Negative content filtering for video application

Abstract

The research on video has been developing along with the research of the digital image processing and technological advances. The development of the internet has led to the increased production of negative images and video content. There are many challenges faced in creating filtering systems for negative content, especially on video. Most researches on the negative content filtering have been based on the skin segmentation of an image. In this research, the method of negative content detection on video (porn video) based on the skin segmentation in the video composer frame was developed. By combining the two color spaces namely RGB and YCbCr color space as the skin detection algorithms to improve the accuracy of the class determination on the video. The sampling method used in this research was the 8 bytes keyframe extraction or about 256 frames with a certain distance based on the total video frame. Based on the number of frames extracted, the porn percentage value was calculated. The data were 102 videos with duration ranging between 2–15 minutes each. The datasets were divided into 60 data, namely 30 porn and 30 nonporn video. The other 42 data were used for accuracy testing. The determination of classes was limited by a porn percentage threshold value (pornographic %). Based on the research, from the result yielded, the video threshold was classified into porn class when the value of porn percentage threshold > 70, and nonporn class when the value of porn percentage threshold < 70. The result of 42 video tests showed the accuracy of about 90.5 %.