Block-based Against Segmentation-based Texture Image Retrieval
Mohammad Faizal Ahmad Fauzi (Multimedia University, Malaysia)
Paul H. Lewis (University of Southampton, United Kingdom)
Abstract: This paper concerns the best approach to the capture of local texture features for use in content-based image retrieval (CBIR) applications. From our previous work, two approaches have been suggested, the multiscale block-based approach and the automatic texture segmentation approach. Performance comparison as well as advantages and disadvantages of the two methods are presented in this paper. The databases used are the Brodatz and VisTex databases, as well as three museum image collections of various sizes and contents, with each collection presenting different challenges to the CBIR systems. Experimental observations suggest that the two approaches both perform well, with the multiscale technique having the edge in retrieval performance and scale invariance, while the segmentation technique has the edge in lighter computational complexity as well as having the shape information for later purposes. The choice between the two approaches thus depends on application.
Keywords: content-based image retrieval, discrete wavelet frames, multiscale technique, texture, texture segmentation
Categories: H.3.1, H.3.3