FACE DETECTION APPLICATION

brasscoffeeAI and Robotics

Nov 17, 2013 (3 years and 9 months ago)

61 views

LOGO

FACE DETECTION
APPLICATION

Member: Vu Hoang Dung



Vu Ha Linh



Le Minh Tung



Nguyen Duy Tan



Chu Duy Linh



Uong Thanh Ngoc



CAPSTONE PROJECT

Supervisor:
Phan

Duy

Hung

FDA TEAM

Contents

Introduction

1

Plan

2

Requirements

3

3

Implementation

4

4

Conclusions

5

1. Introduction


Existing Algorithm:














FDA Team

FDA TEAM

Elastic Bunch Graph Matching (EBGM)

3
-
D Morphable Model.

Boosting & Ensemble Solutions

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.71.97
50&rep=rep1&type=pdf


http://www.mpi
-
inf.mpg.de/~blanz/html/data/morphmod2.pdf


http://www.face
-
rec.org/algorithms/Boosting
-
Ensemble/16981346.pdf

1. Introduction


Existing product:














FDA Team

FDA TEAM

OpenCV


Intel’s Open Source
Computer Vision initiative

Face Tracking DLL from Camegie Mellon

Real
-
time face detection program from FhG
-
II

http://opencv.willowgarage.com/wiki/


http://chenlab.ece.cornell.edu/projects/FaceTrackin
g/#Download


http://www.iis.fraunhofer.de/bf/bv/ks/gpe/

1. Introduction


Idea:


Develop an application to detect Face in Image


Fast speed


Reliable


Can integrated with other products


FDA Team

FDA TEAM

Objective System

FDA Team

FDA TEAM

2. Plan


2.1 Roles and Responsibilities

FDA Team

PM
(
DungVH
)
Design sub
-
team
(
TanND
)
QA
(
LinhCD
)
Implemetation
sub
-
team
(
DungVH
)
Risk
management
(
TungLM
)
Configuration
management
(
LinhVH
)
Testing sub
-
team
(
NgocUT
)
Research
sub
-
team
(
TungLM
)
Document
maintainer
(
LinhVH
)
Document
writer
(
NgocUT
)
TungLM
TanND
DungVH
TanND
LinhCD
TungLM
TungLM
LinhVH
LinhVH
NgocUT
TungLM
DungVH
NgocUT
LinhCD
NgocUT
LinhVH
FDA TEAM

2. Plan

2.2 Software Process Model


Iterative Approach to Development

FDA Team

FDA TEAM

2. Plan


System Requirement










Tool Requirement


Visual Studio 2008.


SQL Server 2008.


.Net Framework 3.5.


Google code project site.

FDA Team

Operating

System

(OS)

Hardware

Microsoft

Windows

XP/

7

(
32

or

64

Bit)

/

Vista


1
.
5

GHz

32
-
bit

(x
86
)/
64
-
bit

(x
64
)

or

higher


1

GB

RAM

(
32
-
bit)

or

higher


2
GB

HDD

free

FDA TEAM

3.1 Functional Requirements


User friendly
-

user can easily understand and handle in first use



Support small
-

big size image with different quality



Support format files: JPG, BMP, PNG, JPEG



Allows user to test the algorithms of image processing.



The processing must have a sequence as Image Original


Convert
to HSV


Test H and V value of each pixel


Use 8 connected
neighbor to find different regions


Identify region of face.


FDA Team

FDA TEAM

3.2 Non
-
functional Requirements


The processing time of each function of image processing should be
about 2 seconds



The result of searching face in images is processed less than 3
seconds



Time processing of searching a faces in the face database is not
over 3 seconds


FDA Team

FDA TEAM

4. Implementation

4.1 System Architectural Design

FDA Team

FDA TEAM

4. Implementation

4.2 Component Diagram

FDA Team

FDA TEAM

4. Implementation

1

Skin pixel
classification

2

Connectivity
analysis

3

Skin region
identified is
a face or
not


4.3 Face Detection Algorithm

FDA TEAM

4. Implementation


Algorithm model process

FDA Team

Image
original

Convert
from RGB
to
HSV

Test
H and
V value of
each
pixel

Using
Threshold

Use
8
connected
neighbor to
find
different
regions

Identify
region of
face

FDA TEAM

4. Implementation

Original image

FDA Team

Image convert to HSV

FDA TEAM

Image convert to HSV with
SoBel

Operator


Filter
Blobs

Draw edge around face

4. Implementation

Draw region found not

filter in HSV image

FDA Team

FDA TEAM

4. Implementation

Binary Matrix

FDA Team

Histogram of image color

All region’s information

Face detected in original
image

FDA TEAM

4. Implementation

4.4 Compare with other software


FDA Team

Test

sample



Size: 42 images
-

121 faces


14 images with 1 faces


13 images with 2 faces


15 images with more than 2 faces


Includes all kind of face: tilt head, obscure by other objects, half of face; in every
kinds of light conditions; from low to high quality.

Result:


Because FDA uses skin color to detect face, we can detect exactly above
70% of test sample with diversity faces. Other software dependent on eyes
so detection's result is above 40%


Also because of that reason, FDA’s wrong ratio above 15% when its
confusion with other skin area. While other software’s wrong ratio about 10%




Test sample result


0
10
20
30
40
50
60
70
80
90
FDA
OpenCV
Neurotec
Exactly
Wrong
FDA TEAM

5. Conclusion

5.1 Advantages & Disadvantages



Advantages


Can handle High Definition Image


Completely open source, can develop in many ways.


Algorithm is fast and can be used in real
-
time applications.


Can detect all natural images under uncontrolled conditions.



Disadvantages


Black and white image


cannot detect skin


Contour distinguish


Confusion of human skin


Confusion of face form


FDA Team

FDA TEAM

5. Conclusion

5.2 Implemented Technical Problems


Recently, threshold to detect face doesn’t has any research can
perfectly detecting all faces.


Convert HSV can’t filter to remove all blobs.


Detect all skin area but can’t distinguish where that area contains eyes
or not.

5.3 Solutions


Need more time to research about algorithm.



FDA Team

Cloud
computing

Using
sample of
eyes

Low
performance

Face
detect

Wrong
detection

Calculate
edge
information

FDA TEAM

5. Conclusion

Maintainability:

Smart software like
Neural network

Performance:

Cloud

computing

Availability:

Code in C,
C++

Reliability:

Collect eyes
sample

FDA TEAM

Demo and Test

Demo FDA

FDA Team

FDA TEAM

Q&A

Question & Answer

FDA Team

FDA TEAM

LOGO

FDA Team