Abstract: In this study, we propose the new pornographic video detection scheme using multimodal feature such as image features of each frame using deep learning architecture, image descriptor features of the frame sequence, motion features using optical flow and audio features extracted from video. By using these various features at once we can detect almost all pornographic events without being confused by a specific element of input video. And as the performance evaluation results we can obtain 100% true positive rate and 67.6% overall accuracy. Although, the overall accuracy is little bit low due to high false positive rate we could successfully detect the pornographic videos which are difficult to detect by using only single modal features.
Kwang Ho Song and Yoo-Sung Kim, 2018. Pornographic Video Detection Scheme Using Multimodal Features. Journal of Engineering and Applied Sciences, 13: 1174-1182. Asian Journal of Information Technology, 18: 250-260.