Hybrid scheme for fast and accurate image retrieval based on color descriptors

Abstract

This paper proposes a new image retrieval system, thatuses only color features and it's based on a hybrid scheme which combines crisp and fuzzy techniques, in order to retrieve color based similar images. The system comprises2 units. The first unit uses Binary Haar Wavelet Descriptor in a histogram that has been proposed in MPEG-7. Anew fuzzy-linking method of color histogram creation isalso proposed, based on the HSV color space. The second unit provides this histogram and decides about the similarity. The system is suitable for accurately retrieving imageseven in distortion cases such as deformations, noise and smoothing. It is tested on a large number of image sselected from proprietary image data bases or randomly retrieved from popular search engines. The retrieval rate approximates 45 images (size of 250 X 250 pixels) per second, assuming no prior stored feature information in the searched image data bases. To evaluate the performance of the proposed system, objective measure called ANMRR is used.