Iris recognition approach for identity verification with DWT and multiclass SVM

Introduction

When accessing secured sites such as ATMs, airports, and datacenters, individuals can prove their identities using standard security systems based on knowledge (e.g. passwords) or possession (e.g. tokens), but they are vulnerable to loss, theft, or guessing (Bhattacharyya et al., 2009). As a result, the researcher focuses his attention on biometrics. In recent years, biometrics has grown in prominence as a result of security concerns. Biometrics is one of scientific branches that uses computer technology to verify or identify individuals based on their physical and behavioral attributes, physiological traits that include the face, iris and retina (El-Sayed, Hassaballah & Abdel-Latif, 2016), as well as fingerprint and palm print, which are related to human’s body parts. Speech, gait, and keystroke are examples of behavioral characteristics that are related to an individual’s actions. Physiological characteristics are also more precise and provide more consistent findings (Jain, Ross & Prabhakar, 2004). Due to its unique composition, the iris is the most ideal physiological biometric characteristic for high security applications (Daugman, 1993; Bowyer, Hollingsworth & Flynn, 2008). This peculiar composition makes every iris unique. Even a person’s right iris is distinct from his or her left (Soliman et al., 2017). There is no genetic breakthrough in the iris makeup. The iris pattern is highly random and does not alter over a person’s lifetime (Muron & Pospisil, 2000). In recent years, iris recognition has proven to be one of the most reliable biometric technologies (Wildes, 1997; Daugman, 2009). Iris minutiae have a characteristic appearance. It’s sophisticated enough to be used as a signature (Daugman, 1994).

As a result, finding two persons with identical iris patterns is impossible (Boles & Boashash, 1998). As a consequence, the iris is commonly used in the identification process. Iris recognition techniques can achieve higher precision as compared to other biometric methods. In iris recognition, errors and rejects are uncommon, and the rate of error is the lowest of all the biometrics branches (Daugman, 1993). The iris recognition concept was first introduced in 1939 by Dr. Frank Burch, and it was used in 1990 for the first time when Dr. John Daugman developed its algorithms. Iris recognition as a biometric technique for identification has been a popular focus of research since this time. Iris recognition has been the subject of a number of researches, both for detection and verification, each with its own collection of benefits and disadvantages. Some of these studies had a high recognition rate but needed more storage due to the large number of extracted features, while others took longer to identify subjects. We focus on extracting only the most effective key features that can lead the system to correct classification in this paper. The first step in iris recognition is iris segmentation, which is followed by normalization and enhancement. To minimize the number of dimensions in which they are displayed, the segmented iris images are subsequently subjected to a feature extraction process. For this phase, the use of (DWT) and Gabor wavelet to extract the key features of the iris was recommended in order to improve the performance of the Iris Recognition System (IRS) and the storage of iris templates. SVM was eventually used to classify the process.

The major aim of this research work is to design an efficient biometric system based on various iris features to ensure precise identification of human. The limitations of traditional biometric system such as high false alarm rate are overcome by improving the accuracy of identification using deep learning techniques.

The following is how the paper is structured: In the second portion the Iris Structure is defined. The third portion introduces the “Relevant Literature”. The proposed approach is described in “Materials and Methods”. The findings of the experiments performed to test the proposed approach are found in “Results and Discussion”, and the conclusion is found in “Conclusions”.

Iris Structure

The iris is a delicately colored component of the eye that is made up of several layers. Because of its pigmented properties, the epithelial layer makes the iris opaque. The pigment tissue, blood veins, and two iris muscles contract the pupil on the stromal surface (Kim, Kim & Kium, 2008).

The Collarette—which has a zigzag pattern—divides two zones: outer ciliary and inner pupillary (area) zones, Fig. 1 illustrates the eye image. Arching ligaments, ridges, furrows, crypts, rings, corona, freckles, and other distinguishing and distinctive features can all be found in the compound and multiplex arrangement. Iris characteristics are unpredictably distributed.

Related Literature

Guo & Zheng (2018) suggested a gray projection algorithm for establishing identity authentication. They used improved Daugman and Point Hough Transform (PHT) algorithms to place the outer and inner boundaries, respectively. Following that, the iris image was normalized using line segment extraction, and the wavelet transform was used to remove the features. Finally, the Hamming Distance between two irises was measured in order to estimate the distance between two irises so that the iris could be recognized. The iris recognition system exhibited has a high level of robustness and is resistant to interference. The device has a high degree of accuracy, according to the trial results.

