A efficient approach for content-based color image retrieval

Image retrieval based on matching of color histograms is a simple yet effective method. Histograms are invariant to translation and rotation about viewing axis, and change only slowly under changes in angle of view, scale and occlusion. The major problem for this method is that colors from two adjacent histogram bins reveal only trivial difference. A small change in illumination condition, or influence of noise can cause pixels to drift from one bin to an adjacent bin. As a result, two images with quite similar color appearance may have two different sets of histogram bins. Matching these histogram bins would fail to retrieve all similar image from database. Furthermore, a traditional histogram method is inappropriate to image which is variable in dimension. In this paper, we focus on generating invariant features and forming similarity function, which includes image histogram creation, histogram normalization and similarity matching generating. Experimental results are used to demonstrate the performance of the method.