Gaussian Mixture Model based Spatial Information Concept for Image Segmentation

Open Access
Author:
Description:
Segmentation of images has found widespread applications in image recognition systems. Over the last two decades, there has been a growing research interest in model-based technique. In this technique, standard Gaussian mixture model (GMM) is a well-known method for image segmentation. The model assumes a common prior distribution, which independently generates the pixel labels. In addition, the spatial relationship between neighboring pixels is not taken into account of the standard GMM. For this reason, its segmentation result is sensitive to noise. To reduce the sensitivity of the segmented result with respect to noise, Markov Random Field (MRF) models provide a powerful way to account for spatial dependencies between image pixels. However, their main drawback is that they are computationally expensive to implement. Based on these considerations, in the first part of this thesis (Chapter 4), we propose an extension of the standard GMM for image segmentation, which utilizes a novel approach to incorporate the spatial relationships between neighboring pixels into the standard GMM. The proposed model is easy to implement and compared with the existing MRF models, requires lesser number of parameters. We also propose a new method to estimate the model parameters in order to minimize the higher bound on the data negative log-likelihood, based on the gradient method. Experimental results obtained on noisy synthetic and real world grayscale images demonstrate the robustness, accuracy and effectiveness of the proposed model in image segmentation. In the final part of this thesis (Chapter 5), another way to incorporate spatial information between the neighboring pixels into the GMM based on MRF is proposed. In comparison to other mixture models that are complex and computationally expensive, the proposed method is robust and fast to implement. In mixture models based on MRF, the M-step of the EM algorithm cannot be directly applied to the prior distribution for maximization of the log-likelihood with respect to the ...
Publisher:
University of Windsor
Year of Publication:
2011-01-01T08:00:00Z
Source:
Electronic Theses and Dissertations
Document Type:
info:eu-repo/semantics/doctoralThesis ; [Doctoral and postdoctoral thesis]
Language:
eng
Subjects:
Electrical engineering
DDC:
006 Special computer methods (computed)
Rights:
info:eu-repo/semantics/openAccess ; http://creativecommons.org/licenses/by-nc-nd/4.0/
Terms of Re-use:
CC-BY-NC-ND
Relations:
URL:
https://scholar.uwindsor.ca/etd/438
https://scholar.uwindsor.ca/context/etd/article/1437/viewcontent/nguyen1j.pdf
Content Provider:
University of Windsor, Ontario: Scholarship at UWindsor  Flag of Canada