Health diagnostic - Reasech Proposal

JAMES COOK AUSTRALIA INSTITUTE OF HIGHER LEARNING

IN SINGAPORE

HEALTH DIAGNOSTIC BY ANALYSING FACE IMAGES USING MOBILE DEVICES

Instructor : Dr. Insu Song

Student : Ho Thi Hoang YenEmail: [email protected]

CONTENT

1. Background

2. Introduction

3. Problems & Solution

4. Preparing DATA

5. Results & Conclusion

6. Future Plan

7. References

FACIAL PALSY

Facial Palsy is the paralysis of nerve VII Þ Recover after 6

months

BUTÞ This is also the

symptom of a

STROKE – which causes many deaths in the world.

STROKE

STROKE: - The third leading

cause of death.- Each year:

approximate 795000 people suffer a stroke

- More than 140,000 people die in the United States.

TOO LATE!

SKIN CANCER & ROSACEA

APPSTORE & GOOGLE PLAY

INTRODUCTION

PROBLEMSServer side :

Build a symptoms

data warehouse

Client side : Take face

image

Analysis the photo if there

is any symptom (server or

client)BUT, there is still a long way from concept to application, we have to face many problems, example:- How fast the analysis can be with a huge database while

the database becomes bigger and bigger by time?- Where should the database locate? On the cloud and

receive the request from the client then process OR on the client side and need to be updated all the time?

- Will people agree to share their personal sensitive information as photos and health condition?

FACIAL POINTS EXTRACTIONFully Automatic Facial Feature Point Detection Using Gabor

Feature Based Boosted Classifiers. (Danijela & Maja)

• Using robust real-time face detection of Viola-Jones (replace adaboost with gentleboost) (shape based method)

• Using the position of iris to divide the face’s into Regions of Interested (ROIs)

• Feature extraction by Gabor WaveletResult :

• Face detection : 100%• Feature points detection : 93%• Outperformed PCA , FLD and LFA

FACIAL POINTS EXTRACTIONFully Automatic Facial Feature Point Detection Using

Gabor Feature Based Boosted Classifiers.

Weakness:- Cannot guarantee on Expressional

faces

FACIAL POINTS EXTRACTIONFacial Feature Tracking under Varying Facial Expressions and Face Poses based on Restricted Boltzmann Machines

(Wu , Wang & Ji) • Capture the distinctions & variations of face shapes due

to facial expression and pose change in a UNIFIED framework

• Using FrontalRBM & PoseRBM • Using 3-way RBM : capture the relationship between

frontal & non-frontal faces.Result• Improve the accuracy and robustness for various

face pose & expressions

DATA COLECTING

RESULT

• Find the solution to the problem:Extract facial feature points

Handle expressional & expressionless faces data by merging

Gabor Filter Method and Restricted Boltzmann Machines.

• Propose a plan to collect trustable data to fulfill the training

task.

CONCLUSION

• Facial Palsy, Skin Cancer, Flu, Stroke,.. can be soon

diagnostic.

• By doing literature review, we proposed a method to build

health diagnostic systems.

• Apply improved Jones-Viola method & Restricted Boltmann

Machines method to detect faces and extract facial

features.

• Grouping the data by diseases and applying to do searching

match sickness with input face data.

FUTURE PLAN

• Implement the two methods into a mobile application (on

iOS/Android)

• Implement the server system (stored sample data)

• Identify problems and search for solutions.

• Do more research on algorithms to improve the performance

of the program.

• Improve data

• Test and release.

REFERENCES

• http://www.facialparalysisinstitute.com/photo-gallery.html

• http://nccd.cdc.gov

• Vukadinovic, D., & Pantic, M. (2005). fully automatic facial feature point

detection using gabor feature based boosted classifiers. systems, man

and cybernetics, 2005 IEEE international conference on. IEEE.

• Wu, Y., Wang, Z., & Ji, Q. (2013). Facial feature tracking under varying

facial expressions and face poses based on restricted boltzmann

machines. Computer Vision and Pattern Recognition (CVPR), 2013

IEEE Conference on. IEEE.

THANK YOUQuestions ?