Content-based retrieval for image databases

Advances in digital storage and processing speed have made feasible the creation of large image databases with rapid access to individual items stored therein. The huge data sizes of images and the enormous number of images in a typical image database, coupled with inexact nature and subjective interpretations, have called for content-based retrieval systems. Fast and accurate retrievals are crucial for such systems to be used efficiently.; This project provides an overview on the Content-based Image Retrieval (CBIR) techniques developed recently. Research directions and current available CBIR systems are presented. Important issues such as image segmentation algorithms, image logic structure and spatial relationships, spatial access methods and Query by Visual Example techniques (QVE) are discussed in detail.; A prototype image retrieval system called IMAGESEEK is implemented using the JAVA programming language. The system enables the search of natural colour images and demonstrates the various ideas of Query by Visual Example techniques. A framework for CBIR systems is proposed. Experimental results of different QVE algorithms are discussed and compared with each other. The system has been successful in retrieving images from our sample data sets by their global and local colours. The user-friendly interface of IMAGESEEK allows the user to tailor and refine the query interactively by changing the retrieval algorithm, the threshold value, the weights, and the selected region of the query image. IMAGESEEK provides us a way to understand the key issues of CBIR techniques. It is a small but valuable component in the collection of multimedia retrieval systems.