Search

Aimed at that there is often a video paragraph of human body's reciprocating motion in the pornographic videos, a novel method of pornographic videos detection is proposed. On the basis of calculating the optical flow field, we extract the characteristics points of the moving target, and classify optical flow direction of these points and get the statistics of them. Then we establish the optical flow direction's statistical histogram. According to the changes in the direction of moving target and the changes in the frequency of it, we can determine there is pornographic information or not. The experimental results show that the method is simple and convenient, fast and efficient, high in the recognition rate and well in the robustness.

This paper proposes a method of identified reciprocating motion in pornographic video from other human action using Hidden Markov Model (HMM). The motion vectors are obtained by decoding the compressed MPEG video. Then the feature vectors are extracted by calculating the direction and the magnitude of the motion vectors. The feature vectors are fed to Hidden Markov Model for training and classification of actions. Six actions were trained with distinct HMM for classification. The correct recognition result is up to 90%.