13-07-2012, 04:03 PM
Two-Stage Enhancement Scheme for Low-Quality Fingerprint Images by Learning From the Images
[attachment=27581]
INTRODUCTION
BIOMETRICS, i.e., described as the science of recognizing
an individual based on his or her physical or behavioral
traits, is beginning to gain acceptance as a legitimate method
for the determination of an individual’s identity [1]. Biometric
systems have now been deployed in various commercial, civilian,
and forensic applications as a means of establishing identity
Manuscript received April 28, 2011; revised July 4, 2011 and September 12,
2011; accepted October 5, 2011. This work was supported in part by the National
Natural Science Foundation of China under Grant 61063035. This paper
was recommended by Associate Editor E. R. Weippl.
J. Yang is with the Institute of Computer Science and Information Engineering,
Tianjin University of Science and Technology, Tianjin 300222,
China and also with the School of Information Technology, Jiangxi University
of Finance and Economics, Nanchang 330013, China (e-mail:
yangjucheng[at]hotmail.com).
RELATED WORKS ON FINGERPRINT ENHANCEMENT
Human experts routinely use the context information of fingerprint
images, such as ridge continuity and regularity to help
in identifying them. This means that the underlying morphogenetic
process that produced the ridges does not allow for
irregular breaks in the ridges except at ridge endings. Because
of the regularity and continuity properties of the fingerprint image,
occluded and corrupted regions can be recovered using the
contextual information from the surrounding area. Such regions
are regarded as “recoverable” regions. The efficiency of an automated
enhancement algorithm depends on the extent to which
they utilize the contextual information. Some filters for these
enhancement tasks are classified either in the spatial domain
or in the frequency domain. According to the classification of
the filters, the existing enhancement processing is roughly classified
into either spatial-domain filtering or frequency-domain
filtering.
PROPOSED TWO-STAGE ENHANCEMENT SCHEME
According to the analysis of the traditional prior works, the
existing enhancement algorithms are not always satisfactory
in enhancing low-quality fingerprint images. To overcome the
demerits of these methods, we propose a new and effective
scheme in this paper for the enhancement of the images using
two consecutive stages as shown in Fig. 1. The algorithm first
enhances the images in the spatial domain with a spatial ridgecompensation
filter and, then, enhances the images in the frequency
domain. The parameters (ridge direction and frequency)
for the frequency bandpass filters are estimated from the original
image and the first-stage enhanced image. The details are
introduced as follows.
EXPERIMENTAL RESULTS
The public fingerprint image databases of the Fingerprint
Verification Competition (FVC2000, FVC2002, FVC2004, and
FVC2006) [38] were established with the aim of providing
a benchmark to test state-of-the-art techniques in fingerprintrecognition
applications. The fingerprint images are well suited
for testing a contemporary (online) fingerprint system with
prints that are acquired using modern capacitive and optical
scanners. Focusing on the low-quality image, the FVC2004
databases contains more poor images than FVC2000 and
FVC2002 databases, and it is easier to access than FVC2006
databases. Therefore, the database that is used in this experiment
is the FVC2004 database set_a [39], which contains four distinct
subdatabases with four different scanners: DB1_a, DB2_a,
DB3_a, and DB4_a. The resolution of DB1_a, DB2_a, and
DB4_a is 500 dpi and that of DB3_a is 512 dpi. Each database
consists of 800 fingerprint images (i.e., there are 100 persons,
and each individual has eight fingerprints) in 256 gray-scale levels.
DB1_a and DB2_a are collected by the optical sensor, while
DB3_a is collected by thermal sweeping sensor and DB4_a
is synthetic fingerprint generation.
CONCLUSION AND FURTHER WORK
In this paper, an effective two-stage enhancement scheme
in both the spatial domain and the frequency domain for lowquality
fingerprint images by learning from the images has been
proposed. Emphasizing the enhancement of the low-quality images,
the first-stage enhancement scheme has been designed to
use the context information of the local ridges to connect or
separate the ridges. Based on this spatial filtering, the broken
ridges will be connected and the merged ridges will be separated
effectively; thus, the fingerprint ridges can be remedied
and recovered well. In the second-stage processing, the filter
is separable in the radial and angular domains, respectively.
