Automatic Extraction And Recognition Of Logo

Logo automatic extraction and recognition in such fields of content retrieval, dissemination and security of broadcast television video has widely applications. Television logo extraction is a hot technology process to locate the logo region and extract the logo using some of their inherent properties, that is separate the logo and background in the image, extract the interest logo from the background. Logo recognition is a real-time procedure to determine whether the presence of the logo based on the extracted logo template in the first step. This paper has using some identification algorithms do logo recognition. The whole issue involving the technical fields of pattern recognition and image processing, in the foreseeable future the real systems can use it.The main work of this paper include the following aspects:(1)Design Logo extraction and logo recognition algorithms overall. Completed the entire algorithm design work, develop a technology roadmap and identify critical and key technologies.(2)Detect and extract logo based on weighted video frames. With a known knowledge of fixed logo position in the television video, the subject determines the sample logo region and extract logo based on weighted video frames using edge information. The results of this experimental show that this method makes the extracted logo images more complete and clearer.(3)Recognize television logo based on the color space and SURF feature matching. This approach includes coarse matching phase using color space information and precise identification phase by SURF feature matching. Coarse recognition stage is to use the HSV color space histogram features to determine the candidate logo set, precise identification phase is using SURF feature match to identify logo in the candidate set. The result of this method has a general rate, between 70% and 80%.(4) Recognition television logo based on HOG feature and SVM classifier method. After the establishment of the station sample logo database, conduct sample training using HOG feature, then take a video frame which is to be identified in the SVM classifier by HOG feature for classification. The results of the experiment show that compared to the feature match algorithm, this method is slightly improved rate of recognition, about increased by 5%.(5) Recognition television logo based on pixel by pixel feature and BP neural network method. After image’s pretreatment, extract the feature of image pixel by pixel, and then as sample inputs for BP network training, video which to be identified is putted into the trained specific neural network to get output. The result of this method is faster, better, but there are limits on applying, however, limitation is that the image size and ratio can not be changed, so do the image scale.Thinking all the experimental results, our proposed methods can extract the logo full and clearly, and recognition rate can be accepted, but there is a gap between experimental program and practical application.