A Layered Approach for Trademark Image Retrieval using Texture and SURF features
Trademark images are the copyright of any individual or commercial organizations. To protect the copyright, retrieval of Trademark image is required which retrieves similar registered trademarks. This paper proposes a method which extract the entropy of the trademark images as a first layer to remove least similar images and then global shape and local texture feature extractor have been applied to retrieve most similar images of the query image. Zernike moments has been used as shape descriptor and Speeded-Up Robust Feature (SURF) as texture feature which is based on key-points of the trademark image. Weighted average is computed of both the local and global feature then, Euclidean distance is applied to retrieve the rank1, rank5, rank10, rank15 and rank20 most similar images. Experiments have been carried out on proposed dataset and found that proposed approach works better and improves the accuracy.