Printer Friendly

Performance analysis of palmprint pattern recognition techniques.


In recent years, biometric applications present an extensive and important method for efficient identification of person through physical or behavioral characteristics. Several biometric characteristics fingerprints, palm print, hand geometry, iris, retina, face, hand vein, facial thermo gram, signature, voice, etc. are researched by many research communities. The often used modalities are fingerprint and face recognition. However, for face authentication people have issues of pose and illumination invariance where fingerprint fails to contain a good psychological effect on the user due to its large utilization in crime investigations. If any biometric modality is to perform well in future, it has to contain the traits like uniqueness, accuracy, richness, ease of acquisition, reliability and user acceptance.

Palm print based pattern recognition is observed as an efficient method for individual user identification with high confidence. Palm print with larger inner surface of hand includes many features such as principle lines, ridges, minutiae points, singular points and textures. The key tasks in designing a palm print identification system comprises of: feature extraction, feature selection and feature matching. The first two tasks are used to extract a set of features from each palm print image and estimate the degree of matching between two palm prints by calculating the distance between their feature sets.

The rest of the paper is organized as follows. Section 2 enumerates survey of related work. Section 3 describes proposed methodologies. Section 4 gives an analysis of various metrics. Finally, section 5 presents the conclusion.

Literature Review:

Fine Ridge Structure Dictionary (FRSD) [1] applied total variation model and dictionary based approach to balance robustness and accuracy while segmenting latent images. But, lack of confidence on robust patch quality compromised the computation efficiency of palm print images. Personal Identification using Left and Right Palm Print images (PI-LRPP) [2] combined both left and right palm print images to improve the matching score level fusion. But time to obtain the features was not considered. Minutia propagation algorithm introduced in [3] provided measures for reducing the time taken during palmprint matching. The features considered were limited which is solved by applying different features like principle lines, ridges, minutiae points and textures in our method.

Robust key point detection using Scale Invariant Feature Transform (SIFT) [4] resulted in improvement in retrieval performance, but at the cost of time which has been addressed in our method using time effective bit parallel ordering. Fei Liu., et al., [5] planned singular value decomposition based minutiae matching method for finger vein recognition. But, the accuracy of the methods is inadequate.

Proposed Methodologies:

Palm print identification system has several steps. Initial work designs Feature Segmentation and Extraction for palm print images to increase the accuracy and robustness of palm print features extracted. Additionally, to increase the feature extraction accuracy, Palm print Feature Selection mechanism is designed to perform efficient and exact matching with the extracted minutiae palm print features by means of genetic operations for instance population, selection, crossover and mutation. Final work introduced feature pattern matching resulted in significant improvement over the modern methods.

Palm Segmentation And Extraction:

Palm print segmentation is a significant task for palm print recognition. Using the segmentation operation, predictable area of each palm print is extracted for later feature extraction and matching.

A Gaussian Measure Curvelet based Feature Segmentation and Extraction (GMC-SE) method is developed for preprocessing of palm print images using Edge Based Tangent model and obtaining region of interest and is shown in figure 3.1. This results in the execution time reduction or removal of distortions in the images.


Then, segmentation of features from the preprocessed images is performed. Effective segmentation of features is performed by applying Gaussian Measure Feature Segmentation. The application of GMC-SE model improves the computation efficiency by obtaining the spectral minutiae based on the local and global features. Finally, efficient feature extraction is performed through Curvelet based model by extensive selection of ridge candidates which results in feature extraction accuracy.

Feature Selection:

Extracting minutiae features from segmented palm print images is considered to be the significant process in automatic palm print recognition system. For effective minutiae based feature selection on segmented palm print image Fuzzy Minutiae Palm print Feature Selection (FMPFS) Using Radial Basis Interpolation Function is employed.


(a) Ridge Ending (b) Ridge bifurcation

(c) Ridge Enclosure (d) Independent Ridges

As illustrated in Figure 3.2, the minutiae features such as ridge ending, ridge bifurcation, ridge enclosure, independent ridges are selected in the segmented palm print image using fuzzy rule principle to improve the processing speed without any overlap.

