Image retrieval based on optimized visual dictionary and adaptive soft assignment

Abstract

In this work, we propose a new image retrieval scheme by identifying better visual representations and fusing multiple similarities based on multiple features. For visual representation, we propose a new coarse-to-fine visual dictionary construction method based on the bag-of-features model. An adaptive soft assignment technique is developed to assign one local descriptor to several nearest visual words. To leverage the advantages of different features, a fusion strategy based on similarities is introduced to fuse multiple features. Experimental results demonstrate the effectiveness of the proposed method for image retrieval.