Learning and Adaptive Characterization of Visual Contents in Image Retrieval Systems

This chapter examines the adoption of neural network techniques for implementing the relevance feedback module and the feature extraction module in the figure. It investigates the adoption of neural network techniques to the task of the characterization of edges, which is considered one of the most significant features in images and is essential for the further extraction of texture and shape information for the purpose of retrieval. The chapter explores a learning technique which enables the radial basis function network to progressively model the notion of image similarity for effective searching. Accurate characterization of visual information is an important requirement for constructing effective content-based image retrieval (CBIR) system. Image content in a CBIR system is usually expressed in the form of a set of features which characterizes the color, texture, and shape information of individual images. In the retrieval process, the user may have difficulty specifying a query that represents important aspects of the desired image or class of images.