Meetei & Begum (2016) applied a BioHasing-based annullable iris biometric to iris biometric. The iris picture is segmented and normalized using the Hough transform and Daugman’s rubber sheet model. To create iris templates, minutiae are retrieved using DWT, ICA, and PCA. The iris template is converted using the suggested annullable iris biometric. In addition, the transformed iris templates are classified using SVM and RBF Kernels. Jagadeesh & Patil (2017) used a simple approach of segmentation that included a Gaussian filter followed by a canny edge detector following the pre-processing step. It makes it simple to locate the essential borders of the eye from the iris boundary. The Curvelets transform is used to extract features from segmented images so that the iris can be represented in significant dimensions. SVM was used to classify the data. Vyas et al. (2017) developed a unique iris recognition approach based on gray level co-occurrence matrices (GLCM) for feature extraction and a neural network as a classifier. The proposed technique maintained a high level of accuracy.

Kaudki & Bhurchandi (2018) offer a unique technique for iris detection based on fuzzy logic edge extraction and iris localization utilizing the Circular Hough Transform (CHT). Due to its computational ease, the Haar wavelet transform system is also utilized to extract the features. To compute template matching, the Hamming distance method is utilized. Archana et al. (2018) developed an open-source iris identification system to authenticate each iris’ identity and biometric performance. They used the Hough Transform for iris segmentation and the Gabor filter for feature extraction. Finally, the Hamming distance was employed to identify iris templates. Pattar (2019) proposed a breakthrough model as an enhancement of Chan-Vese technique by including Bspline approach for iris segmentation. To extract features in this experiment, the Local Binary Pattern (LBP) and GLCM processes were applied. Finally, SVM was utilized to determine whether or not the individual is approved. The proposed technique was evaluated using the NICE.I iris image database.

Entropy-based Occlusion Removal (Likitha, Gupta & Manikantan, 2015), proposed by Likitha, Gupta & Manikantan (2015), is a new technique for removing eyelashes that is utilized after iris localization. The Dual Tree Complex Wavelet Transform (DTCWT) is used in conjunction with the Discrete Cosine Transform (DCT) to extract shift-invariant features. Binary Particle Swarm Optimization (BPSO), which chooses the significant features extracted by DTCWT and DCT, is utilized for feature selection. The degree of similarity between the two photos is calculated using the Euclidean distance classifier.

Rana et al. (2019) proposed a technique that used Principal Component Analysis (PCA) and Discrete Wavelet Transformation (DWT) to identify key characteristics of an iris while reducing iris classification time. The pictures of iris were classified using SVM.

Deep learning based unified framework is proposed in Zhao & Kumar (2019) for accurate detection, recognition and segmentation of iris from raw eye images. The system architecture includes iris specific Mask R-CNN, normalization layer and feature extraction. This mask is used for iris segmentation. The training samples are along with its ground truth images. Normalization layer is used to perform iris and mask normalization before leaning the features of iris. After detecting iris, contrast enhancement process is proposed which is used to enhance the quality of the images. The features are extracted using fully connected layer of FeatNet. The features are learned by Extended Triplet Loss (ETL) function.

Qasmieh, Alquran & Alqudah (2021) proposed an accurate and fast IRS to address the noisy iris images specifically the noises come from eye occlusion and from specular reflection. It embraces a self-customized support SVM and CNN classification models. It used an optimized images processing techniques like iterative randomized Hough transform (IRHT) for iris region segmentation and used few significant features based on singular value decomposition (SVD) analysis for the classification. It used the Hamming distance to identify the subject’s unique iris pattern with high accuracy, security, and reliability.

Desoky, Ali & Abdel-Hamid (2012) proposed an IR algorithm in which a set of iris images of a given eye are fused to generate a final template using the most consistent feature data. Features consistency weight matrix is determined according to the noise level presented in the considered images. A new metric measure formula using Hamming distance is proposed.

Raffei, Dzulkifli & Rahman (2020) proposed a combination with neural network to improve the ability to match the iris features in mobile iris recognition system. The proposed method, a combination of Hamming distance and neural network has achieved a better result where it increases the existing method in term of accuracy for mobile iris recognition under non-cooperative environment.

