Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.281795
Title: Texture analysis and synthesis using the multiresolution Fourier transform
Author: Xu, Daoyi
ISNI:       0000 0001 3573 8732
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 1994
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
In this thesis, a new frequency domain approach to analysis of texture is presented, in which both the statistical and structural aspects of the problem are combined in a unified framework, the Multiresolution Fourier Transform. (MFT). The analysis scheme consists of two main components: texture synthesis and texture segmentation. The synthesis method works by identifying, for pairs of texture ‘patches’ of a given size, the affine co-ordinate transformation which gives the best match between them. This allows the analysis to take account of the geometric warping which is typically found in images of natural textures. By variation of scale, using the MFT, it is possible to identify the scale of the texels giving the best results in the matching process. The technique has the potential to deal effectively with textures having varying amounts of structure and can be used both for segmentation and resynthesis of textures from a single prototype block. The texture segmentation makes use of the localisation properties of the MFT to detect texture boundaries using the MFT coefficient magnitudes, which incorporate both boundary and region information, in order to place texture boundaries accurately. A segmentation algorithm is described, starting with pre-smoothing to reduce texture fluctuations followed by an edge detection based on a Sobel operator to give an initial texture boundary estimate. Both boundary enhancement and region averaging are accomplished by adopting a 'link probability function’ to introduce dependence on neighbouring data, allowing the boundary to be refined successively to achieve segmentation. The method effectively uses the spatial consistency of boundary estimates at larger scales to propagate more reliable information across scales to improve the accuracy of the boundary estimate. Experimental results are presented for a number of synthetic and natural images having varying degrees of structural regularity and show the efficacy of the methods.
Supervisor: Not available Sponsor: Government of China ; Xing zheng yuan
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.281795  DOI: Not available
Keywords: QA76 Electronic computers. Computer science. Computer software ; T Technology (General)
Share: