efficient approaches for fingerprint and palmprint recognition

We further compare the recognition accuracy of the proposed approach with triplet-based matching on the FVC 2004 and FVC continuous datasets. We demonstrate the efficiency of the proposed method on the standard NIST SD-27 fingerprint database.

Fig. 1.1: Types of biometrics: Physiological (a) face (b) fingerprint (c) palm (d) iris.

Introduction to fingerprints

Issues in fingerprint recognition

Some of the unsolved issues in automatic fingerprint identification are: There are so many sensors that are there to capture the fingerprint. In addition, the moisture content of the fingertip can change over time which affects the quality of the fingerprint image obtained from a user.

Introduction to palmprints

Issues in palmprint recognition

Palmprints share some common characteristics, such as ridges, depressions, and points of minutiae. This makes palmprint recognition a challenging problem for high-resolution image capture.

Performance of biometric based recognition system

A typical palm print will be twenty times the size of a fingerprint captured at 500 ppi.

Applications of biometrics

Issues addressed in this thesis

So there is a need to develop algorithms that contain few characteristics of minutiae points to reduce space and time complexities. So there is a need to develop the algorithms to deal with fake details in the palm prints.

Organization of the thesis

Suitability of fingerprint as a biometric
Early history of fingerprint biometrics
Fingerprint recognition system
Basic patterns
Minutiae features
Fingerprint readers
Fingerprint images

Most of the automatic fingerprint recognition systems use minutiae points for fingerprint comparison. The false acceptance rate (FAR) mainly depends on the quality of the fingerprint.

Fig. 2.1: Pattern classes of fingerprints ([9])

Fingerprint recognition

Local ridge orientation and frequency
Segmentation
Enhancement and binarization
Minutiae extraction
Fingerprint matching
Recognition of latent fingerprint
Issues addressed in fingerprint recognition

This method is the most acceptable and most modern method of local minutiae matching technique. Finally, most of the existing latent fingerprint matching algorithms require manually marked minutiae on latent fingerprints for recognition.

Palmprints

Early history of palmprint biometrics

Palmprint recognition works similarly to fingerprint recognition, in that it uses the uneven surfaces of edges and valleys of the hand. An advantage of scanning palm prints is that it provides a large number of details for comparison, much more than just fingerprints, which can lead to greater accuracy. 2004 – The first statewide computerized palmprint databases are deployed in Connecticut, Rhode Island and California.

Palmprint recognition system

Line-based approaches
Subspace-based approaches
Statistical approaches
Issues addressed in palmprint recognition

They use the zero crossings of the first order derivatives to identify the edge points and associated directions. The magnitude of the corresponding second-order derivative is taken as the magnitude of the lines. The weighted sum of the local direction magnitude is considered as an element in the feature vector.

Summary

Capacitive sensors

The cell operates in two phases: first, the charge amplifier is reset, short-circuiting input and output of the inverter. Since feedback capacitance is inversely proportional to the distance from the skin, a linear dependence of output voltage on skin distance is expected. An array of cells is addressed in a raster mode by means of horizontal and vertical scanners.

Optical sensors

Related work on fingerprint matching

80] proposed a method for fast singularity search using the delta field Poincare index and a fast classification algorithm to classify the fingerprint into five classes. 84] have proposed the comprehensive cross-matching algorithm for the fingerprint images captured with different sensor technologies. The experiments are conducted on the fingerprint data captured using Personal Identity Verification (PIV) certified authentication devices [85].

Fig. 3.3: Fingerprints captured from cogent CSD200

Local and global adaptive binarization

Experimental results and analysis

3.8: (a) Bozorth analysis and proposed matching using images captured with the biomini device, (b) Bozorth analysis and proposed matching using images captured with the Cogent CSD200 device. 3.9: (a) Bozorth analysis and proposed matching using images captured with the Upek device, (b) Bozorth analysis and proposed matching using images captured by all three devices. 3.10: (a) Analysis on the Bozorth match using images captured by all three devices, (b) Analysis on the proposed match using images captured by all three devices.

Fig. 3.8: (a) Analysis on Bozorth and proposed matcher using the images captured with biomini device, (b) Analysis on Bozorth and proposed matcher using the images captured with Cogent CSD200 device

Summary

In this chapter, we aim to optimize the resource usage (space and time) of fingerprint matching algorithms by using only a few features of minutiae points. We propose a hybrid fingerprint matching algorithm using k-nearest neighbor matching and minutiae quadruplets with few features of minutiae points for effective resource utilization. In the context of large-scale fingerprint recognition systems, the fingerprint matching system must utilize the minimum computational complexity and memory space, otherwise the system faces scalability problems.

