1. Identification of Facial Features on Android Platforms
Josette. C. Tagatio Mawafo1, W.A. Clarke1 and P.E. Robinson1
University of Johannesburg, Johannesburg, South Africa
Published in: Industrial Technology (ICIT), 2013 IEEE International Conference

2. Abstract
present and investigate the performance of an algorithm designed to identify facial features on an android mobile platform

3. Feature Point Extraction
A. Skin Seqmentation
The RGB color image is transformed into the YCrCb color image using the following formula:

4. Feature Point Extraction
A. Skin Seqmentation

5. Feature Point Extraction(cont.)
B. Eye Extraction
How to extract the face？
Skin region
=1.618
Including face , neck and part of chest
width of the head
Height of the head

6. Feature Point Extraction(cont.)
B. Eye Extraction

7. Feature Point Extraction(cont.)
B. Eye Extraction

8. Feature Point Extraction(cont.)
To page 16
Feature Point Extraction(cont.)
B. Eye Extraction

9. Feature Point Extraction(cont.)
B. Eye Extraction
mark
Left iris
Right iris

10. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(1) N / D = 0.33
D：The distance between
the two eyes D
N： vertical distance between the eyes and the nose tip N

11. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(2) M / D = 1.10
D：The distance between
the two eyes D
M： the vertical distance between the eyes and the mouth center M

12. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(3) E = 0.8 D
D：The distance between
the two eyes D
E：The width of the nose E

13. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(4) E / L = Phi
Phi = 1.618
E = 0.8D
L = the distance from the nose to the mouth E L

14. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(5) K / E = Phi
Phi = 1.618
E：0.8 D
K：the length of the lips E K

15. Feature Point Extraction(cont.)
C. Prediction Feature Regions
(6) The width of the eyes is similar to the nose width
width of the eyes
E = 0.8D

16. Feature Point Extraction(cont.)
To：Page 7
Feature Point Extraction(cont.)
D. Localizing Features Points
On the binary image obtained from applying the threshold to the non- maximum suppressed image we notice that we have an edge detected image where only the most important parts of the face are highlighted.

17. Experimental Result A. Database And Testing Methology
(1) 500 facial images of different individuals from the UJ(University Of Johannesburg) database.
(2) The size of each image is 480 * 624.
(3) The images are named 0000x.jpg, where x ranges for 000 to 500.

18. Experimental Result(cont.)
A. Database And Testing Methology
(4) images are taken under different conditions: facial expression open-eyes; smiling / non-smiling and facial details glasses / no glasses.
(5) Eyes need to be open
(6) images are grey homogeneous background and are in upright, frontal.

19. Experimental Result(cont.)
B. Prediction Accuracy Of The Nose And Mouth

20. Experimental Result(cont.)
C. Performance Of The Feature Detector
Android device：Samsung galaxy 10.1 tablet
processing capabilities： (1 GHz dual core NVIDIA Tegra 2 processor, and 1GB RAM )
The java implementation was run on 45 images, the average processing time is recorded in table II