A two-stage matching algorithm on heterogeneous platforms for detecting Advertisement Logo in videos

With the growing number of video media advertising elements, the detection for time and length of Advertisement Logo has become a new demand of video monitoring sections. In response to this demand as well as the lots of background interference problems in video frames, this paper put forward a two-stage Advertisement Logo matching approach on heterogeneous platforms in videos based on global feature. In the first stage, matching-template which represents shape feature of image is established online by self-adaptive method so that background interference in the target image can be eliminated. In the second stage, the description for texture feature based on the rotation-invariant ELBP operator is used to make up for the inadequacy of the previous method and reduce fault detection resulted of the similarity between column name or TV logo and the advertisement logo. The experimental results prove that our approach can balance retrieval performance with efficiency, which meet the requirements of Advertisement Logo detection in videos.