An efficient multimedia image retrieval system

This paper describes a computationally efficient method for fast retrieval of color images of multimedia and imaging databases. Although the proposed algorithm can operate in an n-dimensional (n-D) feature space for search, in our experiments we use only one three-dimensional (3-D) vector as key for indexing and searching color pictures of the selected archives. A new feature extraction and matching technique is developed based on the first-order statistics of color image data. Eigenvalue analysis provides an effective way of reducing 3-D color data to a one-dimensional (1-D) array. This feature extraction and reduction step is performed only once when an (R,G,B) color picture is submitted for storage or query. As for similarity measure, the Tanimoto coefficient is selected to be a computationally high performance matching algorithm to evaluate the search results. It is shown in the paper that the idea of projection-based retrieval is similar to the well-known histogram intersection operation of Swain and Ballard(8). The algorithm described here has been tested on eleven different databases, each of which consists of various color images of different scenes stored in a content addressable stack. The efficacy of retrieval was determined using the percentage efficiency measure eta=p/P, where p is the number of similar pictures retrieved in a short list and P is the total number of similar pictures in an archive. The experimental results yield almost 90% average retrieval efficiency for the eleven databases searched with the 3-D index or key vector. The feature extraction and matching parts of the method are not affected by major changes in the databases since the extracted features depend on the image statistics and vary with the color data. Moreover, the similarity ratio selected here is invariant to changes in image brightness which are a scaling factor for the feature values. These properties make the algorithm robust and highly efficient for many multimedia imaging database applications (e.g., trademark registration, fingerprint analysis, face identification, etc.).