Szymkowski et al. (2020) proposed retina diagnosing using biometric methods with SVM and clustering methods. The main aim of this research is to extract the features from retina images. First process is preprocessing which includes, denoising, contrast enhancement and normalization for improving image quality. Next process is vessel segmentation, binarization and thinning which is done by Gaussian matched filter with binarization using local entropy thresholding. Final process is minutiae extraction which is done by crossing number algorithm. After extract the minutiae the proposed method automatically counts the number of minutiae which helps to diagnose the disease. SVM is used to classify the detected eyes are healthy or unhealthy.

Li et al. (2021), authors proposed a segmentation method for precise segmentation of iris images by incorporating deep learning model. The importance of proper segmentation of iris for accurate recognition was addressed. The reasons for noises in the input image such as low resolution, light reflections, blur, motion and occlusions affecting the accuracy of segmentation was analyzed. The interleaved residual U-Net (IRU-Net) was implemented to construct the outer boundary and inner boundary of iris based on salient features. The scanning and clustering of salient features was performed to construct the external boundary whereas the inner boundary was constructed by intersecting the salient features obtained from the deep learning model.

Materials and Methods

In general, there are four processes to iris recognition: (1) Acquisition of the iris Image; (2) pre-treatment (Segmentation, normalization and Enhancement): (3) extraction of features: (4) then there's the process of matching. The suggested approach can be explained as exhibited in Fig. 2, in which an iris template is formed by performing the required steps of iris recognition and then compared to the templates stored in the iris database. The segmented iris area is normalized to a rectangular form with firm polar dimensions using Daugman’s rubber sheet model. The characteristics were extracted using a Gabor Wavelet and DWT combination. For classification, a multiclass SVM was utilized.

Hough transformation

In this study, the Hough Transformation was used to determine the iris area. This technique was used to define the parameters of geometric shapes in an image, such as circles and lines. The radius, center, and pupil area of the iris may all be calculated using the circular Hough transform. The Hough algorithm creates an edge map by computing the first derivatives of intensity values or gradients in an eye image. The surrounding points on the circle at various radii are taken for each edge pixel in the edge map, and votes are cast to acquire the maximum values that create the parameters of circles in the Hough space. The parameters x_c, y_c which are the center coordinates and the radius ‘r’ can be found through the equation: $x_c^2+y_c^2-r^2=0$. The maximum point corresponds to the center coordinates ( $x_c$, $y_c$) in the Hough space, and the radius of the circle can be given by the edge points. We used vertical gradients to determine the outer boundary (iris-sclera) and to reduce the impact of the eyelids, which were horizontally oriented to determine the iris boundary and vertical gradients to limit the impact of the eyelids. All of the pixels on the edges are altered when a circular Hough Transformation is applied. In the Hough space, there are fewer edge points from which to cast votes, which not only improves circle localization accuracy but also efficiency (Rana et al., 2019).

Iris normalization

The normalization procedure produces iris areas with constant size after properly segmenting the iris area. This technique can be carried out using the Daugman rubber sheet model (Daugman, 1994; Ma et al., 2004; Podder et al., 2015). The reference point is the pupil center, and radial vectors run across the iris area, as illustrated in Fig. 3. Angular resolution refers to the radial lines that circle the iris area. Because the pupil is not concentric with the iris, rearranging points based on the orientation around the circle necessitates the use of a basic formula. The grey values of these pixels identify the entire iris area. This information’s can be determined by coordinate’s combines of inner and outer boundaries. This model rescales each point inside the iris region to a pair of polar coordinates (r, θ) where ‘r’ lies in the interval [0, 1] and ‘θ’ is an angle in the range of ‘0’ to ‘360’ (2 $\pi$) degree.

So. If s(x, y) is an iris image represented in Cartesian coordinates and s(r, θ) is the representation in polar coordinate. And if (x_i, y_i) and (x_o, y_o) is the inner boundary and outer boundary unit in Cartesian coordinates respectively, then

(1) $$s\left(x\left(r,\theta \right),y(r,\theta )\right)\to f(r,\theta )$$

(2) $$x\left(r,\theta \right)=\left(1-r\right)x_i\left(\theta \right)+r{x}_o(\theta )$$

(3) $$y\left(r,\theta \right)=\left(1-r\right)y_i\left(\theta \right)+r{y}_o(\theta )$$

In the above equation, $r=\frac{P}{M+1}$, where $p=1,2,\dots M$ and $\theta =\frac{q}{N}$, $q=1,2,\dots N$.

The sample rate, as well as the angle and radial direction, are represented by M and N, respectively. The following is a description of the normalizing algorithm:

• Step 1: Based on iris boundary localization of iris image $s\left(x,y\right)$, calculate the parameters of $(x_i,y_i,r_i)$ and $(x_0,y_0,r_0)$. Where the subscript i and o means the inner and outer boundary.

