Segmentation Approach for a Noisy Iris Images Based on Block Statistical Parameters

The Iris localization plays a big role in the performance of an iris recognition system. This is due to the dependent of the next steps up on it, and the incorrect segmentation might lead to inexact normalization and improper feature extraction from less discriminatory parts (eyelids, eyelashes, pupil, etc.) so the execution of system will diminish. An effective method for locating the iris of the eye is suggested in this paper. At first a mixture of gamma transform and contrast enhancing mechanisms are used to guarantee a precise renovation of eye image to become an iris area easy to isolate. The next step is relayed on calculating the statistical image parameters (i.e., the mean and standard deviation) which are employed as a feature to detect outer iris boundary. The integro-differential operational technique is used with further pre-processing processes to detect the inner boundaries of iris. The noisy iris UBIRIS.v1 dataset was used in the experiment. Thee conducted results indicated that the proposed technique has a good performance, which is improved accuracy of iris localization step for noisy dataset.


INTRODUCTION
Biometrics is a technique that uses physiological or behavioral features to authenticate a person's identity [1] to identify individuals. Biometrics is used instantly based upon different physiological characteristics, including fingerprints, face, movement, iris, ear, and sound. Due to its unique nature and its fixed patter throughout the life of the individual the human iris pattern can be used as an ideal biometrical feature [2], [3]. An iris offers high precision in applications relating to personal identification. These applications represent secure access to the ATM machines bank accounts, border controls at national borders, safe access to condominium buildings, passport controls etc. [4]. In spite of there is a fact that iris patterns may provide high confidence for safe identification and personal MAICT Journal of Physics: Conference Series 1530 (2020) 012021 IOP Publishing doi: 10.1088/1742-6596/1530/1/012021 2 authentication but still many challenges ahead of researchers. One of the key issues of iris recognition research is to achieve distinct features of the iris that may maintain its stability to the users concerned on a large scale and under different interference conditions, such as lighting differences, aging, as well as eyelids and eyelashes effecting.

RELATED WORKS
The most common current methods of iris localization are boundary-based. These approaches at first find the region of pupil as the internal border of the iris, and then discovery the eyelid and limb regions, Isolating the iris from it. The Hough transform (HT) [4] and combined Daugman's integro differential operator (IDO) are these kinds of approaches [5].
Wildes approach used HT which is discovered the circular by voting on the edge-map within a certain range of radius. HT-based detection methods are also used [6], [7]. The integro-differential operator for Daugman , looks for the maximum variance in the pixel's value on circular contours ds with changing the radius r and center (x0, y0) over the whole of the image domain. The Gσ is a smoothing function and the symbol * indecates the convolution. The advanced method was developed in 2012 [8] Despite its high efficiency, when the iris input is noisy, the Daugman operator does not always locate the iris precisely. The noise represented by a reflection image, or not fully eye opened. