Multimodal Biometric Recognition using Gabor Wavelet, Harris Corner and Random Forest Classifier

(1)

Multimodal Biometric Recognition using Gabor Wavelet, Harris Corner and Random Forest Classifier

Prachi Agarwal M. Tech. Scholar

Department of Electronics and Comm. Engg.

Shrinathji Institute of Technology & Engineering, Nathdwara, Rajasthan (India)

[email protected]

Pankaj Rathi HOD

Department of Electronics and Comm. Engg.

Shrinathji Institute of Technology & Engineering, Nathdwara, Rajasthan (India)

[email protected]

Abstract – Biometric recognition systems use certain human characteristics such as voice, facial features, fingerprint, iris or hand geometry to identify an individual or verify their identity. These systems have been developed individually for each of these biometric modalities until they achieve remarkable levels of performance. Multimodal biometric systems combine different modalities in a unique recognition system.

The multimodal fusion allows to improve the results obtained by a single biometric characteristic and make the system more robust to noise and interference and more resistant to possible attacks. The fusion can be carried out at the level of the signals acquired by the different sensors, of the parameters obtained for each modality, of the scores provided by unimodal experts or of the decision taken by said experts. In the fusion at the level of parameters or scores it is necessary to homogenize the characteristics coming from the different biometric modalities prior to the fusion process. This paper presents the development of a multimodal biometric identification system based on two biometrics namely, the iris and the fingerprint.

Feature extraction is done using Gabor Wavelet and Harris Corner Method and classification is accomplished using Random forest classifier.

Keywords –Gabor wavelet, Harris Corner Method, Multimodal biometrics, Random Forest classifier.

I. INTRODUCTION

The identification of individuals corresponds to the search for the identity of the person who appears in a database. It can be used to authorize the use of services, for example to control access to a highly secure area for which only a limited number of people (registered in a database) have access authorization, as it can be used to recognise criminals and terrorists.

To meet these needs, biometrics seems to be a practical, effective solution whose cost in effort and money is constantly decreasing. In fact, biometrics is developing rapidly. This infatuation leads to the

growth of a wide variety of biometric methods: from the classical ones, such as the study of fingerprints [1] or iris [2], to more exotic ones like the recognition of the gait [3], recognition of the shape of the ear [4]. Manufacturers are increasingly proposing, for problems requiring a great deal of security, to no longer use a single feature but to set up a system based on combinations of different biometric means in order to further increase safety.

Despite this rapid development, biometrics has points of imperfections. Indeed, at present, there are still very few reflections before the implementation of a biometric solution, whether at the level of the chosen method, the constraints imposed on users or the level of security chosen. There are sometimes aberrations: high-performance sensors coupled with obsolete recognition algorithms, all to control access to a school restaurant.

In addition to this lack of prior reflection, the other critical point of biometric systems concerns their reliability and the mechanisms of recognition or authentication to be implemented.

In his daily environment, an individual needs to identify himself in a multitude of contexts: to enter his building or access his workplace, to withdraw money from a distributor or pay in stores, to request a service social. So many codes and passwords to memorize and protect.

In order to develop the means of recognition, the research has been in recent years a spectacular revival and has a major interest in "biometric" data, that is to say the characteristics of each person: his voice, fingerprints, and features of his face, the shape of his hand, his signature and even his DNA.

Thus, biometric-based techniques enjoy a general enthusiasm favoured by a fashion phenomenon, mainly conveyed by films in the cinema and on television.

However, more recently, the increase in identity fraud has created a growing need for biometric

(2)

technology in a number of applications requiring a high degree of security: access to sensitive sites, airport surveillance [5-8].

The process of biometric identification systems has two phases: the enrolment during which the biometric information of a person is added to the system and the identification during which new biometric information is compared with those already recorded. The enrolment phase requires several samples of biometrics to define the constitutive models of the database. These models must be a set of useful, discriminating and non- redundant information. In this paper, fusion of Harris corner method and Gabor wavelet is proposed and performing the results with Random Forest Classifier [9].

II. PROPOSED METHODOLOGY

The multimodal method executes different algorithm according to different input patterns. The automated system is executed and different id will be generated according to training and testing input.

A. Pre-Processing Block

The phases involved in pre-processing are shown in the flow diagram in Figure 1. After resizing, the RGB image is transformed to a gray scale image using rgb2gray function.

B. Morphological Operations

Mathematical morphology processes images with the help of previously chosen special shapes, which are generally smaller than the image and are called the structural element, which acts as an operator on an image to produce a result.

The shape, size and orientation of the structural element are chosen based on prior knowledge taking into account the relevant geometric structures that are present in the image and the objective pursued with the morphological operation implemented.

Each structure element requires the definition of a point of origin (or reference) for its application as a morphological operator. This allows the structure element to be related in a particular way to the pixels of the image.

