The Homologous Videos Detection On Internet Public Opinion

As the Internet develops rapidly in the recent years, people communicate information via networks in the type of images, audios and videos is becoming more and more popular. This kind of information set is defined as Internet Public Opinion, which assembles people's perceive, attitude, emotion and behavior towards hot topics on the Internet. Moreover, this Internet Public Opinion has an enhanced-mapping relationship with people in the real world,that is to say, people's participation in a large scale may affect the events and on the other hand, these hot topics can influence people's attitude. Therefore, it is very important to accurately analyze and effectively forecast the spreading roots of the sensitive or harmful multimedia information on the Internet. However, most of the current researches focus on texts, very few people pay attention to audios or videos. In this regard, a method for detecting homologous videos is proposed.Taking the detection complexity and difficulty into account, a method based on content instead of watermarks is proposed. This proposed method selects key-frames from original videos and the pre-processing procedure is a breakthrough. Then local visual features are extracted for frames. By comparing the similarity of the frames, homologous videos can be matched. The method makes full use of visual cues of the frames and is easy to be carried out.Frame-based method talked above neglect temporal cues, which is the fundamental difference between videos and images. Considering these limitations, an alternative homologous videos detection system is proposed. The novelty lies in the features extracted are spatial as well as temporal.Having been pre-processed, the input videos are cut into segments and then extracted spatio-temporal features. These features are more discriminative and robust. In addition, the computation cost is greatly reduced.Finally, a complete homologous video detection system based on spatio-temporal features is designed, and simulation results have shown effectiveness of this system.Moreover, an interface is given to vividly show the detection procedure.