Comprehensive evaluation of image color features for CBIR systems

Yiyue Gao (2009). Comprehensive evaluation of image color features for CBIR systems MPhil Thesis, School of Information Technol and Elec Engineering, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
n41315828_MPhil_abstract.pdf Abstract application/pdf 89.30KB 2
n41315828_MPhil_totalthesis.pdf Final Thesis Lodgement application/pdf 3.06MB 6
Author Yiyue Gao
Thesis Title Comprehensive evaluation of image color features for CBIR systems
School, Centre or Institute School of Information Technol and Elec Engineering
Institution The University of Queensland
Publication date 2009-03
Thesis type MPhil Thesis
Supervisor Xiaofang Zhou
Heng Tao Shen
Total pages 104
Total colour pages 10
Total black and white pages 94
Subjects 08 Information and Computing Sciences
Abstract/Summary With the development of computer technologies and specially the advent of Internet, there has been an unprecedented demand in using computer to create, store, transmit, process, and retrieve digital images. Therefore a wide range of image color features have been proposed by various researchers on image processing and information retrieval in the pursuit of automatically creating indexes for image databases. However, with the proposers of each image color feature choosing their own methods and ground truth to prove their effectiveness, acquiring a reliable knowledge of the real performance of each feature becomes quite difficult. There are several evaluation works attempting to compare the performance of these features, whereas the image collections they used to conduct experiments are not large and diverse enough to make convincing and reliable comparisons. This thesis aims at conducting an objective and comprehensive evaluation of the most commonly-used image color features by reducing the subjectivity and randomness in experimental methodology. We firstly conduct experiments to choose the most suitable color space and similarity measurement in order to achieve maximized performance. The image collections we used consist of both photos from the web and professional image collections such as Corel, so that the images are of a wide range of subjects and can more effectively represent the images that users would deal with in real world. In addition, we propose an innovative interface for CBIR system using fisheye view to display retrieval results for improved usability, as we find most currently available CBIR systems are still using traditional grid view to display the retrieved images, which typically are too many to be properly shown in the same size. Our interface displays images simulating the way a fisheye observes the surroundings by showing retrieved images in focus area in larger size and the less important images in consecutively smaller sizes.
Keyword CBIR
color feature
Additional Notes 24,26,30,34,61,78,79,87,89,91

Citation counts: Google Scholar Search Google Scholar
Created: Thu, 24 Sep 2009, 19:51:41 EST by Mr Yiyue Gao on behalf of Library - Information Access Service