A content-based image retrieval system for texture and color queries

Abstract

In recent years, very large collections of images and videos have grown rapidly. In parallel with this growth, content-based retrieval and querying the indexed collections are required to access visual information. Two of the main components of the visual information are texture and color. In this thesis, a content-based image retrieval system is presented that computes texture and color similarity among images. The underlying technique is based on the adaptation of a statistical approach to texture analysis. An optimal set of five second-order texture statistics are extracted from the Spatial Grey Level Dependency Matrix of each image, so as to render the feature vector for each image maximally informative, and yet to obtain a low vector dimensionality for efficiency in computation. The method for color analysis is the color histograms, and the information captured within histograms is extracted after a pre-processing phase that performs color transformation, quantization, and filtering. The features thus extracted and stored within feature vectors are later compared with an intersection-based method. The system is also extended for pre-processing images to segment regions with different textural quality, rather than operating globally over the whole image. The system also includes a framework for object-based color and texture querying, which might be useful for reducing the similarity error while comparing rectangular regions as objects. It is shown through experimental results and precision-recall analysis that the content-based retrieval system is effective in terms of retrieval and scalability.