Research On Logo Matching Technology For Images

Logo is a special type of images. In practical applications, logos contain some additional information, for examples, commodity, organization, team, etc. Nowadays, on account of the information explosion, the logo matching becomes a popular technique in the fields of intelligence collection, business information statistics and news collection etc.Logo matching is an important subfield of image matching. The design for logos is often simple, which has less color, texture information. Therefore, the accurate logo matching in the complex images and videos is often considered difficult.The main contributions in the thesis are as follows:First of all, for high logo matching accuracies in the images and the videos, we analyze the existing relevant works and we propose an algorithmic framework.Secondly, we have a research on the metrics of the feature extraction(1) SIFT(Scale-Invariant Feature Transform, SIFT),(2) SURF(Speeded-Up Robust Features, SURF). In order to choose the suitable feature, we compare the different approaches in both theoretical and experimental analyses.Thirdly, we proposed a matching method which based on connections among feature points. The method of first-matching and Random Sample Consensus(RANSAC) are both introduced. Due to most of current matching methods think over the relationship between the sample features and the target points in isolation, we add the constraint that is the connections between the points of the same point set into the cost function. Then we convert the cost function to a linear problem. We relax the function of the linear problem, so that we can solve the linear problem possibly.Finally, a large number of experiments and analysis are made. We use both standard test set and self-built data set. We demonstrate the applicability and validity of chosen features. Comparison with RANSAC is made to verify the efficiency and robustness of our method.