Trademark image retrieval algorithm based on SIFT feature

Abstract

Image matching is an important way to implement trademark image retrieval. In this paper, we proposed an algorithm based on Scale Invariant Feature Transform (SIFT) for automatic trademark image search. Firstly, feature points were detected and described through calculating the gradient histogram of nearby region. Secondly, Euclidean distances were calculated to obtain the similarity between the SIFT features extracted from the two corresponding images. Experiment illustrated that the proposed method was robust to scale, viewpoint, occlusion and noise interference. © 2012 Springer Science+Business Media B.V.