Iris-Based Biometric Identification Using a Combination of the Right - Left Iris Statistical Features

A combination between the information extracted for both right iris and left iris could increase the efficacy of the biometric recognition systems. In this paper, we propose a biometric identification method based on density of image patterns extracted from human iris images and the combination and comparison of the right iris and the left iris characteristics. The density of the patters approach for processed images can be a new biometric feature used to implement a biometric recognition system with high performance when a small feature dimension is used. In this way, we can maximize the retention of the effective biometric information. The experiments were conducted on the MMU Iris Database containing 225 images of the left eye and 225 images of the right eye. Two morphological Top-hat and Hit or Miss transforms were implemented to find out the particular pattern of pixels. They allow for the enhancement of detail in images. Then, a statistical feature extraction technique is employed to derive the density of the patterns in morphological transformed images. To assess the density of the patterns differences between the right and left iris data groups, the Pearson’s correlation coefficient (PCC) is computed. We reported very good results with a PCC of 0.6164 (strong and positive correlation) for Top-hat morphological operation whilst the Hit or Miss transform returns a PCC of 0.0127 so there is no relationship between the density of the patterns in the right and left irises.


Introduction
Biometrics is the science of recognition and verification or personal identification using biological or behavioral features, traits or distinguishing characteristics of the individual.Biometric identification is the process of associating an identity with the biometric data by comparing it with identities recorded in a database.Among the various biometric features, the iris has proven to be a predominant and popular biometric feature for authentication and identification with a high degree of accuracy and distinctiveness.
Each individual has a unique iris pattern, which does not change over time, but can vary either between the left and right eyes of the same person or between the irises of identical twins.
The textural characteristics of the iris provide unique high-density information that explains why iris recognition is considered the most reliable and accurate biometric identification system available [1].Omran et al. [2] presented an efficient structure for the iris recognition system called IRISNet.It extracts the meaningful features and classifies they automatically without any domain knowledge.Convolutional neural networks (CNNs) are used for this purpose, resulting in a 97.32% identification rate for original images and 96.43% for normalized images.Mayya et al. [3] designed a model for selecting relevant iris features to build a high-performance biometric identification system.The feature extraction step was performed by computing vertical wavelet details and certain statistical metrics of first and second derivatives of the normalized iris image.The most significant features were selected and merged for the classification stage, which was performed using a distance classifier.An accuracy of 100% was obtained for irises segmentation and a recognition rate of 98.7%, as well.The results confirmed the robustness of the segmentation algorithm to the variations in the illumination and rotation and to eyelash and eyelid occlusion.Zhao and Kumar [4] proposed a deep learning technique for accurate iris detection, segmentation and recognition.CNNs were used to generate spatially appropriate iris feature descriptors.The iris detection accuracy ranges from 94.4% to 96.8%, for three studied databases.Sindu et al. [5] presented an algorithm able to extract the individual iris features by using combined biological features.The proposed method used the crypt, pigment layer and Wölfflin node features of the iris.The selection of the individual features increases the reliability and accuracy of the recognition system compared to other biometric systems.Fang et al [6] studied the correlation between a person's left and right iris employing a VGG16 convolutional neural network and using the iris texture features as input samples.They investigated whether both irises (left and right) belonged to the same person or to different people.To this end, the experimental results of the two independent datasets showed a high classification accuracy of 94.67% and 94.83%, respectively.Jusman et al. [7] propose a novel and practical approach for iris recognition using a combination of several algorithms.The method encompasses a segmentation step, a transformation algorithm into polar images, and a quality improving step of the polar images by using a modified Daugman rubber sheet model.The Gray-Level Co-Occurrence Matrix functions were used for texture feature extraction, and a discriminant analysis was used for classification.The recognition rates were higher than 95%, therefore, the proposed algorithm has shown the potential and capability to facilitate the iris recognition and is suitable for the identification process.
This study investigates the structural similarity between the right and left iris of the same person based on pattern density as a texture feature.The main contributions of this paper are as follows: -segmentation of left and right iris from MMU1 database images by using the Otsu binarization; -Hit or Miss and Top-Hat morphological transformations for image quality improvement; -determining the density of the patterns in morphological transformed images; -assessment of the left and right irises similarity based on the Pearson's coefficient.
The remainder of the paper is organized as follows: section 2 presents the image database and details of the used method to obtain the pattern density values in the left and right iris, respectively.Section 3 focuses on the experimental results, followed by the performance evaluation.Finally, Section 4 provides a summary of our research and presents our future intentions.

Database
We used 450 images (225 ocular images of the left eye and 225 ocular images of the right eye, respectively) of 320x240 pixels size.These images belong to the Multimedia University database (MMU1) 1 , which is a public eye image database for iris-based biometrics training models.The experiment was performed in MATLAB R2018a (the MathWorks, Natick, MA, USA).

Methods
The proposed algorithm is shown in Figure 1.

