Discrete Finger and Palmar Feature Extraction for Personal Authentication Junta Doi, Member, IEEE,and Masaaki Yamanaka Advisor:Wen-Shiung Chen Student:

Discrete Finger and Palmar Feature Extraction

for Personal Authentication Junta Doi, Member, IEEE ,and Masaaki Yamanaka

Advisor:Wen-Shiung Chen

Student: Min-Chao Chang

2

Outline

IntroductionImage acquisitionFeature Point DefinitionFeature Extraction & MatchingConclusion

3

Introduction

Biometrics Physiological traits Behavioral traits

finger geometry observation Palmar flexion crease Hand anatomy

Hand geometry is considered to achieve medium security.

4

Introduction

Advantages No time-consumptive image analysis Noncontact Real-time Reliable feature extraction Easily combinable with other traits

5

Image Acquisition

Device: using a monochrome and/or color

video camera Resolution: not require for the faster response /

major crease detection Propose: the palm is placed freely toward the

video camera in front of a low-reflective plate

6

Image Acquisition

Schematic photograph of the palm image acquisition device

7

Image Acquisition

Finger alignment: use an image of the finger-close-together without bending

Enhance creases:

1. By CCD camera with polarizing filter

2. Lighting from a direction of 45 degrees

wrist side

3. near infrared CCD camera

8

Image Acquisition

Image quality: VGA of 640X480, 8bit gray levels the number of palm images is about 500, corrected

from about 50 subjects Noise reduction : use repetitive morphological

operations of erosions and dilations

9

Feature Point Definition

Intersection points with circles

10

Feature Point Definition

Illustration of tangential line at intersection points

11

Feature Point Definition

The way of extract the skeletal line skeletonization thinning algorithm

The way of search the intersection points two dimensional matrix operator

12

Feature Extraction

Finger Spreading and Skeletal Lines

a. the middle finger skeletal axis remains

unchanged

b. when fingers are bring together, the

skeletal lines deviate little

Feature extraction at the intersection points on skeletal lines

13

Feature Extraction

Comparison of intersection points when fingers are spread apart And brought together

14

Feature Extraction

Comparison of each finger skeletal line when fingers are spread apart (white lines) and wider apart (black lines).

15

Feature Extraction

Orientations at the intersection points Examples of detected orientations at the intersection points

16

Feature Extraction

missing points or additional points on the

extended skeletal line in the palm region

may occur in the new entry the middle finger matching is found to be the

most reliable among the four

17

Feature Matching Using Skeletal Lines

For the palm, it consists of the intersection points of the major palmar flexion creases or prominent creases, which are typically three palmar creases , on the extended skeletal line of each finger and also the orientations at the intersection points

18

Feature Matching Using Skeletal Lines

The first feature vector( in middle finger) Distal Middle Proximal

The second, third and fourth feature vector Forefinger Ring finger Little finger

19

Feature Matching Using A Mesh

a mesh is proposed and constructed by connecting laterally the corresponding intersection points on the adjacent skeletal lines

20

Feature Matching Using A Mesh

Each lateral line to line distance depends on the width of the finger

The over all lateral line distances depend on the palm width

All the widths and lengths are personal and are combined with the oriented palmar intersection points

21

Feature Matching Using A Mesh

Mesh Matching for Authentication the middle finger skeleton is selected to align

the meshes for the enrolled and the new Some deviates is caused by a palm image

variation due to the palm bending, though all the fingers are brought together

Compare of the enrolled and the new of the same palm

22

Feature Matching Using A Mesh

Compare of mashes for different palms

23

Feature Matching Using A Mesh

The mesh deviation between the two, is evaluated by calculating the root mean square deviation (rmsd) value.

δi is the positional difference at each mesh point

N is the total number of the mesh points to be compared

The magnitude of the difference is measured in pixels and thereafter normalized by the parameters of the finger length and the palm width

24

Feature Matching Using A Mesh

Rings and Mesh Points The ring wear has little effect on the feature

matching, if it is limited in size and number

25

Feature Matching Using A Mesh

“finger-brought-together” image instead of the pegs

“stretched-or-straightened” image instead of the flat bottom plate

the bending is not so fatal, if it is urged to stretch or straighten out

26

Results

Database ： 50 users

Each user’s hand ： 10 images were captured (total of 500 images ).

27

Feature Matching Using A Mesh

Genuine and imposter rmsd distribution

28

Conclusion

Our matching is multistaged: the first stage is matching for the authentication the second stage is based on four-finger procedure as a

usual matching the third stage is based on more detailed geometric

parameters such as the shape factors of each finger section or the palm

29

Conclusion

This point-based matching brings about a robust and real-time processing of less than one second

The “brought-together fingers” and “stretched-and-straightened-out palm” are our instructions to the user

this noncontacting personal feature extraction method will easily in combination with the hand geometry, palm vascular pattern, and/or facial processing

30

References

A. K. Jain and A. Ross, “A prototype hand geometry-based verification system,” in Proc. 2nd Int. Conf. Audio- and Video-based Biometric Personal Authentication (AVBPA), 1999, pp. 166–171

N. Duta, A. K. Jain, and K. Mardia, “Matching of palmprint,” Pattern Recognit. Lett., vol. 23, pp. 477–485, 2002

R. Sanchez-Reillo, C. Sanchez-Avila, and A. Gonzalez-Marcos, “Biometric identification threou hand geometry measurements,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 10, pp. 1168–1171, Oct. 2000