Efficient Techniques for Processing Image Databases

Abstract

In this work, we study the problem of Content Based Image Retrieval as an evolution of text-based and classification systems. Some of the recently developed systems in the field were reviewed and the weak points of these systems were highlighted. These systems were classified into color based and shape based solutions. The problem of database browsing and navigation was also addressed. Two novel shape based systems are proposed. In the first system, a new shape-based search technique using the Radon transform is considered. During the indexing phase, a proposed compact feature vector is extracted from the image to be indexed after computing its radon transform; during the retrieval phase, the system uses the vector previously stored as a key for finding a short listed result set that will be introduced to the second phase of the search process. In the Second system, a new fast search technique for images stored in databases using similarity retrieval depending on shape is presented. A short feature vector is extracted from the query image with small computational complexity. This vector is compared against all blocks of the same area in the target image, if the comparison yields an acceptable result, then a more detailed distance measure is calculated, the process continues until all areas are compared. The block having minimum distance is chosen as a representative of the distance measure between the query image and the target image. In color based systems, two novel approaches are proposed. In the first system, we propose an image indexing algorithm based on color histogram bins and a new Image Retrieval Query Language (IRQL).