Context Based Similarity Matching

Through this work a new method is introduced for the similarity matching of logos. Logos appear in images of real world indoor or outdoor scenes super imposed on objects of any geometry. In most of the cases these images are subjected to perspective transformations and often corrupted, which have low resolution and quality. Based on context dependent similarity kernel that directly incorporates the spatial local features, a novel solution for logo detection and recognition is attained. Using the CDS kernel, the geometric layout of local regions can be compared across images which show contiguous and repeating local structures as often in the case of graphic logos. This solution is proved to be highly effective and responds to the requirements of similarity matching of logos in real world videos.