(fall 2012) compression vs performance in fingerprint images

COMPRESSION VS. PERFORMANCE IN FINGERPRINT IMAGES The purpose of this study is to evaluate the performance of fingerprint image matching based on various compression levels. The fingerprint images were taken from a Purdue University database and each were compressed to various Quality of Compression levels (QoC 99, QoC 90, QoC 80, QoC 70, QoC 50, QoC 25, QoC 15, QoC 10, QoC 5, QoC 1) using XnConvert. The compressed images were then run through Megamatcher and Oxford Wave to determine the Equal Error Rate (EER). Faisal Al-Khalidi, Troy Davis, Christopher Guntz, Michael Brockly, Stephen Elliott Overview Methodology • Gather right index fingerprint images from a thermal swipe sensor • Compress the images to ten different quality levels formatted as .jpg • Run each compressed image set through Megamatcher to obtain genuine and impostor match scores at each compression level • Process the data in Oxford Wave to produce readable performance graphs and obtain EER’s • Analyze the results from the graphs Compression Sizes EERs of Compression Levels Fingerprint Comparison Observations • We noticed that the EER, our measure of performance, varied between each compression level. There was little observable trend between compression level and EER, as the EER values varied between 0.1861% and 0.2289% across all compression levels. • Uncompressed had the largest file size and QoC1 had the smallest file size • Unexpectedly, the results do not appear to be linear. One example is that QoC15 has a lower EER than QoC99. This may suggest that the average file size is not the best predictor of performance. • Furthermore, the EER range varied only 0.0698%. This suggests that compression may not have the impact that literature we reviewed suggested. Possible reasons could be that the compression method chosen does create much fragmenting or distortion when compression is applied *EER: Equal Error Rate: The acceptance rate at which False Acceptance Rate and False Rejection Rate are equal. A lower EER is considered to have better performance QoC 100 QoC 99 QoC 90 QoC 80 QoC 70 QoC 50 QoC 25 QoC 15 QoC 10 QoC 5 QoC 1 EER at QoC 0.1591 0.216 0.1882 0.1727 0.2124 0.2195 0.1963 0.1861 0.2093 0.2223 0.2289 0.12 0.14 0.16 0.18 0.2 0.22 0.24 Equal Error Rate EER at Quality of Compression

Upload: international-center-for-biometric-research

Post on 07-Aug-2015

52 views

Category:

Technology

1 download

Report

Download

Tags:

Embed Size (px):

TRANSCRIPT

COMPRESSION VS. PERFORMANCE IN FINGERPRINT IMAGES

The purpose of this study is to evaluate the performance of fingerprint image matching based on various compression levels. The fingerprint images were taken from a Purdue University database and each were compressed to various Quality of Compression levels (QoC 99, QoC 90, QoC 80, QoC 70, QoC 50, QoC 25, QoC 15, QoC 10, QoC 5, QoC 1) using XnConvert. The compressed images were then run through Megamatcher and Oxford Wave to determine the Equal Error Rate (EER).

Faisal Al-Khalidi, Troy Davis, Christopher Guntz, Michael Brockly, Stephen Elliott

Overview

Methodology • Gather right index fingerprint images from a thermal swipe

sensor • Compress the images to ten different quality levels

formatted as .jpg • Run each compressed image set through Megamatcher to

obtain genuine and impostor match scores at each compression level

• Process the data in Oxford Wave to produce readable performance graphs and obtain EER’s

• Analyze the results from the graphs

Compression Sizes EERs of Compression Levels

Fingerprint Comparison Observations • We noticed that the EER, our measure of performance, varied between each

compression level. There was little observable trend between compression level and EER, as the EER values varied between 0.1861% and 0.2289% across all compression levels.

• Uncompressed had the largest file size and QoC1 had the smallest file size • Unexpectedly, the results do not appear to be linear. One example is that QoC15

has a lower EER than QoC99. This may suggest that the average file size is not the best predictor of performance.

• Furthermore, the EER range varied only 0.0698%. This suggests that compression may not have the impact that literature we reviewed suggested. Possible reasons could be that the compression method chosen does create much fragmenting or distortion when compression is applied

*EER: Equal Error Rate: The acceptance rate at which False Acceptance Rate and False Rejection Rate are equal. A lower EER is considered to have better performance

QoC100 QoC 99 QoC 90 QoC 80 QoC 70 QoC 50 QoC 25 QoC 15 QoC 10 QoC 5 QoC 1

EER at QoC 0.1591 0.216 0.1882 0.1727 0.2124 0.2195 0.1963 0.1861 0.2093 0.2223 0.2289

0.12

0.14

0.16

0.18

0.2

0.22

0.24

Equa

l Err

or R