Research On Video Commercial Detection Algorithm

Nowadays,multimedia data has permeated into our daily-life.It is a challenging task to get the information we want effectively in the face of such enormous multimedia information.In order to achieve this objective,Content-Based Multimedia Information Retrieval(CMIR) has been studied as a new integrated application referring to many subjects and theories.Now commercial video information plays a more and more important role in transfers of commerce information.However, the technology of advertisement detection still hangs behind comparatively.The reason is that it is difficult for computers to understand the semantic meaning of advertisements.On the one hand,the producing and expressing skills are complicated and diverse without uniform rules.On the other hand,it is hard to distinct the periods of video characteristics because the duration of advertisements is too short.So it's difficult to detect.Considering the characteristics and structures of videos in commercial programs,this paper presents the robust video commercial detection system to detect commercial videos.And the research can be summarized as follows:(1) Unknown repeated video sequence detection.A fixed size window is adopted to segment the videos into a series of video units represented by frames.Each frame is divided into MxN(8x8 ) equal size blocks,from which features are extracted to be analyzed in terms of self-correlation or cross-correlation based on the features.Two detectors in cascade structure are employed to achieve fast and accurate detection,by which not only long repeated video sequences,but also very shot ones(＜1 s) can be detected,while the overall accuracy remains high.(2) We introduce a new indexing method for approximate nearest neighbor with dependence on the data size even for high-dimensional data.Instead of using space partitioning,it relies on a new method called locality sensitive hashing(LSH).The key idea is to hash the points using several hash functions so as to ensure that,for each function,the probability of collision is much higher for objects which are close to each other than for those which are far apart.Then,one can determine near neighbors by hashing the query point and retrieving elements stored in buckets containing that point.