Research On Auto Detection Of Video Advertisements

With the deep development of market economy, video advertisements for enterprises become more and more important. It gradually comes to be a daily work of companies. To be sure that the advertisements do their work enterprises have to assign someone to watch TV all time to find when and in which TV channel the advertisements are broadcasted. At the same time, culture supervising deputies would demand TV stations to broadcast some commonweal advertisements, video bulletins and video messages according to governments. Nowadays this job usually is done by manpower. It is no doubt great waste of the resource of man power and finance.The tiding development of digitized video makes ISPs(information service provider) and TV stations rely on video classification more and more. As an independent field of video classification, advertisement videos' auto detection should give a good example to it.This paper has researched the techniques and algorithms to detect the advertisement from video and implement advertisements' auto detection by computer. Video advertisements detection includes two aspects: advertisement breaks detection and advertisements units detection. As a resolve to detect the advertisement breaks in the videos, this paper proposes an algorithm using a multi-features integrated vector slidding window. To find if an example advertisement unit is broadcasted in a channel by computers, we propose an algorithm using block-based frames matching and similar shots matching.The main jobs in this paper lie in:? It has studied the features of advertisement videos differentiated from other types of video on visual and audio levels. It also finds the low level features responding to video structural features, video object features and audio features. Then it proposes a technique framework for advertisement detecting.? It has proposed an algorithm corresponding to the demand detection of advertisement breaks' boundaries ... multi.features integration based sliding window algorithm. It has researched the algorithms to gain the features of the vector of advertisement videos: compressed field based shot boundary detection, text position to video detection and the algorithm of matching of the advertisement vectors. It also analyses the weights of the features by which the vector of advertisement breaks are composed.? It has proposed the algorithms corresponding to the demand detection of known advertisement units. The algorithm is integrated by frame matching and shot length series matching. It has researched the frame.based videosegment matching algorithm and the shot.based video segments matchingalgorithm. ? It has designed and implemented a system to contain the function ofadvertisement breaks detection and known advertisement units detection.Experiments show the plausibility of the algorithms.