Image retrieval using an improved similarity measure: SRIC Similarity with Region Importance and Consistency

Content based image retrieval has become a major research interest recently. This paper presents an improved image similarity measure for image retrieval system. In the region based image comparison, two images are usually compared in terms of sum of the Euclidean distances among their regions. In this work, the image similarity measure is enhanced through a fuzzyfication of regions' importance and inter-region similarity. First, a “Mutual Consistency” function is defined for regions' comparison to avoid undesired inter-region similarity. Thereafter, an improved “Region Importance” function is developed to weight the regions of an image for overall image comparison. Utilizing fuzzy concepts of size and shape features of the regions, these two functions impose additional constrains on similarity measure that helps to improve the image retrieval results. Comparative results with the well-known IRM method are given to illustrate the effectiveness of the proposed approach.