Otsu binarization
Segmentation divides the image into areas with a strong correlation in terms of the presence of objects in the image [2].The Otsu thresholding method takes the maximum inter-class variance between the background and the target image as the threshold selection rule in consideration.The Otsu method assumes that the image under study contains two classes, foreground and background.The threshold is used to maximize their inter-class variance so that the background and foreground become clearly visible.To select an optimal threshold, a measure of the homogeneity of the gray distribution in the region of interest, called variance, is used.For each potential gray level threshold t, the following steps are considered: -Divide the pixels into two classes based on the threshold value: C1 ∈ [0, 1,..., t] and C2 ∈ [t+1,..., L-1], where L=255.-Compute the probability and mean of the class C1 and C2.
-Determine the inter-class variance based on the maximum judgment criterion, to obtain the threshold t.The used statistical parameters are as follows [8]: The probabilities associated with the classes: () where 1() is weighting of background class, 2() is weighting of foreground class and P(i) denote the probability of occurrence of the grey level i.
The means of the background and foreground classes: The variance between classes to determine the optimal threshold: The figure below represents a conclusive example of the difference between the raw image and the binarized image.

Morphological transforms
The human iris contains a large amount of information which leads to a variety of features.Morphological operators were applied to binary and grayscale eye images to identify patterns in the iris.The morphological operators map each image and improve the quality by reconstructing damaged areas.They also obtain iris's structures which facilitate the local features extraction [9].Morphological methods apply a structuring element to an input image and give an output image of the same size.In binary morphology, an image is represented by a scalar function A(x, y) and the structuring element B is a small image or array of pixels, that determines the neighborhood relationship of the pixels in terms of shape analysis.Dilation of an image A with a structuring element B, denoted A⊕B, replaces the gray level values of the image with the maximum value determined within a mask defined by B: The erosion operator reduces the size of objects.It removes very small details of that image.Usually, the bright areas surrounded by dark areas shrink while the dark areas surrounded by bright areas grow in size.The final image appears darker than the original image.

 
The opening operator implies the application of the erosion and dilatation operations on the same image, while the closing operator works in reverse.Both operators smooth the edges of an image.Opening is useful in removing noise while closing removes the small holes inside the foreground objects, or small black dots on the object.The opening operation, denoted by •, as well as the closing operation, indicated by the symbol •, are defined by [10]: Top-Hat morphological transformation is used to improve the image contrast.This transformation acts as a high-pass filter and extracts bright areas of the image that are smaller than the mask suggested by the structural element B. The Top-Hat transform is achieved by subtracting the aperture of the original image from the actual image [11]: Hit or Miss morphological transformation is a basic tool for shape detection or pattern recognition.It is a combination of two erosion operations and results in template matching, in which an input is matched to a template or mask containing a sub-image of interest.The hit-or-miss transformation of A by B is denoted as [11]:

Results
We adopt the commonly used Pearson's correlation coefficient (PCC) to measure the results of image segmentation and morphological processing for right and left iris images in MMU1 dataset.The statistical correlation through the instrumentality of PCC allows a sound analysis of the similarity between the left and right eyes [12]: where X = x1, x2,..., xn and Y = y1, y2, ..., yn are the measured values, and x ̅ , y ̅ are the sample means of the respective series.Figure 4 shows the qualitative results of four images, randomly selected from dataset.We compared the performance of morphological operators and Table 1 shows the PCC values.The iris recognition system with PCC close to 1 is considered to be a more accurate one.A high correlation between the left and right iris for images processed using the Top-Hat morphological operation is reported.On the other hand, a very low correlation, almost non-existent, of the images processed with the Hit or Miss transform is achieved.These results suggest that Top-Hat method can achieve superior performance in preserving the salient features of the input images due to its very good contrast improvement ability.This operator uses the neighborhood ranking from two different size regions and highlights white object against dark background.The brightest value from an interior region is compared to the brightest value in a surrounding region.If the difference exceeds a threshold level, the analyzed area is kept.On the other hand, the Hit or Miss operator removes more local patters and did not show a good performance of the iris recognition system.

Conclusions
In this study, we propose a biometric identification method based on density of image patterns extracted from human iris images and the combination and comparison of the right iris and the left iris characteristics, as well.The method is based on image binarization, morphological transformations, and PCC statistical correlation for density of the patterns, for both left and right iris images.
The results show a very good PCC performance rate of 0.616 for images processed for the Top-Hat morphological operation.This morphological operation highlighted the texture features, which led to the objective calculation of the pattern density.In our future work, we plan to investigate other methods to improve the iris image quality in order to identify possible diseases related to the iris features.

Figure 1 .
Figure 1.The framework of the proposed approach.It consists of the following steps:  import and cropping images;  segmentation thorough the Otsu's thresholding method, at is a global adaptive binarization threshold image segmentation algorithm;  improving the image quality by using the Top-Hat and Hit or Miss morphological transformations;  compute the density of the patterns in the left and right iris images, for images processed with both Hit or Miss and Top-Hat operators;  compute the Pearson's coefficient for morphologically transformed images.

3 .Figure 3 .
iris's structural changes produced by the two morphological transformations are depicted in figure a) Image after Otsu segmentation; b) Image after segmentation and Hit or Miss morphological processing; c) Image after segmentation and Top-Hat morphological processing.

Figure 4 .
Figure 4. Images processing results based on the proposed algorithm.

Table 1 .
Pearson's coefficient for each correlation type.