Existing fingerprint minutiae matching methods

Correlation-based matching

Usually, a cross-correlation measure representing similarity between the two images (such as sum of squared difference of intensity values) is calculated. But the direct calculation of correlation is not a good solution due to the following reasons. Skin condition and finger pressure cause the images' brightness, contrast and ridge thickness to vary significantly across different impressions of the same finger.

Ridge features based matching

Unfortunately, most of the global matching algorithms are computationally demanding and lack robustness with respect to nonlinear distortions. Many of the above techniques [95] require prior alignment of the two fingerprint images, which is computationally expensive. Since nonlinear distortions cause impressions of the same finger to differ with respect to global structure, these techniques are unable to handle local nonlinear distortions.

Minutiae based

Fixed radius based structures
Minutiae cylinder code

However, like most fixed-beam approaches, their method suffers from boundary errors. The technique proposed by Feng [93] does not suffer from boundary errors and can be regarded as an advanced fixed-radius local matching algorithm. Most fixed radius approaches lead to a variable length descriptor (since the number of minutiae in the sphere will depend on the minutiae density around the central minutia), which is more complex to match.

Fig. 4.5: The k-directional nearest neighbor (k-DNN) structure proposed by Kwon (for k=8)

Proposed hybrid fingerprint recognition

If P and Q are set of probe and reference minutes, then the probe minutiae point P and query minutiae point Q are defined as P1.m ={Distp, Dirp} and Q1.n = {DistqDirq}, where Distp is the Euclidean distance to nearest minutiae points, and Dirp is the direction difference with nearest minutiae points.

Global minutiae matching using minutia quadruplets

Find the partners for each minutia in the query and probe template using k-nearest neighbors (k = 7,8,9 and 10), using Euclidean distance and direction differences. Let M be the set of local minutia pairs corresponding to the query and probe template. Sort minutia pair information in the query and explore in ascending order using Euclidean distance.

Experimental results

Fig. The graph from FVC shows that the EER is at 1.037 % on the continuous FVC database with the proposed combination of feature extractor and matching, and has achieved reduced space and time, the proposed feature extractor is using advanced techniques of pre-processing. Table 4.4: Equal error rates for fingerprint matching using triplets and quadruplets in the FVC 2004 database.

Fig. 4.9: Results on FVC ongoing: (a) FMR(t) and FNMR(t) graphs (b) Score distributions (c) DET graph ([104])

Summary

It can be noted that the proposed fingerprint matching algorithm performs better in terms of space and time complexity compared to the other algorithms SourceAFIS (independent developer), fpcoreII, psmath and HXKJ published by companies. The problems associated with common fingerprint matching and the proposed approaches to effectively address these problems are discussed in the earlier chapters. These minutiae points are in turn used by a new global minutiae matching approach, which is tolerant to spurious minutiae, for effective fingerprint matching.

Fig. 5.1: Latent fingerprint from NIST special database 27 ([106]) textured surfaces.

Existing matching techniques on fingerprints

A fuzzy similarity measure for two triangles is introduced and extended to construct a similarity vector that includes the triangle-level similarities for all triangles in two fingerprints. Finally, the FFM method maps the similarity vector pair to a normalized value that quantifies the overall image to an image similarity within the real interval [0, 1]. However, marking extensive features (orientation field, reef skeleton, etc.) in poor quality latents is very time consuming and may only be feasible in rare cases.

Latent fingerprint recognition

Latent fingerprint feature extraction

Information redundancy is used in an adaptive matching filter that is applied to each pixel in the fingerprint image. A filter is applied based on the local orientation of ridges around each pixel to improve the orientation of ridges in the same direction of the same location. The ridges can be extracted after the fingerprint image is enhanced by a denoising process.

Semi-automated latent fingerprint matching

Experimental results

It is observed that the matching accuracy is better in the group of LF-1 latent fingerprints where 60% of the cases are identified in the top 10 search results, 30% of the cases are identified in the top 100 search results and about 10% are identified. not identified. Similarly, latent fingerprints in LF-2 are identified approximately 40% in top 10 search results, 30% in top 100 search results; and the remaining cases are not identified. It is observed that the matching performance for LF-1 group of quality latents is significantly improved compared to the latent fingerprints belonging to the other two groups LF-2 and LF-3.

Table 5.1 gives the matching performance of latent fingerprints from NIST SD27 database

Summary

This work considers palm print recognition using high-resolution images, which is an emerging area of research. Due to the large size of high-resolution palmprint images, algorithms with smaller memory footprint and low computational cost are needed. There are mainly two different approaches for palmprint matching on high-resolution palmprints, namely minutiae-based [87] and ridge feature-based [88].

