Use this URL to cite or link to this record in EThOS:
Title: Region segmentation of images using a scale-space approach
Author: Iyer, B. K.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2002
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
Region segmentation is the process by which an image is segmented into its constituent regions, where a region is a group of connected pixels with similar properties. Closely related to region segmentation is edge detection where the edges form the boundaries between regions. Region segmentation is important in many areas of computer vision and image processing. It is often the first step towards interpreting an image, where the regions correspond to objects in the image. The first part of this dissertation introduces the topic of region segmentation of images and also discusses the common data structures by which regions are represented. Region segmentation by scale-space methods, namely the datasieve (a recursive non-linear morphological filter), is proposed and the next part of the dissertation introduces the datasieve. Datasieves, as the name suggests, are methods by which data can be "sieved" based on the size of elements of the data. Datasieves have been found to be robust and are computationally efficient. They have been used in a number of applications such as position estimation of objects in an image, object recognition, motion estimation, and region segmentation. A scale-tree of the image can be generated based on the scale-space analysis using the datasieve. The dissertation demonstrates some problems in using the datasieve-based scale-tree, in its direct form, for region segmentation of images. The concept of the "seed node" from the datasieve scale-tree is introduced, based on which the datasieve is used as the primary tool for developing methods for segmenting grey scale and color images. Emphasis is also given to demonstrate segmentation results of a multi-scale nature using the datasieve. The possibilities of using the properties of the seed node for applications such as object tracking are shown. The theme of the dissertation then shifts to incorporate non-linear image pyramids in developing faster methods for region segmentation and edge detection.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available