Image Retrieval Based on an Optimum Matched Cluster-pairs Set

A new image retrieval method, which is based on an optimum matched cluster-pairs set between two images and referred to as IROMCS, is proposed. it uses assignment model to seek an optimum matched cluster-pairs set. Based on an optimum matched cluster-pairs set, a similarity measure is defined. Since the matching algorithm study the similarity of two clusters based on these two scales both color and size, the matching between two clusters contains relatively more complete information, so that the similarity based on an optimum matched cluster-pairs set theoretically should be able to more effectively remove irrelevant images from retrieval result and can achieve the better retrieval effect. In comparison with most widely used methods based on color histogram, experimental results show that both recall and precision with IROMCS are improved significantly, in particular it can effectively retrieve such two images that their structures are similar but their colors not identical, however, those methods can´t achieve such an retrieval result.