Abstract: This paper addresses the problem of detecting objectionable videos. An optical flow based key-frame selection algorithm is presented. It makes use of the key-frame and frames near it to select the foreground. With the view of bad illumination, a new gamma correction method before the skin detection is introduced. An effective determining rule is given. The experimental results show that the method’s correct detection is up to 80% and it is fit for the video detection environment.

Key words: Optical flow, Key-frame, Gamma correction

摘要： 提出了一种适用于流媒体中的不良视频检测算法，提出了4点约束光流算法，利用局部光流信息提取关键帧，对关键帧中的前景部分进行肤色检测，考虑了光照不足的情况，在肤色检测前加入伽马矫正，提出了一种有效的判决准则，实现整段视频的判别。实验表明，该算法识别率达到80%，鲁棒性较好，适用于视频检测环境。

关键词: 光流, 关键帧, 伽马矫正

CLC Number: 

• TP37