A new trademark image retrieval algorithm based on multiple features fusion.

Abstract

The existing single feature based methods cannot effectively retrieve trademark images. Therefore, in this paper, we propose a new trademark image retrieval algorithm using fused features to solve the problem, which combines shape features, improved SIFT features, and color features. Our idea is to first extract the local shape features of the input trademark image, and then use the improved SIFT algorithm to detect local key features. After that, the dominant color descriptor is utilized to describe the global color features of the input image. Based on the above extracted different sorts of features, a fused feature vector with different weights describing the overall information of the input trademark image is derived. Finally, the cosine similarity is employed to compute the similarity between the searched trademark and the databases. To evaluate the performance of the proposed algorithm, we have conducted a set of query experiments on a data set containing more than 12K trademark images. Experimental results show that the proposed method improves the performance of trademark image retrieval with good robustness, and is significantly better than the state-of-the-art algorithms in query accuracy and efficiency. It has good practical value in the field of well-known trademark detection and trademark copyright protection.