Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

A wide variety of systems require reliable personal authentication schemes to either confirm or determine the identity of individuals requesting their services. The purpose of such schemes is to ensure that the rendered services are accessed by a legitimate user, and not anyone else. Traditionally, passwords (knowledge-based security) and identification cards (token-based security) have been used to restrict access to systems. However, security can be easily breached in these systems when a password is divulged to an unauthorized user or a card is stolen by an impostor. Furthermore, simple passwords are easy to guess (by an impostor) and difficult passwords may be hard to recall (by a legitimate user). The emergence of biometrics has addressed the problems that plague traditional verification [1]. Among all biometric traits, palmprint has one of the highest levels of reliability and has been extensively used by forensic experts in criminal investigations. A palmprint is obtained from the central surface of the hand, between the wrist and the fingers. Figure 85.1 shows the shape and details of a palmprint.

Fig. 85.1
figure 1

Shape and details of a palmprint

Full size image

A simple biometric system has four important modules [2], which are the following:

  • The Sensor Module, which captures the biometric data of an individual. An example is a palmprint sensor that captures palmprint impressions of a user.

  • The Feature Extraction Module, in which the acquired data is processed to extract feature values. For example, extract shape feature by using moments algorithm.

  • The Matching Module, in which the feature values are compared against those in the template by generating a matching score. For example, in this module, the number of matching minutiae between the query and the template can be computed and treated as a matching score.

  • The Decision-making Module, in which the user’s claimed identity is either accepted or rejected based on the matching score generated in the matching module.

The rest of paper organized as follows: Sect. 85.2 explains in details the methodology that used in this paper. Section 85.3 introduces the results and discussion. Finally, Sect. 85.4 concludes the paper.

2 Methodology

This study explains and discusses the proposed method in performing palmprint recognition. The first stage is the acquisition of hand images from the CASIA. The second stage is pre-processing, which affects two main steps: including the binarizing hand image and extracting the palmprint region from the whole hand. This stage helps strip images from noise. The final stage includes feature extraction methods and matching techniques, such as invariant moments, used in computer vision applications. In this stage, the shape descriptor approach is employed in extracting global features of the palmprint, based on the wavelet transform, by decomposing each palmprint image into four bands and deriving seven invariant moments for each band. Finally, matching is carried out to perform identification. This method can generate more accurate results in less time. Points of interest in the identification of the palmprint are the features of an image, the change in the scale and rotation of the image’s features.

2.1 Hand Image Acquisition (Data Set)

The databases were obtained from the CASIA. Images of the hand were obtained using a flatbed scanner set at 256 grayscale. Grayscale images are mere representations of the image using only one color, which is gray, to cover the entire image. In a grayscale image, the density of the color component is used to represent the image. This component carries more information about the image. It also reduces the size of the image dimensions, for example, instead of three bytes, the image will be reduced gradually to a single byte (which is 0–255 in the decimal system). The image, when converted to this type of color range, it becomes easier to explain and promote. This property is also another feature of the grayscale mode [3]. In this study, we used the CASIA database. The CASIA database contains 800 images of the hands of 50 subjects. All images measure 640 × 480 pixels. Each user submitted eight pictures of their right hand and eight pictures of their left hand. The images show 256 gray levels on the flatbed scanner. The scanner is link-free, and therefore users are free to put their hands anywhere on the scanner as long as their hands are placed face down. Figure 85.2 shows samples of the left hand images. To ensure that the reproduction of your illustrations is of a reasonable quality, we advise against the use of shading. The contrast should be as pronounced as possible.

Fig. 85.2
figure 2

Samples of the left hand images

Full size image

2.2 Image Pre-processing

Image processing is an area described by the need for intensified efforts to create an experimental feasibility of the proposed solutions to a specific problem. Image processing is a discipline in computer science. Each image has a value of more than tens of thousands of words and provides a number of meanings. The digital images can be defined through the various definitions. For example, a digital image can be formulated as a function of two dimensions, and (X, Y), where x and y are spatial coordinates [4]. Values F (X, Y) represent the clarity or jamming of the image at a certain point. All the digital images of different elements, each element have a location and a value. These elements are referred to as image elements and pixels. The image processing is the study of the use of a variety of algorithms that were conducted on the pictures, where all of the input and output of the algorithms is the picture. First, we binarized hand image from the original image, at this stage a global threshold used to extract the image from rare hand. Figure 85.3 shows the original image and binarized image.

Fig. 85.3
figure 3

Binarized hand images of the left and hands

Full size image

After that, extraction of the region from the whole hand will be done [5]. Figure 85.4 shows a sample of the extracted images.

Fig. 85.4
figure 4

Samples of extracted left palmprint of one user from the CASIA database

Full size image

2.3 2D Wavelet Transform

After extracting the region from the whole hand, 2D filter banks will apply to these regions, to get decomposed images into four bands [6], low–low pass filter, low–high pass filter, high–low pass filter and high-high pass filter (LL, LH, HL, HH) (Fig. 85.5).

Fig. 85.5
figure 5

2D filter bank analysis

Full size image

2.4 Invariant Moments Algorithm

Recognition of objects efficiently can extract feature points of objects (palmprint) to create a powerful feature descriptor or the representation of objects. Hu [7] has introduced a technique to extract the shape features form of images, which are called invariant moments. These features are invariance to translation, rotation, and change in scaling.

The advantage of the moments over other techniques is the implementation of the previous descriptors is straight forward, and they also carry a “physical” interpretation of boundary shape. Hu was first to set out the mathematical foundation for two-dimensional moment invariant and demonstrated their applications to shape recognition. Invariant moments were first applied to aircraft shapes and have been proven to be fast and reliable. These values are invariant with respect to translation, rotation and shape scaling.

Hu defines seven of these shape descriptor values computed from central moments through order three that are independent to object translation, scale and rotation.

$$ \Phi 1 =\upeta20 +\upeta02 $$
(85.1)
$$ \Phi 2 = \left( {\upeta 20 -\upeta0 2} \right) 2+ 4\upeta 2 1 1 $$
(85.2)
$$ \Phi 3 { } = \, \left( {\upeta 30 - 3\upeta 1 2} \right) 2\, { + }\,\left( { 3\upeta 2 1-\upeta0 3} \right) 2 $$
(85.3)
$$ \Phi 4= \left( {\upeta 30\,{ + }\,\upeta 1 2} \right) 2 \,{ + }\,\left( {\upeta 2 1 \,{ + }\,\upeta0 3} \right) 2 $$
(85.4)
$$ \begin{aligned}\Phi 5 { } = & \, \left( {\upeta 30 \, - 3\upeta 1 2} \right) \, \left( {\upeta 30\,{ + }\,\upeta 1 2} \right) \, [\left( {\upeta 30\,{ + \,}\upeta 1 2} \right) 2\\ & - 3\left( {\upeta 2 1 \,{ + }\,\upeta0 3} \right) 2]\,{ + }\,\left( { 3\upeta 2 1{-}\upeta0 3} \right)\left( {\upeta 2 1 \,{ + }\,\upeta0 3} \right) \\ & \left[ { 3\left( {\upeta 30\,{ + }\,\upeta 1 2} \right) 2 { }{-}\left( {\upeta 2 1 \,{ + }\,\upeta0 3} \right) 2} \right] \\ \end{aligned} $$
(85.5)
$$ \begin{aligned}\Phi 6 { } = & \, \left( {\upeta 20{-}\upeta0 2} \right)\left[ {\left( {\upeta 30{ + }\upeta 1 2} \right) 2 { }{-}\left( {\upeta 2 1 { + }\upeta0 3} \right) 2} \right] \\ & { + 4}\upeta 1 1\left( {\upeta 30{ + }\upeta 1 2} \right)\left( {\upeta 2 1 { + }\upeta0 3} \right) \\ \end{aligned} $$
(85.6)
$$ \begin{aligned} \Phi 7 { } \,= & \, \left( { 3\upeta 2 1-\upeta0 3} \right)\left( {\upeta 30{ + }\upeta 1 2} \right)[\left( {\upeta 30{ + }\upeta 1 2} \right) 2\\ & - 3\left( {\upeta 2 1 { + }\upeta0 3} \right) 2]{ + }\left( { 3\upeta 1 2-\upeta 30} \right) \, \left( {\upeta 2 1 { + }\upeta0 3} \right) \\ & \left[ { 3\left( {\upeta 30{ + }\upeta 1 2} \right) 2 { } - \left( {\upeta 2 1 { + }\upeta0 3} \right) 2} \right] \\ \end{aligned} $$
(85.7)