Radial Basis Interpolation Function used in FMPFS method offers an interpolated functional value on feature extraction at irregular positional points. Herewith, the interpolation function produces approximate result with the center of basis palm print function. The extracted minutiae palm print features are then matched with the database using genetic operations to achieve high and accurate palm print recognition through feature matching process.

3.3 Feature Matching:

Rabin-karp Palmprint Pattern Matching (RPPM) method is used for improve the performance of palm print feature matching. The RPPM method used double hashing technique for evaluate the value of palm print pattern with different features. Double hashing is also used to avoid the hash collision with different set of value position on the test and training samples. Double hashing is employed to achieve the same hash key used to improve the accuracy of pattern matching features with different angle of position. RPPM method matches with multiple patterns of features simultaneously between the test and training samples using the Aho-Corasick Multiple Feature matching procedure. Aho-Corasick Multiple Feature locates the position of the features with finite set of bit values as an input text. Finally, a time efficient bit parallel ordering present an efficient variation on matching the palm print features of test and training samples with minimal time.

In order to test the proposed method, experiments on the PolyU 2D Palm print Database and CASIA database is performed. The PolyU 2D Palm print Database includes 8000 samples collected from 400 different palms. Twenty samples from each of these palms were collected in two separated sessions, where 10 samples were captured in each session, respectively. The training model for CASIA database consists of all palm print images are 8-bit gray level JPEG files providing the images of palm print using the database CASIA. The sample palm print images are taken from the CASIA database which used for improved feature pattern matching accuracy.

Analysis Of The Parameters Based On Gmc-Se, Fmpfs And Rppm:

Experiments are conducted for GMC-SE, FMPFS and RPPM technique. Simulations are carried out to measure in terms of computation efficiency, Palm print Biometric Precision Ratio, Pattern matching efficiency, false positive rate, Cumulative accuracy on hashing are measured to prove the real time usage in practical. Detailed result analyses of these metrics are elaborated in further section.

6.1 Computation Efficiency:

The Gaussian Measure Curvelet based Feature Segmentation and Extraction (GMC-SE) is used to perform an efficient segmentation process for improves the computation efficiency. Gaussian Measure Feature Segmentation uses square regions for effective segmentation of high resolution palm print images. Computational efficiency is defined as the ratio of number segmented images to the total number of images. It is measured in terms of percentage (%).

Computation Efficiency = segmented Images/Total No. of images * 100

Table 6.1 determines the computation efficiency with respect to different number of images and is measured in terms of percentage (%). Number of images used for experimental purposes varies from 10 to 70 and comparison is made with proposed GMC-SE technique and existing SVD by Fei Liu., et al., (2014) and FRSD by Kai Cao., et al., (2014), PI-LRPP by Yong Xu ., et al .,(2015). Higher the computation efficiency the GMC-SE technique is said to be more efficient.


Figure 6.1 shows the computation efficiency with respect to different palm print images in the range of 10 to 70. From the figure it is illustrative that with the increase in the number of images, the computation efficiency is increased in all the methods. But, comparatively, the computation efficiency is higher using proposed GMCSE. This is because with the application of Gaussian Measure Feature Segmentation, both local and global palm print features are combined by evaluating the spectral minutiae in the square regions. As a result, the GMC-SE method improves the computation efficiency by 17 % as compared to existing FRSD by Kai Cao., et al., (2014). In addition, the computation efficiency of proposed GMC-SE method is 25% higher as compared to existing PILRPP by Yong Xu ., et al .,(2015).

6.2 Palm print Biometric Precision Ratio:

The set of features are selected through fuzzy logic principle applied in FMPFS with the segmented palm print input image. Irrelevant features are ignored using FMPFS method with Lukasiewicz Fuzzy Logic Rule at the initial stage.

