Content-based image retrieval using associative memories

The rapid growth in the number of large-scale repositories has brought the need for efficient and effective content-based image retrieval (CBIR) systems. The state of the art in the CBIR systems is to search images in database that are "close" to the query image using some similarity measure. The current CBIR systems capture image features that represent properties such as color, texture, and/or shape of the objects in the query image and try to retrieve images from the database with similar features. In this paper, we propose a new architecture for a CBIR system. We try to mimic the human memory. We use generalized bi-directional associative memory (BAMg) to store and retrieve images from the database. We store and retrieve images based on association. We present three topologies of the generalized bi-directional associative memory that are similar to the local area network topologies: the bus, ring, and tree. We have developed software to implement the CBIR system. As an illustration, we have considered three sets of images. The results of our simulation are presented in the paper.