Ａｂｓｔｒａｃｔ
Ａｂｓｔｒａｃｔ
Ｗｉｔｈ
ｔｈｅ
ａｄｖｅｎｔ ｏｆ
ｔｈｅ
Ｉｎｔｅｍｅｔ
Ｅｒａ．Ｂｉｏｍｅｔｒｉｃ
ｉｄｅｎｔｉｆｉｃａｔｉｏｎ
ｈａｓ
ａｎｒａｃｔｅｄ
ｌｏｔｓ
ｏｆ
ａｔｔｅｎｔｉｏｎ
ａｎｄ
ｍａｄｅ
ｇｒｅａｔ
ｐｒｏｇｒｅｓｓ
ｓｉｎｃｅ
ｔｈｅ
８０ｓ．Ｆｉｎｇｅ印ｒｉｎｔ，ｏｎｅ
ｏｆ
ｔｈｅ
ｆｏｃｕｓｅｓ
ｉｎ
ｔｈｅ
ｐｅｒｓｏｎａｌ
ｉｄｅｎｔｉｆｌｃａｔｉｏｎ
ｆｉｅｌｄ，ｉｓ
ｔｈｅ
ｈｏｔｓｐｏｔ
ａｎｄ
ｄｉｍｃｕｌｔｙ
ｔｏ
ｒｅｓｅａｒｃｈｅｒｓ
ａｔ
ｈｏｍｅ
ａｎｄ
ａｂｒｏａｄ
ｆｏｒ
ａ
ｌｏｎｇ
ｔｉｍｅ．Ｃｕｒｒｅｎｔｌｙ＇ａｎ
ｉｎｃｒｅａｓｉｎｇ
ｎｕｍｂｅｒ
ｏｆ
ａｕｔｏｍａｔｉｃ
ｆｉｇｅ印ｒｉｎｔ
ｉｄｅｎｔｉｆｉｃａｔｉｏｎ
ｓｙｓｔｅｍｓ（ＡＦＩＳ）ｈａＶｅ
ｂｅｅｎ
ｄｅＶｅｌｏｐｅｄ
ａｎｄ
ｅｘｐａｎｄｅｄ
ｉｎ
ｐｅｏｐｌｅ’ｓ
ｄａｉｌｙ
ｌｉｆｅ，
ｐｒｏｖｉｄｉｎｇ
ｉｄｅｎｔｉｔｙ
ａｕｍｅｎｔｉｃａｔｉｏｎ
ａｎｄ
ｓｅｃｕｒｉｔｙ
ｐｒｏｔｅｃｔｉｏｎ．ＡｌｔＩｌｏｕｇｈ
ｔｈｅ
ｔｈｅｏｒｅｔｉｃａｌ
ｒｅｓｅａｒｃｈ
ｈａｓ
ｍａｄｅ
ｓｉｇｎｉｆｉｃａｎｔ
ａｃｈｉｖｅｍｅｎｔｓ，ＡＦＩＳ
ｂａＳｅｄ
ｏｎ ａ
１２ｕｒｇｅ—ｓｃａｌｅ
ｆｉｎｇｅｒｐｒｉｎｔ
ｄａｔａｂａｓｅ
ｉｓ
ｓｔｉｌｌ
ｔｏ
ｂｅ胁ｈｅｒ
ｉｍｐｒｏｖｅｄ
ｏｎ
ａｃｃｕｒａｃｙ
ａｎｄ
ｅｆ五ｃｉｅｎｃｙ．Ｃｏｍｂｉｎｅｄ
ｗｉｔｈ
ｔｈｅ
ｃｕｒｒｅｎｔ
ｄｅＶｅｌｏｐｍｅｎｔ
ｏｆ
ｄｉｇｉｔａｌｉｍａｇｅ
ｐｒｏｃｅｓｓｉｎｇ
ａＩｌｄ
ｐａｔｔｅｍ
ｒｅｃｏｇＩｌｉｔｉｏｎ
ｔｅｃｌｌｎｉｑｕｅｓ，
ｔｈｉｓ
ｔｈｅｓｉｓ
ｒｅｓｅａｒｃｈｅｄ
ｓｅｖｅｒａｌ
ｋｅｙ
ａｌｇｏｒｉｔｌｌｍｓ
ｂａｓｅｄ
ｏｎ
ｌａｒｇｅ－ｓｃａｌｅ
ｆｉｎｇｅ叩ｒｉｎｔ
ｄａｔａｂａｓｅ，
ｍａｉｎｌｙ
ｆｏｃｕｓｅｄ
ｏｎ
ｔｈｅ
ｏｐｔｉｍｉｚａｔｉｏｎ
ｏｆ
ｔｈｅ
ｆｉｎｇｅ印ｒｉｎｔ
ｉｍａｇｅ
ｐｒｅｐｒｏｃｅｓｓｉｎｇ，ｆｉｎｇｅ平ｒ主ｎｔ
ｃｌａＳｓｍｃａｔｉｏｎ
ａＪｌｄ
ｆｉｎｇｅ印ｒｉｎｔ
ｉｄｅｎｔｉｆｉｃａｔｉｏｎ．Ｔｈｅ
ｍａｉｎ
ｃｏｎｔｒｉｂｕｔｉｏｎｓ
ｏｆ
ｔｈｅ
ｄｉｓｓｅｒｔａｔｉｏｎ
ａｒｅ
ａＳ
ｆｏｌｌｏｗｉｎｇ：
１．Ｄｕｅ
ｔｏ
ｄｉ蔬ｒｅｎｔ
ｋｉｎｄｓ
ｏｆ
ｎｏｉｓｅｓ，ｔｈｅ
ｆｉｎｇｅ印ｒｉｎｔ
ｉｍａｇｅ
ｐｒｅｒｏｃｅｓｓｉｎｇ
ｉｓ
Ｖｅｒｙ
ｉｍｐｏｒｔａｌｌｔ
ｔｏ
ｉｍｐｒｏＶｅ
ｔｈｅ
ｑｕａｌｉｔｙ
ｏｆ
ｔｈｅ
ｃｏｌｌｅｃｔｅｄ
ｉｍａｇｅｓ．０ｎ
ｔｈｅ
ｂａＳｉｓ
ｏｆ
ｍｅ
ａｎａｌｙｓｉｓ
ｏｆ
ｃｕｒｒｅｎｔ
ｆｉｎｇｅｒｐｒｉｎｔ
ｉｄｅｎｔｉｆｉｃａｔｉｏｎ
ｔｅｃｈｎｏｌｏｇｙ
ｄｉｓｓｅｒｔａｔｉｏｎｓ，ｔｈｉｓ
ｔｈｅｓｉｓ
ｒｅａｌｉｚｅｓ
ｔｈｅ
ｋｅｙ
ｓｔｅｐｓ
ｏｆ
ｔｈｅ
ｆＩｎｇｅｒｐｍｔ
ｉｍａｇｅ
ｐｒｅｐｒｏｃｅｓｓｉｎｇ．Ｉｎ
ｔ１１ｅ
ｒｅｇａＬｒｄ，ａ
ｓｅｒｉｅｓ
ｏｆ
ａ１９０ｒｉｔｈｍ
ｍｏｄｉｆｉｃａｔｉｏｎｓ
ａｂｏｕｔ
ｈｉｓｔｏｇｒａｍ
ｅｑｕａｌｉｚａｔｉｏｎ，ｏｒｉｅｎｔａｔｉｏｎ
ｆｉｅｌｄ
ｃａｌｃｕｌａｔｉｏｎ
ａＩｌｄ
ｉｍａｇｅ
ｅＩｌｌｌａｎｃｅｍｅｎｔ
ｉｓ
ｐｒｏｐｏｓｅｄ．Ｔｈｉｓ
ａｌｇｏｒｉｔ№ｓｉｇｎｉｆｉｃａｎｔｌｙ
ｅｎ】１ａｎｃｅｓ
ｔｈｅ
ｃｏｎｔｒａｓｔ
ｏｆ
ｔｌｌｅ
ｒｉｄｇｅ
ａｎｄ
ｔｈｅ
Ｖａｌｌｅｙ
ｌｉｎｅ，ｅｎｓｕｒｅ
ｔｈｅ
ａｃｃｕｒａｃｙ
ａｎｄ
ｒｅｌｉａｂｉｌｉｔｙ
ｏｆ
ｔｈｅ
ｄｅｔａｉｌｓ
ｏｆ
ｔ１１ｅ
ｆ．ｅａｔｕｒｅ
ｅｘｔｒａｃｔｉｏｎ．
２．Ｕｓｕａｌｌｙ
ｔｈｅ
ｃｌａＳｓｉｃ
ｆｉｎｇｅｒｐｒｉｎｔ
ｃｌａＳｓｉｆｉｃａｔｉｏｎ
ａｌｇｏｒｉｔｈｍ
ｎｅｅｄｓ
ｔｏ
ｅｘｔｒａｃｔ
ｃｏｒｅ
ａｎｄ
ｄｅｌｔａ
ｐｏｉｎｔｓ．Ｂｕｔ
ｉｎ
ｐｒａｃｔｉｃａｌ
ａｐｐｌｉｃａｔｉｏｎ，ｉｔ’ｓ
ｈｉｇｈｌｙ
ｐｏｓｓｉｂｌｅ
ｔｈａｔ
ｄｅｌｔａ
ｐｏｉｎｔｓ
ｄｏｎ’ｔ
ａｐｐｅａｒ
ｉｎ
ｔｈｅ
ｃｏｌｌｅｃｔｅｄ
ｉｍａｇｅ，ｓｏ
ｅｘｔｒａｃｔｉｎｇ
ｍｏｒｅ
ｃｌａｓｓｉｆｉｃａｔｉｏｎ
ｆｅａｔｕｒｅｓ
ｔｏ
ｍａｌ（ｅ
ｕｐ
ｆｏｒ
ｔｈｅ
ｄｅｆｅｃｔ
ｗｉｌｌ
ｂｅ
ｖｅｒｙ
ｎｅｃｅｓｓａｒｙ。Ａ
ｎｅｗ印ｐｒｏａｃｈ
ｗｈｉｃｈ
ｃｏｍｂｉｎｅｓ
ｔｈｅ
ｍｅｄｉａｌｌ
ｌｉｎｅ
ｆｉｔｔｉｎｇ
ｗｉｔｈ
ｔｌｌｅ
ｍｉｎｉｍｕｍ
ｂｏｕｎｄｉｎｇ
ｅｌｌｉｐｓｅ
ｍｅｔｈｏｄ
ｉｓ
ｐｒｅｓｅｎｔｅｄ
ｔｏ
ｅｘｔｒａｃｔ
ｆｉｎｇｅｒ
ｐｌａｃｅｍｅｎｔ
ｄｉｒｅｃｔｉｏｎ．Ａｍｎｅ
ｔｒａｎｓｆｏｍａｔｉｏｎ
ｉｓ
ｕｓｅｄ
ｆｏｒ
ｒｏｔａｔｉｏｎａｌ
ｐｒｏｃｅｓｓｉｎｇ
ａＲｅｒ
ｃａｌｃｕｌａｔｉｎｇ
ｔｈｅ
ｃｏｍｐｅｎｓａｔｉｏｎ
ｏｆ
ａ１１９ｌｅ．Ｉｎ
ｔｈｅ
ａＳｐｅｃｔ
ｏｆ
ｆｉｎｇｅ巾ｒｉｎｔ
ｆｅａｔｕｒｅ
ｅｘｔｒａｃｔｉｏｎ，
ｔｗｏ—ｓｔｅｐｓ
ｐｏｓｉｔｏｌｌｉｎｇ
ｍｅｔｈｏｄ
ｔｏ
ｅｘｔｒａｃｔｉｎｇ
ｃｏｒｅ
ｐｏｉｎｔｓ
ｈａＳ
ｂｅｅｎ
ｓｔｕｄｉｅｄ，ｗｈｉｃｈ
ｉｓ
ｏｎ
ｔｈｅ
ｂａＳｉｓ
ｏｆ
ｔｈｅ
ｏｒｉｅｎｔａｔｉｏｎ
６ｅｌｄ
ｃｏｎｓｉｓｉｔｅｎｃｖ
ｃｏｅｆｆｉｃｉｅｎｔ
ａ１１ｄ
ｍｏｄｉ６ｅｄ
Ｐｏｉｎｃａｒｅ
ｉｎｄｅｘ
ａｌ罩Ｆｏｒｉｔｈｎｌ．Ｅ；ｅｃａｕｓｅ
ｏｆ
ｔｈｅ
ｐｓｅｕｄｏ
ｐｏｉｎｔｓ，ｔｈｉｓ
ｔｌｌｅｓｉｓ
ｈａｓ
ｐｒｏｐｏｓｅｄ
ａ
ｓｏｕｎｄ
ｓｅｔ
ｏｆ
ｍｅｔｈｏｄｓ
ｂａｓｅｄ
ｏｎ
ｈ啪ａＩｌ
ｅｘｐｅｒｉｅｎｃｅ
ｔｏ
ｈｅｌｐ
ｅｎｈａｎｃｅ
ｔｈｅ
ｃｌａｓｓｉｆｉｃａｔｉｏｎ
ａｃｃｕｒａｃｙ．
Ｆｉｎａｌｌｙ＇
ａ
ｔｈｒｅｅ．１ｅｖｅｌ
ｆｉｎ２ｅｒｐｒｉｎｔ
ｃｌａｓｓｉｆｉｃａｔｉｏｎ
ｓｙｓｔｅｍ
ｂａＳｅｄ
ｏｎ
ｔｈｅ
ｎ啪ｂｅｒｓ，
ｏｒｉｅｎｔａｔｉｏｎ，ｗｈｏｒｌ‘ｓ
ｒａｄｉｕｓ
ａ１１ｄ
ｔｈｅ
ａｒｃｈ’ｓ
ｄｉｒｅｃｔｉｏｎ
ｏｆ
ｍｅ
ｃｏｒｅ
ｐｏｉｎｔ，ｉｓ
ｐｒｅｓｅｎｔｅｄ．Ｉｔ
ｃａｎ
ｂｅ ｕｓｅｄ
ｔｏ
ｉｍｐｌｅｍｅｎｔ
ｔｈｅ
ｃｌａｓｓｉｆｉｃａｔｉｏｎ
ｏｎ
ｔｈｅ
ｌａｕｒｇｅ．ｓｃａｌｅ
ｄａｔａｂａｓｅ．
３．Ｉｎ
ｏｒｄｅｒ
ｔｏ
ｏｐｔｉｍｉｚｅ
ｍｅ
ｃｏｍｍｏｎｌｙ
ｆｉｎｇｅｒｐｒｉｎｔ
ｉｄｅｎｔｉｆｉｃａｔｉｏｎ
ａｌｇｏｒｉｔ№ｓ
ｂａｓｅｄ
ｏｎ
ｔｈｅ
ｍｉｎｕｔｉａｅ
ｍａｔｃｌｌｉｎｇ，ａ
ｔｗｏ—ｌｅＶｅｌ
ｆｉｎｇｅ印ｒｉｎｔ
ｍａｔＣｈｉｎｇ
ａｌｇｏｒｉｔｌｌＩＩｌ
ｉｓ
ｄｅＶｅｌｏｐｅｄ．
Ｆｉｒｓｔ．ａ
ｄｉｓｃｒｉｍ
ｉｎａｔｉｖｅ
ｓｕｂ．ｓｔｍｃｔｕｒｅ
ｉｓ
ｄｅｆｉｎｅｄ
ｆ－ｏｒ
ｅａｃｈ
ｍｉｎｕｔｉａ，、Ｖｈｉｃｈ
ｉｎｃｌｕｄｅｓ
ａ
ｅｘｔｅｍａｌ
ｃｏｎｇｍｅｎｔ
ｔｒｉａｎｇｌｅ
ａＩｌｄ
ａ
ｔｒｉａｎｇｌｅ
ｍａｄｅ
ｕｐ
ｏｆ
ｔｈｅ
ｎｅｉｇｈｂｏｒ
ｍｉｎｕｔｉａｅｓ．Ｔｈｅｎ
ａ
ｓｉｍｉｌａｒｉｔｙ
ｓｃｏｒｅ
ｉｓ
ｅｍｐｌｏｙｅｄ
ｔｏ
ｄｅｃｉｄｅ
ｗｈｅｔｈｅｒ
ｉｔ
ｉｓ
ｎｅｃｅｓｓａ叫ｔｏ
ｇｅｔ
ｔｏ
ｔｈｅ
ｇｌｏｂａｌ
ｍａｔｃｈｉｎｇ．Ｉｆ
ｔｈｅ
ｓｉｍｉｌａｒｉｔｙ
ｓｃｏｒｅ
ｄｏｅｓｎ’ｔ
ｒｅａｃｈ
ａ
ｃｅｒｔａｉｎ
ｔｂｒｅｓｈ０１ｄ，ｔｈｅ
ｐｏｔｅｎｔｉａｌ
ｉｍａｇｅｓ
ａｎｄ
ｃａｎｄｉｄａｔｅ
ｒｅｆＩｅｒｅｎｃｅ
ｐａｉｒｓ
ｗｈｉｃｈ
ｈａｖｅ
ｇｏｔｔｅｎ
ｈｉｇｈｅｒ
ｓｃｏｒｅｓ
ｗｉｌｌ
ｂｅ ｕｓｅｄ
ｉｎ
ｔｈｅ
Abstract
Abstract
With the advent
of
the Internet Era, Biometric identification has attracted lots
of
attention and made great progress since the 80s. Fingerprint, one
of
the focuses in the
personal identification field,
is
the hotspot and difficulty to researchers at home and
abroad for a long time. Currently, an increasing number
of
automatic figerprint
identification systems (AFIS) have been developed and expanded in people's daily life,
providing identity authentication and security protection. Although the theoretical
research has made significant achivements, AFIS based on a large-scale fingerprint
database is still to be further improved on accuracy and efficiency. Combined with the
current development
of
digital image processing and pattern recognition techniques,
this thesis researched several key algorithms based on large-scale fingerprint database,
mainly focused on the optimization
of
the fingerprint image preprocessing, fingerprint
classification and fingerprint identification. The main contributions
of
the dissertation
are as following:
I. Due to different kinds
of
noises, the fingerprint image prerocessing is very
important to improve the quality
of
the collected images. On the basis
of
the analysis
of
current fingerprint identification technology dissertations, this thesis realizes the
key steps
of
the fingerpmt image preprocessing. In the regard, a series
of
algorithm
modifications about histogram equalization, orientation field calculation and image
enhancement is proposed. This algorithm significantly enhances the contrast
of
the
ridge and the valley line, ensure the accuracy and reliability
of
the details
of
the
feature extraction.
2.
Usually the classic fingerprint classification algorithm needs
to
extract core
and delta points. But in practical application, it's highly possible that delta points don't
appear in the collected image,
so
extracting more classification features to make up
for the defect will be very necessary. A new approach which combines the median line
fitting with the minimum bounding ellipse method
is
presented to extract finger
placement direction. Affine transformation is used for rotational processing after
calculating the compensation
of
angle. In the aspect
of
fingerprint feature extraction,
two-steps positoning method
to
extracting core points has been studied, which is on
the basis
of
the orientation field consisitency coefficient and modified Poincare index
algorithm. Because
of
the pseudo points, this thesis has proposed a sound set
of
methods based on human experience to help enhance the classification accuracy.
Finally, a three-level fingerprint classification system based on the numbers,
orientation, whorl's radius and the arch's direction
of
the core point, is presented. It
can be used to implement the classification on the large-scale database.
3.
In
order to optimize the commonly fingerprint identification algorithms based
on the minutiae matching, a two-level fmgerprint matching algorithm is developed.
First, a discrim inative sub-structure is defined for each minutia, which includes a
external congruent triangle and a triangle made up
of
the neighbor minutiaes. Then a
similarity score
is
employed to decide whether it is necessary to get to the global
matching.
If
the similarity score doesn't reach a certain threshold, the potential images
and candidate reference pairs which have gotten higher scores will be used in the
II
评论0