Feature selection is obtained through Precision ratio which is defined as the ratio of number of true feature retrieval to the number of false feature retrieval and true feature retrieval which is measured in terms of percentage (%).

P = (No. of true features retrieved)/(No. of true features retrieved)+ (No. of false features retrieved) * 100

The Palmprint Biometric Precision Ratio of FMPFS method is presented in table 6.2. Comparison made with two other existing schemes namely, MLPM method and HRMPR Method The proposed FMPFS technique provides high precision ratio to improve the feature selection. The features are varied from 20 to 140. For all the varying features, the proposed FMPFS technique provides high precision ratio than the existing approaches. The table values are plotted in graph shown in figure 6.2.


Figure 6.2 depicts the palm print biometric precision ratio determined based on the number of relevant features extracted from the given palm print image database. Compared to the existing MLPM and HRMPR, the proposed FMPFS method provides higher rate of biometric precision ratio. From the figure 4.6, it is illustrative that the proposed FMPFS framework increases the feature selection based on the retrieval of relevant images and removal of irrelevant images. This is because by applying the Lukasiewicz Fuzzy Logic Rule that works well with 'n' features. This fuzzy logic rule perform both the processes of removal of irrelevant features and retrieved the relevant features improving the palm print biometric precision ratio by 7% when compared to existing MLPM by Eryun Liu., et al., (2013) and 20.5 % as compared to existing HRMPR by Raffaele Cappelli., et al ., (2012) respectively.

6.3 Pattern matching efficiency:

The pattern matching efficiency rate in RPPM is the amount of patterns efficiency matched using the double hashing procedure. The pattern efficiency rate is measured in terms of percentage (%) and is the ratio of the number of features matched to the total number of features provided.

PME = Matched features/No. of features * 100

Table 6.3 shows the pattern matching efficiency over different features provided as input using MATLAB. From the figure, with an increase in the number of features, the pattern matching efficiency also increases. As these images are not similar, the changes in the pattern matching efficiency are also being observed. As a result, the percentage increase or decrease in pattern matching efficiency does not remain the same.

Figure 6.3 shows the pattern matching efficiency is improved using the proposed method RPPM. Pattern matching efficiency rate in RPPM method is improved with the application of double hashing procedure that uses different angular position of minutiae points and results in higher pattern matching rate by 12.5% as compared to existing SVD by Fei Liu ., et al., (2014) on working with the test and training sample images.


In addition using RPPM method, based on the positional changes of the features, with the help of alter key and hash table, the features are not only matched with the single hash value but with different hash key resulting in the improvement of pattern matching efficiency by 17.5% as compared to MSR by Sumit Shekhar., et al., (2013).

6.4 Feature extraction accuracy:

Feature extraction accuracy is defined as the ratio of number of features being extracted to the total number of images. It is measured in terms of percentage (%).

Feature extraction accuracy = No. of Features extracted/Total no. of images * 100

Table 6.4 shows that feature extraction accuracy with respect to 70 images given as input. Comparison is made with the two existing schemes namely, FRSD by Kai Cao., et al., (2014), PI-LRPP by Yong Xu ., et al., (2015) and proposed GMC-SE method. From the table, it is clear that the proposed GMC-SE framework achieved high Feature extraction accuracy while increasing the number of images. Each image is examined and the combined performance is plotted. All the 70 images were extracted and then it stored with their characteristics.


Figure 6.4 illustrates that the feature extraction accuracy is compared with proposed GMC-SE and existing FRSD and PI-LRPP methods. The proposed GMC-SE method provides higher feature extraction accuracy than the state-of-art-methods. This is because of the application of Curvelet based Feature Extraction Model that significantly extracts the feature improving the accuracy rate. By applying Curvelet based Feature Extraction Model three minute lines are considered to extract the exact texture of palm print images improving the accuracy rate by 6.5% as compared to FRSD by Kai Cao., et al., (2014). Furthermore, ridge candidate selection uses both local and global features based on scaling, angular movement and orientation improving the accuracy by 13% as compared to existing PI-LRPP by Yong Xu ., et al .,(2015). Therefore, the proposed GMC-SE method attained high feature extraction accuracy among the other proposed methods.