• Step 2: The following formula can be used to calculate the distance between the pupil’s center and the iris’s center: $r\mathrm{\prime }=\sqrt{{(x_i-x_o)}^2+{(y_i-y_o)}^2}$

• Step 3: Connection direction angle is also calculated. $\text{φ}=arc\tan \left(\frac{y_i-y_o}{x_i-x_o}\right)$

• Step 4: Choose the pupil’s center as your pole. In the polar system $r(\theta )=r_p$. For the outer boundary of the iris, $\theta =q\frac{\pi }{180}$, where $j=1,2,\dots N,R\left(\theta \right)=r^{\mathrm{\prime }}\cos \left(\pi -\theta -\text{φ}\right)+\sqrt{r_o^2-r^{\mathrm{\prime }2}+(r^{\mathrm{\prime }}\cos (\pi -\theta -\text{φ}{)}^2)}$

• Step 5: The grey values of the normalizing iris for each pixel can be calculated using those grey $\left(x,y\right)$ coordinates and the following equations:

(4) $$R_p=(1-\frac{p}{M+1})\ast r\left(\theta \right)+\frac{p}{M+1}R(\theta )$$

(5) $$x=X_p+R_{p}cos\theta ,y=Y_p+R_{p}sin\theta$$

Enhancement

Depending on the application, enhancement is a technique that is used to increase the image quality. In iris recognition, the low contrast image of a segmented iris must be corrected. As a result, Adaptive Histogram Equalization, Image Adjustment and Image Sharpen are applied in that order. Photographs at various levels are depicted in Fig. 4. It is observed from that figure how features in image is enhanced in each stage of preprocessing.

Feature extraction/encoding

The feature extraction procedure is a type of dimension reduction that is unique. Feature extraction is the process of converting the input data into a set of features. When the features are carefully selected, the features collection is anticipated to extract the critical information from the input data in order to complete the required procedure using this reduced representation rather than the full-size input. Analyze the Principal Components (PCA), Kernel PCA, Linear Discriminate Analysis (LDA), and SVM are examples of prominent pixel-based feature extraction algorithms that produce good results. The fundamental drawback of these methods is that features are inversely proportional to speed and proportionate to accuracy. It means that as the number of selected features increases, the probability of accuracy increases as well, although the rate of speed decreases. It took extra time to calculate additional features. Wavelets devised a third method for identifying the best solution to such issues (Sheikh & Thakare, 2016). The segmented iris portion is used to extract various features. They are as follows: Gabor Features. This is a linear filter that detects edges. This filter representation is similar to that of the human visual system, and it has the ability to reflect texture and discriminate it successfully. As a result, these features are used in this research. Wavelet Features. Wavelet transform is a sub sampling technique that separates an image into several sub bands. Because the image is separated into sub bands, this altered version of the image can extract an n-number of characteristics. Image features such as mean, standard deviation, low band, and high band are called features, and they are merged to generate the ultimate feature vector (Tallapragada & Rajan, 2010). Finally, all of the retrieved features are merged and formed into a high-dimensional feature vector. The purpose of creating such a large feature vector is to improve classification accuracy.

Wavelet transform

The Wavelet Transform will exemplify the signal in terms of time and frequency. It may clarify important aspects of a signal, such as trends, breakdown points, higher derivative discontinuities, and self-similarity (Misiti et al., 2001). The major benefit of the wavelet transform is its local and multi-resolution feature extraction capability. The wavelet transforms often called a ‘mathematical microscope’. The ways by which wavelet transform can be developed are Continuous wavelets transform (CWT) and discrete wavelet transform (DWT).

The Discrete Wavelet Transform (DWT) is a rapid Wavelet Transform calculation method based on sub-band coding. It is simple to use and saves time and resources while performing calculations. When techniques for decomposing discrete-time signals were devised in 1976, DWT was founded (Birgale & Kokare, 2009). As shown in Fig. 5, the DWT is performed by filtering the discrete time-domain signal with successive low-pass and high-pass filters.

When more precise low-frequency information and compact regions are required, wavelet analysis allows the use of long-time intervals. It takes a lot of time and effort to compute wavelet coefficients at all feasible sizes, and it generates a lot of data. As a result, our calculations are limited to a subset of scales and places. It turns out that if we choose scales and positions based on the powers of two so-called dyadic scales and positions, our analysis will be significantly more efficient and accurate. The DWT provided by provides us with such an analysis Eq. (6).