2.5 Identification

To obtain the identification accuracy of our palmprint system, each of the palmprint images was matched with all of the palmprint images in the database. A matching is noted as a correct matching if two palmprint images are from the same palm. If we suppose the query image is an X, and images in the database are N, we must compare X with N.

3 Results and Discussion

The main objective of this study is to obtain higher authentication accuracy by using a proposed algorithm after performing all previous stages. In this stage, 50 users from the CASIA database had features of their palmprints extracted using the invariant moment algorithm. This stage is considered fundamental in building the proposed system of recognition. To find out how accurate this algorithm is, we need to determine the accuracy rate of the identification phase. For determining accuracy rate, we have used two factors, which were FRR and FAR.

3.1 Results Based on Wavelet Transform

After applying wavelet transform (2D filter bank). The output is images that had been decomposed into four bands (LL, LH, HL, HH). Figure 85.6 shows an example of decomposed palmprint images.

Fig. 85.6
figure 6

Samples of decomposed palmprint for five users

Full size image

3.2 Results Based on Moments Algorithm

After applying wavelet transform on palmprint images and decomposing these images into four bands (the output from the previous stage), The invariant moments algorithm was applied for deriving seven moments from each band. Table 85.1 shows values of the seven moments of four bands for one palmprint user (Table 85.2).

Table 85.1 Values of the seven moments of four bands for one palmprint user
Full size table
Table 85.2 Final seven moments which were derived from Table 85.1 for one palmprint
Full size table

3.3 Accuracy Rate of Identification Phase

For the purpose of calculating the accuracy rate of the proposed method to perform the identification phase by matching query palmprint image with each one stored in the database based on FRR and FAR. To calculate the FRR factor, samples of five sets were taken, the size of each set was 15 palmprint templates. These samples belong to the same samples stored in the database. To calculate the FAR, samples of five sets were taken, the size of each set was 15 palmprint templates. These samples were of out the stored database. Finally, accuracy rate was calculated based on FRR and FAR as: Accuracy = (100 − (FRR + FAR)/2).

According to the results in Table 85.3, it was observable that the highest accuracy was at set no. 3, 4 and 5, with a value of 100 %. The results had the highest FRR, which was 6.66 %, which means, just one palmprint of the fifteen valid users was rejected, and with the same high rate (6.66 %) of the FAR, means that one of fifteen invalid palmprints was accepted. The accuracy rate is 98 % (Fig. 85.7).

Table 85.3 Accuracy type
Full size table
Fig. 85.7
figure 7

Accuracy rate for CASIA database of identification phase

Full size image

4 Conclusion

All methods used in this study are based on important steps in the processing and recognizing of palmprints. The major important steps include preprocessing, feature extraction, and matching. All of the aforementioned processes were carried out in a variety of ways and algorithms, which are suitable for the structural condition of the image itself. The results of the experiment show that the overall goal of using invariant moments on the basis of the wavelet transform is successfully implemented on palmprints. It was found that moments of extracting palmprint to be invariant to rotation, translation and scaling. Transformation was applied to decompose palmprint images, that made moments more stable implemented to achieve its goal.