Initial Face Images With Glasses
Glasses Detection Based on Bayes Decision Rules
The Green Points are glasses points extracted based on Bayes Decision Rules.
Facial Images after Glasses Points Removal
[ Home | About Me | Research | Resume | Publications | Album | Link | Contact me]
Results and Experiments
Aim
To achieve aface recognition system robust to the presence of glasses, I developed a glasses detectio, extraction and removal algorithm.
Problems
It's difficult to classify glasses points and non-glasses points based on edge information, since there are too many confusion edge points beside glasses points, including eyes, nose, eyebrows, and wrinkle.
Results of Samples
Initial Face Images With Glasses
Glasses Detection Based on Bayes Decision Rules
The Green Points are glasses points extracted based on Bayes Decision Rules.
Facial Images after Glasses Points Removal
Experiments
Comparison of Deformable Contour method and Linear Interpolation based on Glasses Detection using Bayes Rule
| Detection Algorithm | Average Error of Left Part Glasses in 29 Facial Images | Average Error of Right Part Glasses Facial Images |
| Dynamic Programming | 27.3297 | 28.9375 |
| Based on Bayes Rule | 11.0963 | 19.627 |
After glasses points detection based on Bayes Rules, a simple linear interpolation algorithm is employed to obtain the contour of glasses frame. The comparison results of average errors of glasses contour are shown in above Table.
The
average error is computed by:
(8)
Where,
is the ith point of glasses chain,
which is obtained by manual labeling,
is the glasses point computed from glasses detection results by quantization of
direction and coordinate averaging to identify corresponding points between T
and G.
Comparison of False Detection of Glasses Existence
| Detection Algorithm | Based on Edge of Glasses | Based on Bayes Rules |
| False Detection | 2/419 | 0/419 |
This comparison is between "Towards Detection of Glasses in Facial Images"* and my methods in KRDL Facial Database.
Results Showed On Paper
The first line is original facial images with glasses.
The glasses points which are highlight have been extracted based on Bayes Rules, in the second line.
The glasses points are replaced by skin gray level neighboring with them.
* X. Jiang, M. Binkert, B. Achermann, H. Bunke, “Towards Detection of Glasses in Facial Images”, Proc. 14th ICPR, Brisbane, 1998, 1071 - 1073
[Go Top | Home | About Me | Research | Resume | Publications | Album | Link | Contact me]