(6) $$DWT=\sum _{k=1}^{\mathrm{\infty }}\sum _{l=-\mathrm{\infty }}^{+\mathrm{\infty }}q\left(k,l\right)\psi (2^{-k}t-l)$$

Filters were introduced in 1988 as an effective approach to implement this method. This approach is known in the signal processing field as a two-channel sub-band coder. This is a helpful filtering approach that constructs a box through which a signal flows and wavelet coefficients emerge fast using a simple wavelet transform, let’s say that

(7) $$\mathrm{\varnothing }(x)=\sum _n{h}_{\mathrm{\varnothing }}\left(n\right)\sqrt{2}\mathrm{\varnothing }(2x-n)\mathrm{a}\mathrm{n}\mathrm{d}$$ (8) $$\psi (x)=\sum _n{h}_{\psi }\left(n\right)\sqrt{2}\psi (2x-n)$$

$\mathrm{\varnothing }(x)$ and $\psi \left(x\right)$ can be expressed as linear combinations of double-resolution copies of themselves. Here $h_{\mathrm{\varnothing }}$ in (7) and $h_{\psi }$ in (8) the expansion coefficients are called scaling and wavelet vectors, respectively. They are the filter coefficients of fast wavelet transform (FWT), $W_{\mathrm{\varnothing }}\left(j,m,n\right)$ Approximate coefficients, $W_{\psi }^H\left(j,m,n\right)$ Horizontal coefficients, $W_{\psi }^V\left(j,m,n\right)$ Vertical coefficients and $W_{\psi }^D\left(j,m,n\right)$ Diagonal coefficients. Here $W_{\mathrm{\varnothing }}\left(j,m,n\right)$ the original image whose DWT is to be computed (Birgale & Kokare, 2009).

Gabor wavelet

Gabor wavelets (filters) have frequency and orientation representations that are akin to those of the human visual system, and they were created with texture representation and discrimination. Pattern analysis applications make extensive use of Gabor filters (Shen, Bai & Fairhurst, 2007; Liu & Wechsler, 2002; Meshgini, Aghagolzadeh & Seyedarabi, 2013). Another notable advantage of Gabor filters is their invariance to lighting, rotation, size, and translation. Photometric disturbances like light variations and picture noise are also tolerated. In the spatial domain, a two-dimensional Gabor filter is defined as a Gaussian kernel function modulated by a complex sinusoidal plane wave, as shown:

(9) $$G\left(x,y;\lambda ,\theta ,\psi ,\sigma ,\gamma \right)=\frac{{\lambda }^2}{\pi \gamma }\exp \left(\frac{-1}{2{\sigma }^2}\left(x^{\mathrm{\prime }2}+{\gamma }^2{y}^{\mathrm{\prime }2}\right)\right)\exp \left(i\left(\frac{2\pi x^{\mathrm{\prime }}}{\lambda }+\psi \right)\right)$$where,

(10) $$x\mathrm{\prime }=x\cos \left(\theta \right)+y\sin \left(\theta \right),\begin{array}{c}y\mathrm{\prime }=-x\sin \left(\theta \right)+y\cos \left(\theta \right)\hfill \end{array}$$where $\lambda$ is the wavelength of the sinusoidal factor, $\theta$ represents the orientation of the normal to the parallel stripes of a Gabor function, $\psi$ is the phase offset, $\sigma$ is the standard deviation of the Gaussian envelope and $\gamma$ is the spatial aspect ratio which specifies the ellipticity of the Gabor function’s support. Log Gabor filters (Shen, Bai & Fairhurst, 2007; Kumar & Rao, 2014) are used in the proposed algorithm, and their frequency response is as follows:

(11) $$G\left(f\right)=exp\left(\frac{-{(log(f/f_o))}^2}{2{(log(\sigma /f_o))}^2}\right)$$where $f_o$ represents the center frequency, and $\sigma$ gives the bandwidth of the filter.

Generating features

1. Practically, the function waveletTransform utilities a wavelet transform and dwt2 function in matlab form and generates 40 features. An image inputs to process and extract wavelet coefficients. The output is 1 × 20 feature vector containing the first 2 moments of wavelet coefficients (mean and standard deviation coefficients).