6.5 False positive rate:

The false positive rate on matching the patterns is the ratio of false positive to the false positive and true negative value. It is measured in terms of percentage (%).

FPR = FP/FP + TN * 100

The false positive rate for 35 test images with varying principle lines, ridges, minutiae points and textures are considered and compared with two other existing SVD by Fei Liu ., et al., (2014) and MSR methods are shown in table 6.5. Lower the false positive ratio, more efficient the RPPM method is said to be.


Figure 6.5 illustrates the results of false positive rate on matching the patterns using RPPM method is compared with two state-of-the-art methods SVD and MSR. The figure 6.5 is presented for visual comparison based on the initialization of features. The Aho-Corasick procedure used in RPPM method designs the finite state machine in an accurate manner for performing easy matching functions for multiple features without any backtracking process. Therefore, it minimizes the false positive rate on pattern matching using RPPM method by 17.5 % compared to SVD by Fei Liu ., et al., (2014). Furthermore, the linear form of multiple feature matching operation further enhances the accuracy and therefore reduces the false positive rate on pattern matching by 27% as compared to MSR by Sumit Shekhar., et al., (2013).

6.6 Cumulative accuracy on hashing:

The cumulative accuracy on hashing is ratio of difference between the actual features and measured features to the actual features. It measured in terms of percentage (%).

CAH = (1 - [] - []/[]) * 100

The cumulative accuracy on hashing is performed between the proposed RPPM and the existing SVD by Fei Liu ., et al., (2014) and MSR by Sumit Shekhar., et al., (2013) method is shown in table 6.6. The number of features ranges from 10 to 70 where the experiments were conducted using MATLAB. The convergence plot with differing samples is depicted in above table 6.6. From the table, proposed RPPM method had better cumulative accuracy on hashing.


Figure 6.6 shows that the plot for measuring the cumulative accuracy on hashing with respect to number of features. From the figure, proposed RPPM method converge high cumulative accuracy on hashing than existing SVD and other existing methods with the application of multiple feature pattern matching. This is effectively carried out using probing sequence of bits for different images that efficiency performs different separation to match the specific features using double hashing table. This result in the increase of cumulative accuracy on hashing using RPPM method by 14% compared to SVD by Fei Liu ., et al., (2014). The proposed RPPM methods also improved the accuracy rate of hashing by 19% as compared to MSR by Sumit Shekhar., et al., (2013).


A perfect illustration is discussed on analysis of proposed GMC-SE, FMPFS, and RPPM frameworks. The theoretical analysis and experiment result shows that, the proposed GMC-SE method increased feature extraction accuracy by 21%. The Gaussian Measure Curvelet based Feature Segmentation and Extraction algorithm effectively increased computation efficiency, feature extraction accuracy on several test sets and also reduced the execution time using Edge Based Tangent (EBT) compared to existing state-of-the-art work. The FMPFS framework is improved the matching accuracy, average approximation value and palm print biometric precision ratio. The proposed FMPFS used Radial Basis Interpolation Function at irregular positional points the average approximation is increased 16% obtains the center functional point through Gaussian function. In addition, Pattern matching efficiency rate in RPPM method is significantly improved with the application of double hashing procedure by 14% using CASIA Palm print image database.


[1.] Kai Cao, Eryun Liu, Member, IEEE and Anil K. Jain, Fellow, IEEE, 2014. "Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary", IEEE Transactions on Pattern Analysis and Machine Intelligence, 36.9.

[2.] Yong Xu, Member, IEEE, Lunke Fei and David Zhang, 2015. 'Combining Left and Right Palmprint Images for More Accurate Personal Identification", IEEE TRANSACTIONS ON IMAGE PROCESSING, 24: 2.

