recognition for TV logo recognition to improve accuracy according to different TV logo feature of illegal video. At the same time, a variety of image enhancement techniques are adopted to decrease the influence of uncertain factors like resolution, definition and background on the identification results, improving the recall rate of illegal video and recall accuracy. Experimental results show that this system could effectively reduce the human input and material resource by screening out the security video and suspected illegal video. In addition, FFMPEG is used to decode and draw key frame,which can improve the efficiency at 30%
出处
《计算机技术与发展》
2016年第10期87-91,共5页
Computer Technology and Development
基金
北京市科委项目(Z141100006014002)
