Web-based Image Retrieval with a Case Study

Abstract. Advances in content-based image retrieval(CBIR)lead to numerous efficient techniques for retrieving images based on their content features, such as colours, textures and shapes. However, CBIR to date has been mainly focused on a centralised environment, ignoring the rapidly increasing image collection in the world, the images on the Web. In this paper, we study the problem of distributed CBIR in the environment of the Web where image collections are represented as normal and typically autonomous websites. After an analysis of challenging issues in applying current CBIR techniques to this new environment, we explore architectural possibilities and discuss their advantages and disadvantages. Finally we present a case study of distributed CBIR based exclusively on texture features. A new method to derive texture-based global similarity ranking suggests that, with a deep understanding of feature extraction algorithms, it is possible to have a better and more predictable way to merge local rankings from heterogeneous sources than using the commonly used method of assigning different weights.