Advanced search options

Advanced Search Options 🞨

Browse by author name (“Author name starts with…”).

Find ETDs with:

in
/  
in
/  
in
/  
in

Written in Published in Earliest date Latest date

Sorted by

Results per page:

You searched for +publisher:"University of New South Wales" +contributor:("Sowmya, Acrot, Computer Science & Engineering, Faculty of Engineering, UNSW"). One record found.

Search Limiters

Last 2 Years | English Only

No search limiters apply to these results.

▼ Search Limiters


University of New South Wales

1. Chen, Yu. Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model.

Degree: Computer Science & Engineering, 2015, University of New South Wales

Considerable research has been done on automatic extraction of salient features from human faces in images over the last 20 years. For many existing systems, a prior face shape model has been utilized to align with unknown faces and locate important facial features, represented by a set of labelled points around the eyes, nose and mouth. Constraints are applied to the shape model to facilitate reliable face alignment. However, these constraints may lead to inaccuracy and failure in alignment due to complex situations that commonly occur in the real world. This thesis is aimed at extending a current face alignment method, namely Active Shape Model (ASM) proposed by Coates et al. (1995). ASM is a robust approach to accommodate non-rigid object segmentation in the field of computer vision and has been widely used to segment and acquire visual information from the world. In this thesis, some novel variations have been applied to classical ASM to improve its performance. More specifically, machine learning techniques and dynamic tracking ir. the model space are successfully combined with ASM to achieve improved segmentation, and a new hierarchical clustering framework is integrated to accommodate multi-view face shape alignment. In the proposed framework, three elements account for the main contributions. Firstly, the accuracy and robustness of the classical Active Shape Model are enhanced by learning nonlinear classifiers through machine learning algorithms. followed by a boosting method to automatically select the best classifier. Secondly, tracking of the motion of facial feature points benefits from dynamic modelling and particle filtering. Finally, a Gaussian mixture model is adopted to learn clusters of views, and parameters are updated by EM algorithm. Beyond the investigations of the framework, the most significant achievement is the idea of incorporating dynamic information in deformable shape models which distinguishes it from conventional template-based segmentation methods. A number of experiments have been conducted to evaluate the performance of the novel method. Different parameters have been tuned in search of optimal solutions. To assess generalizability, various popular public face databases are utilized. Comparison with published results of other segmentation methods have been carried out. Stability is demonstrated by a set of statistical tests. The proposed approach produces distinctive results on most of the major databases, also outperforms other algorithm both in accuracy and robustness. Advisors/Committee Members: Sowmya, Acrot, Computer Science & Engineering, Faculty of Engineering, UNSW.

Subjects/Keywords: Particle Filter; Active Shape Model; Machine Learning

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

Chen, Y. (2015). Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model. (Masters Thesis). University of New South Wales. Retrieved from http://handle.unsw.edu.au/1959.4/54529 ; https://unsworks.unsw.edu.au/fapi/datastream/unsworks:35191/SOURCE02?view=true

Chicago Manual of Style (16th Edition):

Chen, Yu. “Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model.” 2015. Masters Thesis, University of New South Wales. Accessed March 24, 2019. http://handle.unsw.edu.au/1959.4/54529 ; https://unsworks.unsw.edu.au/fapi/datastream/unsworks:35191/SOURCE02?view=true.

MLA Handbook (7th Edition):

Chen, Yu. “Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model.” 2015. Web. 24 Mar 2019.

Vancouver:

Chen Y. Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model. [Internet] [Masters thesis]. University of New South Wales; 2015. [cited 2019 Mar 24]. Available from: http://handle.unsw.edu.au/1959.4/54529 ; https://unsworks.unsw.edu.au/fapi/datastream/unsworks:35191/SOURCE02?view=true.

Council of Science Editors:

Chen Y. Facial Segmentation Using Boosted Dynamic Mixture Active Shape Model. [Masters Thesis]. University of New South Wales; 2015. Available from: http://handle.unsw.edu.au/1959.4/54529 ; https://unsworks.unsw.edu.au/fapi/datastream/unsworks:35191/SOURCE02?view=true

.