2. The following Function for computing Gabor features of a gray-scale image This function calculates Gabor features and generates 48 features, 24 features generated by Mean Amplitude for each scale and orientation is returned, and 24 features generated by Square Energy. Mean-squared energy. There are potentially many arguments, here is all parameters have defaults and the usage can be as simple as gaborWavelet(im); For maximum speed the input image should have dimensions that correspond to powers of 2, but the code will operate on images of arbitrary size.

• Filters are constructed in terms of two components. The radial component, which controls the frequency band that the filter responds to. The angular component, which controls the orientation that the filter responds to. The two components are multiplied together to construct the overall filter. Construct the radial filter components. First construct a low-pass filter that is as large as possible, yet falls away to zero at the boundaries. All log Gabor filters are multiplied by this to ensure no extra frequencies at the ‘corners’ of the FFT are incorporated as this seems to upset the normalization process when calculating phase congruency. For each point in the filter matrix calculate the angular distance from the specified filter orientation. To overcome the angular wrap-around problem sine difference and cosine difference values are first computed and then the atan2 function is used to determine angular distance.

Results and Discussion

Images dataset

The proposed iris recognition system is evaluated in this paper using one of the most widely available datasets: the IITD V1.0 dataset (Kumar & Passi, 2010; https://www4.comp.polyu.edu.hk/~csajaykr/IITD/Database_Iris.htm) which contains iris photos from 224 classes (users). The bitmap (*.bmp) format is used for all of the images. All of the subjects in the dataset are between the ages of 14 and 55, with 176 men and 48 girls. The database contains 1120 photos with a resolution of 320 × 240 pixels. As a result, there are 224 classes in this dataset, each having five sample iris photos, and all of these images were taken indoors using the near-infrared (NIR) wavelength. Figure 6 shows 16-sample images from IITD Iris dataset.

All experiments are conducted on a Windows-based machine with Intel (R) Core (TM) i7-7500U CPU@2.70 GHz processor, 16 GB RAM, and MATLAB R2017a installed on it. In the verification mode, the proposed system's performance is assessed. In this mode, each iris image is examined as a single test image, with all others balanced. Table 1 shows the relationship between recognition rate and various training/testing ratios. Recognition Rate (RR), Obtained for IITD databases by applying the proposed technique, along with other pre-processors for different training: testing ratios. The higher the training ratio, the higher the accuracy of recognition.

Table 1:

Recognition rate of Gabor wavelet, DWT and proposed method using IITD iris dataset.

Technique	No. of features	Training %	Testing %	RR %
Gabor wavelet	48	20	80	57.232143
		40	60	74.553571
		60	40	86.160714
		80	20	91.339286
DWT	40	20	80	67.053571
		40	60	83.571429
		60	40	90.803571
		80	20	95.892857
Proposed method	88	20	80	75.892857
		40	60	90.089286
		60	40	95.714286
		80	20	98.928571

DOI: 10.7717/peerj-cs.919/table-1

The table also shows the number of features derived using the Gabor wavelet and the Discrete Wavelet Transform. On the basis of the accuracy rate, Table 2 contrasts the proposed system to the other current methods. The proposed method’s efficiency in the verification mode is assessed. Each iris image is treated as a single test image that matches all other images in the database in this mode. The association between the recognition rate and various training and testing ratios is shown in Table 1. The more training ratio, the more recognition rate accuracy. Moreover, the table shows the number of features extracted using Gabor wavelet and Discrete Wavelet Transform. Table 2 shows a comparison between the proposed method and the existing methods based on the accuracy rate, using the same dataset (IITD iris dataset).

Table 2:

Accuracy comparison of our proposed technique with existing methods.

Method	(Mrinalini et al., 2015)	(Vyas et al., 2017)	(Ali, Suandi & Abdullah, 2018)	(Vyas, Kanumuri & Sheoran, 2016)	(Qasmieh, Alquran & Alqudah, 2021)	Proposed method
Accuracy (%)	94.04	97.83	96.5	96.33	97.12	98.92

DOI: 10.7717/peerj-cs.919/table-2

Conclusions

A new iris recognition method is introduced in this paper. For the segmentation job, the device used the Hough Transform process, with the Daugman rubber sheet model for normalization, Gabor wavelets, and wavelet transforms for feature extraction. Finally, multiclass SVM was used with caution to complete the matching job. The proposed approach uses only 1 × 88 vectors to create an iris signature. The speed can be improved by using a signature of this length. The developed approach in this study reduced the number of features while maintaining a high degree of accuracy. The proposed technique’s accuracy is 98.92%, which is higher than some other current approaches for accurate comparison.