Performance evaluation of multiple regions-of-interest query for accessing image databases

The paper addresses two fundamental aspects of content based image retrieval (CBIR) systems: visual feature extraction and retrieval system design which uses multiple regions-of-interest (ROIs) as the key to retrieve relevant images. Visual feature extraction is performed on all images in the database where each image is segmented into a number of homogenous regions. Low-level attribute calculation is performed on each region whose color and texture information is obtained using color histogram analysis and wavelet decomposition, respectively. The proposed retrieval system supports queries based on system-user interaction that takes into account the user´s requirements. The implementation of binary color sets is employed to ensure efficiency of the system. Several multiple regions-of-interest query strategies have been adopted which use statistical analysis and a hierarchical framework to improve the retrieval results. These schemes are compared with the single ROI retrieval methods. Experimental results show the Multiple ROI query strategies perform better than other existing methods, such as those using global features or single ROI