Visual Face Recognition Fusion

brasscoffeeΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 7 μήνες)

43 εμφανίσεις

1

Real
-
Time Efficient Parallel Thermal and

Visual Face Recognition Fusion


2009/12/24


陳冠宇

2

Outline


I
ntroduction


G
abor

F
iltering
F
or

F
ace Recognition


-
Feature point calculation

selection

Feature vector generation


-
Similarity calculations


Parallel Architecture For Face Recognition


Limitations And Benefits


Conclusions


3

I
ntroduction


Computer vision has long fascinated

applications in psychology, neural science,
computer

science, and engineering.



A simple feature

extraction

algorithm may
require thousands of basic operations

per
pixel.


As you can see, parallel

computing is essential
to solving such a problem.


4

I
ntroduction


This paper would discuss Task Parallel
processing for fast face recognition system
based on Gabor Filtering technique.

5

Related work in Face Recognition


Images taken from visual band are formed due to


reflectance.



Recently, face recognition on thermal/infrared

spectrum has gained popularity because thermal

images are formed due to emission not reflection
.



6

Related work in Face Recognition




Some of the commonly used face recognition
techniques are Principal Component Analysis
(PCA) , Linear Discriminate Analysis (LDA)
and Gabor Filtering technique.

7

G
abor

F
iltering
F
or

F
ace Recognition


1.Feature point calculation

For point (X, Y), filter response denoted as

R is defined as

8

Feature point calculation

Where
σ
X

and
σ
Y

are the standard deviation of the Gaussian
envelop along the x and y dimensions respectively.

λ
,
θ

and
n

are the wavelength, orientation and no of

orientations respectively.

9

2.Feature point selection

In a particular window of size SxT around

which the behavior or response of Gabor filter

kernel is maximum, as feature point.


10

Feature point selection

Feature point located at any point can be evaluated as

Where
R
j

is the response of the image to the jth Gabor

filter and C is any window.

11

3.Feature vector generation

Feature vectors are generated at feature points as

discussed in previous sections. p
th

feature vector

of i
th

reference face is defined as:

12

Decision Fusion Architectures

where W
v

and W
T

denote weight factors for the

matching scores of visual and thermal modules.


In this paper, W
v
=W
T
=0.5

13

14

Parallel Architecture For Face Recognition


As same face recognition steps are repeated for

visual, thermal and fused image. So it is proposed

that three individual face recognition processes for

each data be carried out on different slave computers.

15

Parallel Architecture For Face Recognition

16

Limitations And Benefits


Complexity


Resource Requirements


Speedups


Portability

17

Conclusions

18

Conclusions

This paper briefly described a parallel design

framework for efficient and real
-
time face

recognition system. It defines new frontiers for

fast and efficient recognition system.


With our design framework, the realtime performance

can be achieved on regular computers,such as those

found in a student cluster.

19

Thank you