An effective content based image retrieval system using relevance search and low level features

Abstract

The enormous growth in the internet and multimedia technology has generated a huge amount of data in the form of images, videos, and audio. This has created the demand of systems which can store and retrieve multimedia data like images in an effective and efficient manner. Content Based Image Retrieval (CBIR) is the searching, navigation and retrieval of images based on their visual content. Visual content of image include color, texture, shape and spatial location of objects depicted in the image. The low level features like color, texture and shape have limited capability for describing visual contents of an image. Therefore semantic gap is observed between visual interpretation and representation of images with low level features. newline Researchers all over the world are trying to fill this semantic gap using low level features and their combinations. This thesis is a next step in this series. A number of problems in existing low level features and retrieval systems are identified and their solutions are proposed. newline Generally the accuracy of a Content Based Image Retrieval system decreases as the number and variety of images increases in the database. Similar images depicting different semantic concepts may be retrieved as a result. In addition, the extraction of shape features requires accurate segmentation of images. However, segmenting image itself is an open problem; therefore extraction of shape features is not much reliable. A CBIR using multistage retrieval strategy is proposed here to deal with these problems. newline Low level features of an image can be local and global. Global features are extracted using whole image and local features are extracted from a local region or part of image. Local features based CBIR systems are called Region Based Image Retrieval systems (RBIR). Accuracy of a RBIR system is higher than corresponding global CBIR in general. The traditional RBIR systems face the problem of precise query formulation and higher response time.