SIMILARITY BASED IMAGE RETRIEVAL

Abstract

With the spurt in computing and communication technology, more and more images are being captured, stored and widely used in multimedia collections. Image Retrieval Systems are required to utilize such collections of images efficiently. The work presented in the report is an effort to propose and implement different mechanism and techniques for efficient retrieval of image databases using low-level features such as texture, color and shape. In the first part, a new approach for Similarity Based Image Retrieval, which is based on well-known and widely used color histograms; is proposed. Contrasting to previous approaches, such as using a single color histogram for the whole image, or local color histograms for a fixed number of image cells, the one being proposed (named Merged Spatial Color) uses a variable number of histograms, depending only on the actual number of colors present in the image. Next, texture feature extraction method based on Statistical methods is presented. It has very less computational complexity and at the same time good retrieval performance. Then, shape feature extraction method is proposed that integrates the two methods elegantly, namely Hu moments and Gradient approach. Experiments have shown that retrieval performance of the integrated approach is better than both the approaches. A technique that computes the integrated feature vector, which combines color, texture and shape features, is presented in the next part. Then we address the issue of clustering and searching the image databases. Top down clustering approach is used for clustering purposes and branch and bound searching is used to search images in the database. A free parameter is proposed that can reduce the search complexity. A case study has been carried out on advertisement images taken from magazines on which the technique, which computes the integrated feature vector for color, texture and shape along with a technique presented in the chapter, is implemented. The project has been implemented. using JDK version 1.3 with Windows XP operating system.