Use this URL to cite or link to this record in EThOS:
Title: Development of a perception oriented texture-based image retrieval system for wallpapers
Author: Qian, Yu
ISNI:       0000 0004 2713 1692
Awarding Body: Middlesex University
Current Institution: Middlesex University
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
Access from Institution:
Due to advances in computer technology, large image collections have been digitised and archived in computers. Image management systems are therefore developed to retrieve relevant images. Because of the limitations of text-based image retrieval systems, Content-Based Image Retrieval (CBIR) systems have been developed. A CBIR system usually extracts global or local contents of colour, shape and texture from an image to form a feature vector that is used to index the image. Plethora methods have been developed to extract these features, however, there is very little in the literature to study the closeness of each method to human perception. This research aims to develop a human perception oriented content-based image retrieval system for the Museum of Domestic Design & Architecture (MoDA) wallpaper images. Since texture has been widely regarded as the main feature for these images and applied in CBIR systems, psychophysical experiments were conducted to study the way human perceive texture and to evaluate five popular computational models for texture representations: Grey Level Co-occurrence Matrices (GLCM), Multi-Resolution Simultaneous Auto-Regressive (MRSAR) model, Fourier Transform (FT), Wavelet Transform (WT) and Gabor Transform (GT). By analyzing experimental results, it was found that people consider directionality and regularity to be more important in terms of texture than coarseness. Unexpectedly, none of the five models appeared to represent human perception of texture very well. It was therefore concluded that classification is needed before retrieval in order to improve retrieval performance and a new classification algorithm based on directionality and regularity for wallpaper images was developed. The experimental result showed that the evaluation algorithm worked effectively and the evaluation experiments confirmed the necessity of the classification step in the development of CBIR system for MoDA collections.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available