Figure 1: Proposed block diagram RGB to

Gray

Morphological operation Pre-Processing

Feature Extraction using Gabor Wavelet Harris Corner method Input

Classification using Random Forest Classifier

Result in terms of Accuracy, Precision and

Sensitivity Input

Testing Training

Store in

(3)

Dilation: It is also called expansion, filling, or growth, produces a thickening effect on the edges of the object. This algorithm is used to increase the contour of the objects and to join the discontinuous lines of these, produced by some filtration, mathematically the binary dilation is defined as:

𝐴 ⊕ 𝐵 = {𝑐 ∈ 𝐸^{𝑁| 𝐶= 𝑎+}𝑏 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑎 ∈ 𝐴 𝑎𝑛𝑑 𝑏 ∈ 𝐵}

(1) Erosion: Erosion is the dual function of expansion, but it is not the reverse, i.e. if erosion is done and then a dilation, the resulting image will not be equal to the actual image, mathematically erosion represented as:

𝐴 ⊝ 𝐵 = {𝑥 ∈ 𝐸^𝑁|𝑥 + 𝑏 ∈ 𝐴 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑏 ∈ 𝐵} (2)

C. Feature Extraction

There are two features have been considered for proposed multimodal biometric recognition. Gabor wavelet and Harris Corner method are used for the feature extraction of the input image.

1) Gabor Wavelet

The directional decompositions for image analysis were probably initiated by Gaussian windowing in the 2D Fourier domain, and continued 4D filters as well as the directional Gabor wavelets proposed by J. Daugman [10]. The use of Gabor's wavelets for the image is often linked to the consideration of the human visual system.

Gabor wavelets are constructed by isotropic Gaussian windowing of a complex plane wave of frequency 𝐹 in direction 𝜃 [11]:

𝜓^𝜃(𝑥) =^𝑒^{−‖𝑥‖2/2}

2𝜋 𝑒^−𝑗(𝑥^𝑇^𝜔⁰⁾ (3) Where, 𝜔0= 𝐹[cos(𝜃) ; sin(𝜃)]^𝑇

The decomposition depends on the number of orientations 𝐾 that one fixes, and that one can distribute evenly in [0, 𝜋].

𝜃 ∈ Θ = {^𝑘𝜋

𝐾 ; 0 ≤ 𝑘 ≤ 𝐾} (4) We can then decompose any real 2D signal (𝑥) by scalar product with the following atoms:

{𝜓_𝑗,𝑢^𝜃 (𝑥) = 2^−𝑗𝜓^𝜃(2^−𝑗(𝑥 − 𝑢))}

𝜃∈Θ,𝑗∈ℤ,𝑢∈ℝ² (5) 2) Harris Corner Method

Harris corner detection is a post-processing technique used for smile recognition by the lip corner data, eye detection and many more [12].

Harris corner detector has to detect the corners in the iris image 𝐼_𝐼. Going stepwise, firstly 𝐼_𝐼 is used to create gradient image through filtering it with Gaussian Mask. And then the Harris corner method is imposed [12].

Harris Corner Algorithm [13]:

1. Find the vertical gradient and horizontal gradient represent as follows:

M = (𝑃 𝑅

𝑅 𝑄) = (𝐼𝑥2 𝐼_𝑥𝑦

𝐼𝑥𝑦 𝐼𝑦2) (6) 2. Compute 𝑥 and 𝑦 derivatives of image:

𝐼_𝑥 = 𝐺_𝜎^𝑥 ∗ 𝐼, 𝐼_𝑦 = 𝐺_𝜎^𝑦 ∗ 𝐼 (7) 3. Compute product of derivatives at every pixel:

𝐼_𝑥²= 𝐼_𝑥∗ 𝐼_𝑥, 𝐼_𝑦²= 𝐼_𝑦∗ 𝐼_𝑦, 𝐼_𝑥𝑦= 𝐼_𝑥∗ 𝐼_𝑦 (8) 4. Apply Gaussian filter to the image:

𝑤μ,v= exp (⁻¹

2 (μ²+ v²) / δ²) (9) Where, 𝑤μ,v is a Gaussian window [13].

5. Calculate the 𝑟 value of the pixel:

𝑅_𝑟= {𝐼_𝑥²+ 𝐼_𝑦²− (𝐼_𝑥𝐼_𝑦)²} − 𝑘{𝐼_𝑥²+ 𝐼_𝑦²}² (10) 6. Selection of local extreme points.

7. Threshold determination and define the corner point.

Figure 2: Flow diagram of Harris Corner detection D. Classification by Random Forest Classifier The above extracted feature values can be combined to get optimal dataset for training of the classifier.

A random forest [14] is a classifier containing of one group of structured tree predictors [𝑇(𝑥,⊝𝑘), 𝑘 = 1, … . ] where the [⊝_𝑘] are random vectors of identical distributions and where every tree provides a unit poll for the furthermost common class of each entry x.

The main advantage of this structure is that it avoids the danger of over-learning for any method of prediction based on induction.

