Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.590422
Title: Image segmentation using deformable spatial priors
Author: Hasan, Basela Sharif
Awarding Body: University of Leeds
Current Institution: University of Leeds
Date of Award: 2012
Availability of Full Text:
Access through EThOS:
Abstract:
Image segmentation is one of the main problems that need to be solved as a component procedure in many computer vision tasks such as recognition, image editing, and indexing. Poor quality segmentation results can markedly deteriorate the performance demonstrated by the whole task. Therefore, a great deal of research heeds to the set of segmentation techniques focused on finding high accuracy segmentations. Existing methods tend to exploit low and high level information about the object in a given image. Incorporating shape priors within the MRF formulation were shown to be extremely helpful in finding desired segmentations. This thesis presents a method for segmenting the parts of a known object within images. The method builds on an existing MRF formulation incorporating a prior shape model and colour distributions for the constituent parts. As a means to tackle this problem when these instances exhibit large variations in projected shape and colour: the proposed approach is to learn a. probabilistic model for variations in the shape of the class of objects and to use this model in segmenting. For efficient search on shape latent parameters, a Branch & Bound approach is formulated to provide upper bounds on the pixelwise prior probabilities over the selected shape space used in this search. Moreover, a simple extension is made to the MRF formulation to deal simultaneously with multiple objects within a global optimisation. Finally, the method is evaluated on a library of images depicting people wearing suits - the aim being to segment the shirt, jacket , tie, and head/face for each individual. Results demonstrate improved performance in terms of accuracy over the state of the art for this task.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.590422  DOI: Not available
Share: