Rao, A Tirupathi and C, Krishna Mohan
(2016)
EFFICIENT APPROACHES FOR FINGERPRINT AND
PALMPRINT RECOGNITION.
PhD thesis, Indian institute of technology Hyderabad.
Abstract
Fingerprint and palmprint are the commonly used biometric traits among all the
biometrics. With the drastic improvement in technology, biometrics are making their
way to mobile handheld devices, with limited storage and computation power. So,
there is a need for recognition approaches that are efficient and optimal in resource
utilization. In this thesis, we propose techniques to improve the overall efficiency
and response time of fingerprint and palmprint recognition systems. To achieve this,
we propose various approaches targeting the different stages in a biometric recognition system, namely, pre-processing, feature extraction, and matching. In this thesis,
the following methods are proposed: 1) enhancements in pre-processing stage to improve the efficiency of extracting the minutiae points that increases the performance
of cross sensors fingerprint matching, 2) efficient fingerprint matching using hybrid fingerprint matching with k-nearest neighbors and minutia quadruplet features, 3) semiautomated latent fingerprint recognition using global minutia matching technique, and
4) efficient minutiae point detection in feature extraction for palmprint recognition,
and efficient minutiae quadruplet matching to improve the accuracy of high resolution
full palmprint matching.
The first approach in this thesis aims to improve the accuracy of minutia detection
using local and global adaptive binarization in pre-processing stage. The performance
of fingerprint sensors deteriorate over time which causes the appearance of noise and
ghost images in the background during capture which in turn produces too many false
minutiae points in feature extraction. To remove these false minutiae, we propose a
local and global adaptive binarization that reduces the noise and ghost images present
in the fingerprint background. A comparative study is conducted to evaluate the
proposed technique across 3 optical sensors, namely, Cogent-200, BioMini-Plus, and
Upek, in the presence of ghosting and noise. We demonstrate that removal of false
minutiae using local and global adaptive binarization improves, the performance of
global minutiae matching and NIST Bozorth matching algorithms.
The existing fingerprint matching algorithms use more information in memory, due
to which the matching process is time consuming. So, we propose a hybrid matching
algorithm with k-nearest neighbor and minutia quadruplets for recognising plain fingerprint. The minutiae quadruplets are calculated considering very few characteristics
of minutiae points to reduce the memory required for storage and subsequently the
time required for matching. The space and time complexity of proposed approach are
evaluated on the finger vendor competition (FVC) ongoing data set with ISO/IEC
19794-2 template matching and verified against other existing minutiae based matching algorithms. We further compare the recognition accuracy of the proposed approach
with triplet based matching on the FVC 2004 and FVC ongoing data sets. Experimental studies suggest that our approach achieves comparable performance while using less
memory and computation time.
In forensic applications, the latent fingerprints are of poor quality and are partial prints(i.e. minimal common area between two captured prints), there by making
recognition and matching a challenging task. Therefore, a fingerprint expert needs
to accurately mark the minutiae points on latent prints before they can be used for
identification. So, in the third approach, we present a semi-automated latent fingerprint recognition algorithm using global minutiae matching technique on the standard
ISO/IEC 19794-2 templates. We demonstrate the efficacy of the proposed method on
the standard NIST SD-27 fingerprint database.
Palmprint recognition closely resembles fingerprint matching as the matching criteria and minutiae feature extraction methods are almost similar. As 30% of latent
prints are palm prints, there is a need for high performance palmprint matching alv
gorithms. Also, in regions of palmprint with high distortion, extraction of genuine
minutia points is a challenge. So, we propose an efficient palmprint feature extraction
and matching algorithm using nearest neighbour minutiae quadruplets, which improves
the efficiency of matching by discarding false minutia points in the identification of
probable matching minutiae candidates. Further, our algorithm is a full palm to full
palm matching technique, which reduces the chance of missing common areas, unlike
existing palmprint matching techniques that are based on segmentation. We show
that the proposed method achieves better equal error rates on the FVC ongoing and
THUPALMLAB data sets.
Finally, we demonstrate the feasibility of using our approaches for a large-scale
fingerprint authentication by evaluating them for public distribution system (PDS)
using point-of-sale (PoS) devices. In the traditional PDS systems, the commodities
distribution is paper based, which lacks transparency and can be easily tampered (misused). So, we propose a system that uses fingerprint based authentication to distribute
the commodities. A PoS device captures a persons fingerprint and authenticates with
reference fingerprints from the Aadhaar central information repository. In the proposed system, genuine beneficiaries can be identified more accurately and misuse of
government subsidies can be avoided.
In summary, this thesis proposes various approaches to improve fingerprint recognition by introducing adaptive binarization for efficient minutia extraction. A hybrid fingerprint matching algorithm using minutiae quadruplets and k-nearest neighbor is proposed that uses less space and time for plain fingerprint recognition. A
semi-automated matching on latent fingerprint is also proposed using global minutiae
matching. This thesis also proposes to use minutiae quadruplets for full palm to palm
matching, there by eliminating the need for segmentation. We also demonstrate that
the methods developed during the course of this thesis can be used for large-scale
e-governance applications.
[error in script]
| IITH Creators: |
| IITH Creators | ORCiD |
|---|
| C, Krishna Mohan | UNSPECIFIED |
|
| Item Type: |
Thesis
(PhD)
|
| Uncontrolled Keywords: |
fingerprint; palmprint; latent fingerprints; minutiae; triplets; quadruplets;
minutiae cylinder codes; optical sensors; binarization; thinning; ridges; valleys; thenar;
cross-sensor; nearest neighbour; identification; verification |
| Subjects: |
Computer science |
| Divisions: |
Department of Computer Science & Engineering |
| Depositing User: |
Team Library
|
| Date Deposited: |
12 Jul 2019 12:08 |
| Last Modified: |
17 Mar 2022 11:18 |
| URI: |
http://raiithold.iith.ac.in/id/eprint/5719 |
| Publisher URL: |
|
| Related URLs: |
|
Actions (login required)
 |
View Item |
Altmetric
Altmetric