An Efficient and Flexible Matching Strategy for Content-based Image Retrieval

With the rapid growth of multimedia applications and digital archives, content-based image retrieval (CBIR) has emerged as an important area and received lots of attentions for the past decades. In practice, there are two major problems raised by a CBIR system: the feature descriptions of images and the expressions of users’ search preferences. To tackle these problems, we present a DCT-based feature descriptor coupled with an efficient way of expression for users’ preferences in this paper. Our approach partitions images into a number of regions with fixed absolute locations. Each region is represented by its low-frequency DCT coefficients in the YUV color space. Two policies are provided in the matching procedure: local match and global match. In the local match, the user formulates a query by selecting the interested region in the image. Candidate images are then analyzed, by inspecting each region in turn, to find the best matching region with the query region. For those query images without clear objects, the user can select the option “global match” instead. The experimental system shows that this approach is generally effective and particularly suited for images with interested regions having features which significantly differ from the global image features. With the help of friendly GUI, our system also allows users of any experience level to effortlessly get interested images from database. [Life Science Journal. 2010; 7(1): 99 – 106] (ISSN: 1097 – 8135).