Content-Based Image Retrieval Using Color Features of Partitioned Images

Content-based image retrieval using low-level features such as color, texture and shape is one of the major challenges in image processing and computer vision. Color as the most important low-level feature, has very wide applications in image retrieval systems. In this paper a new content-based image retrieval method using color feature of image regions is expressed. To this end, at the first step, images are partitioned to five fixed regions include center, top, bottom, left and right. Then the color autocorrelogram for each region is computed separately and kept as a feature vector to compare different images similarities. Because of the importance of the images center, weight of the center region is doubled when comparing similarity of images. For comparing other regions, difference of most similar regions is computed. This comparison makes the algorithm more invariant to rotation and to somehow changing the viewing angle, than the similar works. By combining the output of this method with the output of global color histogram-based retrieval method, system performance and accuracy of the results are more improved.