[3.] Eryun Liu, Anil K. Jain and Jie Tian, 2013. " A Coarse to Fine Minutiae-Based Latent Palmprint Matching", IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 35: 10.

[4.] Sunpreet, S. Arora, Eryun Liu, Kai Cao and Anil K. Jain, 2014. "Latent Fingerprint Matching: Performance Gain via Feedback from Exemplar Prints", IEEE Transactions on Pattern Analysis And Machine Intelligence, 36: 12.

[5.] Fei Liu., Gongping Yang., Yilong Yin., and Shuaiqiang Wang., 2014. "Singular value decomposition based minutiae matching method for finger vein recognition" Neuro computing, Elsevier, 145: 75-89.

(1) Mrs.S. Kanchana and 2Dr.G. Balakrishnan

(1) Associate Professor, Science and Humanities Indra Ganesan College of Engineering, Trichy, Tamilnadu, India.

(2) Professor, Computer Science and Engineering Indra Ganesan College of Engineering, Trichy, Tamilnadu, India.

Received 25 February 2016; Accepted 10 April 2016; Available 15 April 2016

Address For Correspondence:

Mrs.S. Kanchana, Associate Professor, Science and Humanities Indra Ganesan College of Engineering, Trichy, Tamilnadu, India.

Table 6.1: Computation efficiency

Total no. of images   Computation efficiency (%)

                      Proposed GMSE   Existing FRSD   Existing PI-LRPP

10                    90              69              61
20                    92              75              70
30                    93              78              72
40                    95              80              75
50                    96              82              79
60                    97              84              81
70                    98              87              84

Table 6.2: Palmprint Biometric Precision Ratio

Number of features   Palm print Biometric Precision Ratio (%)

(n)                  Proposed FMPFS   Existing MLPM    Existing HRMPR

20                   90.54            80.35            65.35
40                   91.09            82.68            68.95
60                   91.67            84.25            71.25
80                   92.24            86.37            73.21
100                  93.12            87.64            75.32
120                  93.87            90.24            76.12
140                  95.35            91.35            81.21

Table 6.3: Pattern matching efficiency

Number of features   Pattern matching efficiency (%)

                     Proposed RPPM    Existing SVD    Existing MSR

10                   90.24            74.11           69.54
20                   92.05            78.52           71.26
30                   92.75            80.32           75.32
40                   93.12            82.24           78.64
50                   93.98            84.46           80.52
60                   94.13            86.53           81.35
70                   95.36            87.64           82.47

Table 6.4: Feature extraction accuracy

Total no. of images    Feature extraction accuracy (%)

                       Proposed    Existing    Existing
                       GMC-SE      FRSD        PI-LRPP

10                     91          82          75
20                     92          84          78
30                     94          86          80
40                     95          88          82
50                     96          91          84
60                     97          93          86
70                     98          94          89

Table 6.5: Tabulation for false positive rate

Number of images       False positive rate (%)

                       Proposed RPPM    Existing SVD    Existing MSR

5                      40               45              51
10                     42               48              52
15                     45               52              54
20                     46               56              59
25                     48               58              62
30                     51               62              67
35                     52               64              68

Table 6.6: Cumulative accuracy on hashing

Number of Features     Cumulative accuracy on hashing (%)

                       Proposed RPPM    Existing SVD    Existing MSR

10                     89.96            72.38           69.87
20                     91.25            78.69           70.65
30                     92.78            81.35           75.32
40                     95.64            84.62           77.65
50                     96.12            86.62           79.52
60                     96.89            87.62           80.36
70                     97.12            88.21           81.23
COPYRIGHT 2016 American-Eurasian Network for Scientific Information
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Kanchana, S.; Balakrishnan, G.
Publication:Advances in Natural and Applied Sciences
Article Type:Report
Date:Apr 1, 2016
Previous Article:Fabrication of bamboo fibre reinforced polymer matrix composites.
Next Article:Artificial neural network based performance assessment of thermosyphon solar water heating system.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters