Files and links (1)
PDFOpen Access (License Unspecified), Open Access
Dissertation
Unsupervised segmentation of noisy and textured images modeled by Markov random fields
Doctor of Philosophy (Ph.D.), Drexel University
Sep 1992
DOI:
https://doi.org/10.17918/00000826
Abstract
Image segmentation, i.e. the partition of an image into meaningful components, is an important step in low level vision. Here we consider the segmentation of noisy images, natural images with smooth surfaces, and textured images. The presented segmentation algorithms are completely data-driven and no a-priori knowledge about the image model parameters is assumed. Images are modeled by Markov random fields and the problem is cast into a statistical framework. Thus, statistical techniques are employed to achieve the segmentation. The algorithms are derived from the maximization of the joint probability density function of the observations and of the desired segmentation. In the case of the textured images, the image undergoes a transformation which reduces the segmentation problem into one of restoring a vector-valued noisy field. All the parameters involved are estimated by the algorithms, together with the desired segmentation. The algorithms are iterative and toggle between obtaining the maximum likelihood estimates of the parameters and updating the segmentation by using the maximum a-posteriori estimation criterion. The algorithms are general in nature and can be applied to natural images of objects with smooth surfaces, noisy or blurred images and textured images. The time requirement is very reasonable and the algorithms are highly parallelizable when parallel processors are available. The texture segmentation algorithm is not limited to images but it is applicable to any vector-valued signals, such as signals from multiple colors or multiple sensors.
Metrics
29 File views/ downloads
67 Record Views
Details
- Title
- Unsupervised segmentation of noisy and textured images modeled by Markov random fields
- Creators
- Georghios K. Gregoriou
- Contributors
- Oleh John Tretiak (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Doctor of Philosophy (Ph.D.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- xiii, 154 pages
- Resource Type
- Dissertation
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering [Historical]; Drexel University
- Other Identifier
- 991014970311404721