Fig. 6.1: Palmprint Image (a) Interdigital (b) Hypothenar (c) Thenar

Related work

In cam feature-based palmprint matching, features of the palmprint ridge pattern such as local ridge orientation, frequency, and shape are extracted for comparison. The palmprint matching algorithm is correct when there are true matches (true accepts) and true rejections (true non-matches). The match is wrong when there are cheater matches (false accepted) and cheater non-matches (false rejects).

Palmprint recognition

Palmprint feature extraction

Local and global adaptive binarization is the process that combines local average intensity as well as global average intensity information to binarize the fingerprint image. The proposed binarization combines both the global binarization as well as local adaptive binarization, to remove the false smalls formed due to noise and ghosts present in the palm print. This operation is necessary to simplify the subsequent structural analysis of the image for feature extraction.

Fig. 6.2: Variuos phases of feature extraction

The Euclidean distance between minutiae a and b is given as. xa−xb)2−(ya−yb)2, (6.3) where Distab is the Euclidean distance between minutiae points a and b, xa is coordinate x of minutiae point a, ya is coordinate y of minutiae point a, xb is coordinate x of minutiae point b, and yb is coordinate y of minutiae point b.

Computing match score using quadruplets

Experimental setup and results

Find the partners for each minutia in the query and probe template using k-nearest neighbors (k = 5,6,7 and 8), using Euclidean distance and direction differences. Let P and Q be the fingerprint minutiae of probe and gallery images, respectively, and R and S be the minutia edge pair information. It is clear from Table 6.2 that on both THUPALMLAB and FVC datasets, the lowest EER is achieved when six nearest neighbors are considered for matching.

Summary

Aadhaar authentication

The main objective of the Aadhaar project is to provide a unique number based on the person's biometrics such as fingerprints, iris and face. Aadhaar authentication is an online process where the Aadhaar number is submitted to the CIDR along with other attributes including biometric data. During the authentication transaction, the resident record is first selected using the Aadhaar number and then the.

Students attendance in academia

Hand-held fingerprint terminals

Point of sale device

IP-based fingerprint reader

Implementation model

Aadhaar authentication system

The raw materials are allocated to the FP store based on the real-time closing balances. The PoS terminal generates a one-time password (OTP) and is authenticated using the SMS received on the mobile phone.

Student time and attendance

This is a mandatory task that will be used to easily track course attendance from the device at a later time. Once the course authentication is over, the device is passed to each student for their finger authentication, saving the attendance status of each student for that particular course. On the server side, all these details are separated based on course registration data.

Program analysis and experimental results

Aadhaar authentication system

Almost 85 percent of beneficiaries have UID numbers and the remaining 15 percent of beneficiaries have EID (Enrollment Id) numbers. Beneficiaries with EID numbers are authenticated with the OTP (one time password) or UID of authorized village employees on behalf of the beneficiary. In the second month, 99 percent of the beneficiaries authenticated with their fingerprints, while the rest of the beneficiaries could not authenticate due to the poor quality of the fingerprints.

Fig. 7.1: Transaction trend of the proposed system

Student time and attendance system and its proof of concept

It is noticeable that the process is faster than the traditional smart card based authentication method. It is observed that almost 25% savings in urban areas, 15% to 20% in rural areas and almost 30% savings in tribal areas. Attendance percentage compared to manual attendance. Both approaches are validated and produce the same results, with this approach manual work has been reduced.

Fig. 7.3: Monthly authentication percentage, with fingerprints, supervisor, and OTP the attendance percentage for every student is plotted as shown in the figure 7.4

Summary

The proposed semi-automated latent fingerprint recognition is evaluated on the FVC SD standard latent database 27. The experimental study suggests that the proposed algorithm improves the performance of high-resolution palmprint matching. The proposed large-scale real-time fingerprinting application in the public distribution system is able to eradicate the fraud involved with the existing manual system.

Directions for future research

Zhang, "Fuzzy direction element energy feature (fdeef) baseret håndfladeaftryksidentifikation," i International Conference on Pattern Recognition, s. Zhuang, "Palmprint identification using boosting local binary pattern," i International Conference on Pattern Recognition, s. Krishna Mohan, ” Latent Fingerprint Recognition using ISO 19794-2 Fingerprint Templates,” Journal of Recent Trends in Engineering and Technology, bind 11, pp.

Typical block diagram of biometric recognition system

Fingerprint features

Typical block diagram for enrollment, verification and identification

Different types of sensors

Palmprint features ([3])

ROC curve to measure the performance of a biometric system

Pattern classes of fingerprints ([9])

Fingerprint global features ([9])

Palmprint features ([42])

Capacitive sensing technology ([67])

Optical sensing technology ([68])

Fingerprints captured from cogent CSD200