Merged region based image retrieval

In Region based Image Retrieval (RBIR) methods, region matching mainly focuses on region-to-region and image-to-image methods. The former may cause loss of image information and the latter may lead to similar regions being matched repeatedly. To solve these problems, we propose a new image retrieval method based on merged regions, and feature extraction and matching are processed at the category level. Merged regions in an image belong to the same category to some extent, and are obtained by a statistical region merging and affinity propagation (SRM-AP) algorithm. For feature extraction, regional convolution mapping feature (RCMF) based on the convolutional neural networks (CNN) are extracted. RCMF is further combined with the number and distribution of regions to represent the characteristics of merged regions. Moreover, to match the merged regions according to their significance in images, an integrated category matching (ICM) method is designed. Experimental results on Corel-1000 and Caltech-256 show that the proposed method is more effective than some existing RBIR methods.