Performance evaluation of color image retrieval

Abstract

In this paper, we have investigated the capabilities of 4 approaches for image search for a CBIR system. First two approaches are based on comparing the images using color histograms of RGB and HSV spaces, respectively. The other 2 approaches are based on two quantitative image fidelity measurements, Mean Square Error (MSE) and Structural Similarity Index (SSIM), which provide a degree of similarity between two images. The precision performances of approaches have been evaluated by using a public image database containing 1000 images. Finally effectiveness of retrieval has been measured for each method.