Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.384933
Title: Surface defect detection on textured background
Author: Song, Keng Yew
ISNI:       0000 0001 3470 9964
Awarding Body: University of Surrey
Current Institution: University of Surrey
Date of Award: 1993
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
This thesis addresses the problem of defect detection on complex textural surfaces. In general, whether the texture to be inspected is regular or random, in image terms it is characterized by local variations in pixel grey level values. These normal variations render the problem of texture defect detection extremely difficult as defects are often manifested by grey level changes and their detection requires more than mere pixel comparisons. In the thesis, classical techniques on texture representation are studied and various existing texture defect detection algorithms are reviewed. Three novel algorithms have been developed to tackle the problem of defect detection on random or regular textures. The first two are devoted to the problem of crack detection and the third algorithm is devoted to the problem of detecting regional defects. For texture crack detection, a cojoint spatial and spatial frequency representation, that is, wigner distribution is proposed to model the inspected texture surface. A detailed analysis of the wigner distribution, its properties and the effect of windowing on its crack detection performance are carried out. Two postprocessing methods, ie, probabilistic relaxation labelling and linear filtering are incorporated into the crack detection algorithm to refine the results. The potential of the Wigner model has also been explored by modifying the crack detection algorithm so as to detect other types of defects. For real world applications, an efficient crack detection algorithm based on a new distribution is proposed. The algorithm is shown to produce comparable results but in much shorter time. For regional defect detection, a hybrid chromato-structural approach to colour texture representation is proposed where combined colour texture information is extracted from various chromatic classes associated with the inspected surface. In the approach, a unified defect detection framework which combines a new colour clustering scheme, morphological smoothing and blob analysis are used to capture the relevant combined colour texture information. With this framework, good defect detection results are obtained and presented in this thesis.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.384933  DOI: Not available
Keywords: Pattern recognition & image processing
Share: