CONTENT-BASED IMAGE RETRIEVAL TECHNIQUES FOR LARGE IMAGE COLLECTIONS

Abstract

Content-based image retrieval (CBIR) refers to the retrieval of relevant images based on the image attributes like color, shape, texture or text. Such a system helps users (even those unfamiliar with the database) retrieve relevant images based on their content. Application areas in which CBIR is a principal activity are numerous and diverse. Art galleries and museum management, interior design, fabric and fashion design, retailing are just some examples. Color histograms are used to compare images in many applications. Their advantages are efficiency and insensitivity to small changes in camera viewpoint. However, color histograms lack spatial information so images with very different appearances can have similar histograms. Hence in the project, we will look into color-spatial techniques to improve the retrieval results. In a color-spatial retrieval techniques, the color information is integrated with the knowledge of the color's spatial distribution to facilitate content-based image retrieval. Several techniques have been proposed in the literature, but these works have been developed independently without much comparison. In this thesis, we investigate three color-spatial retrieval techniques - the signature-based technique, the partition-based technique and the cluster-based technique. We examine its representation, the similarity measures and the appropriate indexing structures. In addition, we develop relevant feedback mechanism to increase the retrieval performance of the system. Furthermore, we conduct numerous sensitivity analysis experiments to determine the optimal settings for the various techniques. Following that, we perform an experimental evaluation of the 3 techniques based on the retrieval effectiveness and retrieval efficiency. The experimental study is conducted on a large image database consisting of 12,000 images. With the proliferation of image retrieval mechanisms and the lack of extensive performance study, the experimental results can serve as useful guidelines in selecting and designing as new technique.