[attachment=27581]
INTRODUCTION
BIOMETRICS, i.e., described as the science of recognizing
an individual based on his or her physical or behavioral
traits, is beginning to gain acceptance as a legitimate method
for the determination of an individual’s identity [1]. Biometric
systems have now been deployed in various commercial, civilian,
and forensic applications as a means of establishing identity
Manuscript received April 28, 2011; revised July 4, 2011 and September 12,
2011; accepted October 5, 2011. This work was supported in part by the National
Natural Science Foundation of China under Grant 61063035. This paper
was recommended by Associate Editor E. R. Weippl.
J. Yang is with the Institute of Computer Science and Information Engineering,
Tianjin University of Science and Technology, Tianjin 300222,
China and also with the School of Information Technology, Jiangxi University
of Finance and Economics, Nanchang 330013, China (e-mail:
yangjucheng[at]hotmail.com).
RELATED WORKS ON FINGERPRINT ENHANCEMENT
Human experts routinely use the context information of fingerprint
images, such as ridge continuity and regularity to help
in identifying them. This means that the underlying morphogenetic
process that produced the ridges does not allow for
irregular breaks in the ridges except at ridge endings. Because
of the regularity and continuity properties of the fingerprint image,
occluded and corrupted regions can be recovered using the
contextual information from the surrounding area. Such regions
are regarded as “recoverable” regions. The efficiency of an automated
enhancement algorithm depends on the extent to which
they utilize the contextual information. Some filters for these
enhancement tasks are classified either in the spatial domain
or in the frequency domain. According to the classification of
the filters, the existing enhancement processing is roughly classified
into either spatial-domain filtering or frequency-domain
filtering.
PROPOSED TWO-STAGE ENHANCEMENT SCHEME
According to the analysis of the traditional prior works, the
existing enhancement algorithms are not always satisfactory
in enhancing low-quality fingerprint images. To overcome the
demerits of these methods, we propose a new and effective
scheme in this paper for the enhancement of the images using
two consecutive stages as shown in Fig. 1. The algorithm first
enhances the images in the spatial domain with a spatial ridgecompensation
filter and, then, enhances the images in the frequency
domain. The parameters (ridge direction and frequency)
for the frequency bandpass filters are estimated from the original
image and the first-stage enhanced image. The details are
introduced as follows.
EXPERIMENTAL RESULTS
The public fingerprint image databases of the Fingerprint
Verification Competition (FVC2000, FVC2002, FVC2004, and
FVC2006) [38] were established with the aim of providing
a benchmark to test state-of-the-art techniques in fingerprintrecognition
applications. The fingerprint images are well suited
for testing a contemporary (online) fingerprint system with
prints that are acquired using modern capacitive and optical
scanners. Focusing on the low-quality image, the FVC2004
databases contains more poor images than FVC2000 and
FVC2002 databases, and it is easier to access than FVC2006
databases. Therefore, the database that is used in this experiment
is the FVC2004 database set_a [39], which contains four distinct
subdatabases with four different scanners: DB1_a, DB2_a,
DB3_a, and DB4_a. The resolution of DB1_a, DB2_a, and
DB4_a is 500 dpi and that of DB3_a is 512 dpi. Each database
consists of 800 fingerprint images (i.e., there are 100 persons,
and each individual has eight fingerprints) in 256 gray-scale levels.
DB1_a and DB2_a are collected by the optical sensor, while
DB3_a is collected by thermal sweeping sensor and DB4_a
is synthetic fingerprint generation.
CONCLUSION AND FURTHER WORK
In this paper, an effective two-stage enhancement scheme
in both the spatial domain and the frequency domain for lowquality
fingerprint images by learning from the images has been
proposed. Emphasizing the enhancement of the low-quality images,
the first-stage enhancement scheme has been designed to
use the context information of the local ridges to connect or
separate the ridges. Based on this spatial filtering, the broken
ridges will be connected and the merged ridges will be separated
effectively; thus, the fingerprint ridges can be remedied
and recovered well. In the second-stage processing, the filter
is separable in the radial and angular domains, respectively.