Start

Load Image 𝐼_𝐼

RGB to gray value conversion of 𝐼𝐼

Corner Detection of 𝐼_𝐼

Find the required corner point on the basis of threshold value

𝐼𝐼

Corner Points

(4)

III. SIMULATION RESULTS

Figure 3: Confusion matrix plot for iris recognition

Figure 4: Confusion matrix plot for hybrid approach

Figure 5: ROC curve for different classes

Table 1: Comparison with previous work

Method Feature Accuracy

[1] LBP + PCA 99.2%

Proposed Gabor + Harris Corner

100%

IV. CONCLUSION

The proposed multimodal system combines information from two biometric sources: the fingerprint and iris. The approach at the level of features. The results obtained are satisfactory.

Calculated accuracy is 80%% for the iris recognition. The highest accuracy achieved is 100%

when the multimodal biometric approach is used.

All the tests carried out allow us to conclude that we can take advantage of the fusion of the multimodal feature and step to increase system performance remote identification.

REFERENCE

[1] Reddy, C.V.R., Kishore, K.K., Reddy, U.S. and Suneetha, M., 2016, December. Person identification system using feature level fusion of multi-biometrics.

In 2016 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC) (pp. 1-6). IEEE.

[2] Prabhakar, Salil, Sharath Pankanti, and Anil K. Jain.

"Biometric recognition: Security and privacy concerns."

IEEE security & privacy 2 (2003): 33-42.

[3] Ahmed Badawi, Mohamed Mahfouz, Rimon Tadross, and Richard Jantz, "Fingerprint-Based Gender Classification," In Proc. of the International Conference on Image Processing, Computer Vision, Pattern Recognition, Las Vegas, Nevada, USA, Vol. 1, Jun 2006.

[4] Manish Verma and Suneeta Agarwal. “Fingerprint Based Male-Female Classification.’’ in Proceedings of the international workshop on computational intelligence in security for information systems (CISIS’08), Genoa, Italy, pp.251-257, 2008.

[5] Jen Feng Wang, “Gender Determination using Fingertip Features”, Internet Journal of Medical Update, Jul-Dec, 3(2), 22-8, 2008.

[6] Angela Bell, “Loop ridge count differences between genders”. Nebraska Wesleyan University.

(http://www.neiai.org/).

[7] Dr. Prateek Rastogi, Ms. Keerthi R Pillai, “A study of fingerprints in relation to gender and blood group”, J Indian Acad Forensic Med, 32(1), pp-11-14 ISSN 0971- 0973, 2010.

[8] Gnanasivam .P, and Dr. Muttan S, “Fingerprint Gender Classification Using Wavelet Transform and Singular Value Decomposition”. European Journal of Scientific Research ISSN 1450-216X Vol.59 No.2, pp.191-199, 2011.

[9] Gnanasivam .P, and Dr. Muttan S, “Gender Identification Using Fingerprint through Frequency Domain analysis” IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 2, No 3, March 2012.

[10] Daugman, John G. "Complete discrete 2-D Gabor transforms by neural networks for image analysis and

1 2 3 4 5 6 7 8 9 10

2 6.7%

0 0.0%

1 3.3%

0 0.0%

66.7%

33.3%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

2 6.7%

0 0.0%

1 3.3%

0 0.0%

66.7%

33.3%

0 0.0%

2 6.7%

1 3.3%

0 0.0%

66.7%

33.3%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

2 6.7%

1 3.3%

0 0.0%

66.7%

33.3%

0 0.0%

1 3.3%

0 0.0%

2 6.7%

0 0.0%

66.7%

33.3%

0 0.0%

1 3.3%

0 0.0%

2 6.7%

0 0.0%

66.7%

33.3%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

100%

0.0%

100%

0.0%

100%

0.0%

50.0%

66.7%

33.3%

75.0%

25.0%

100%

0.0%

50.0%

100%

0.0%

100%

0.0%

100%

0.0%

80.0%

20.0%

Target Class

Output Class

Confusion Matrix

1 2 3 4 5 6 7 8 9 10

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

0 0.0%

100%

0.0%

0 0.0%

3 10.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

100%

0.0%

Target Class

Output Class

Confusion Matrix

0 0.2 0.4 0.6 0.8 1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

False Positive Rate

True Positive Rate

ROC

Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8 Class 9 Class 10

(5)

compression." IEEE Transactions on acoustics, speech, and signal processing 36, no. 7 (1988): 1169-1179.

[11] Zhou, Zhiping, Huijun Wu, and Qianxing Lv. "A new iris recognition method based on Gabor wavelet neural network." In Intelligent Information Hiding and Multimedia Signal Processing, 2008. IIHMSP'08 International Conference on, pp. 1101-1104. IEEE, 2008.

[12] Chen, Jie, Li-hui Zou, Juan Zhang, and Li-hua Dou.

"The Comparison and Application of Corner Detection Algorithms." Journal of multimedia 4, no. 6, 2009.

[13] Li, Yanshan, Wei Shi, and Ailin Liu, "A Harris corner detection algorithm for multispectral images based on the correlation", pp. 5-5, 2015.

[14] Breiman, Leo. "Manual on setting up, using, and understanding random forests v3. 1." Statistics Department University of California Berkeley, CA, USA 1 (2002).