Video Fingerprint Scheme Of Fusion Bimodal Based On AdaBoost Research

The acceleration of network and development of webpage technology provided wide space for on-line videos, which make it possible for general internet users to upload, download and watch on-line videos easily. But many problems arises during the rapid development of on-line videos, such as more piratical videos, more illegal videos and lots of junk videos which uploaded by users and etc. These factors have adverse effects on the development of on-line videos. At first, the legitimate rights of owners of commercial videos can’t be guaranteed, which also blow the enthusiasm of the producers; Secondly, illegal videos exert a passive and negative influence on internet users. Meanwhile, the junk videos not only bring greater demand for hardware storage and vulgar labels to the servicer of on-line videos but also waste users’time. Therefore, it’s necessary to manage the content of on-line videos effectively, filter out the piratical videos, illegal videos and junk videos and provide a good and sustainable development environment for online videos. For that content-based video identification is a good approach to solve these problems and the video fingerprint fusing multiple features can identification videos more accurately and robustly, this paper focuses on systematic study of the fusion multi features fingerprint technology.First, this paper adopts improved algorithm of AdaBoost to fuse double-modal video fingerprint, its advantage is that it can get the weights of SIFT features,motion features and audio features by adaptive training of sample data, then identify video fingerprint by the weights. Experiment shows that the approaches of this paper proposed can get higher accuracy than others. Moreover, it can be more robust under the attack of noise, change of scale and brightness.Second, this paper according to the character of video fingerprint unite have high effective on search of the LSH algorithm design a kind of search-match method of video database. Firstly, this method use audio fingerprints the front153frames data in the database of videos to set up a LSH hash table by algorithm of LSH for searching; then we can get a small candidate sets by use continuous153frames audio fingerprint data search in query stage; finally fusion the three features and match all of the small candidate sets and we can gain the result. Experiment shows that compared to exhaustion search method this approach of search can improve the speed of video search.