Research On The Method Of Internet Video Classification Based On Multi-information Iteration

In the past years, information networking has greatly promoted the wide spread of multimedia, which ensures rapid information sharing as well as enriches people's life and facilitates people-to-people communications and interactions. However, there is a great deal of bad information existing on Internet in nowadays, which not only poses bad effects on the physical and psychological growth of youths, but also increases unstable factors of society.Deepen and wide researches about the distinguishing of online violent videos are carried out in this paper. In current methods, violent elements in those videos are bases on single visual features or audio features. While these methods ignore the key role that the online environments of the violent videos play. Methods of multi-features for violent videos sorting are put forward by using context information. The visual and audio features, are taken into consideration, the textual features and online context of the video can be further extracted. So that the three features can be effectively combined. Main works in this paper are as follows:1. We propose a pre-sorting method of online videos bases on textual features and voice features. We use the context information of the webs which conclude the videos. The videos samples are sorted into the drafted kinds of videos on the basis of zero-crossing rate, pitch frequency and so on of the audio.2. We propose a method for distinguishing online violent videos on the basis of visual moving features. The motion intensity and the motion angle combined, so that the motion conditions in the videos can be described. We have tested the method through the performance comparison in the primary data base and pre-sorting data base.3. The comparison experiment results shows that the right detection rate is significantly enhanced after the pre-sorting and sorting of videos on the basis of visual information, which indicates the multi-features method proposed in this paper is more effective in online video sorting.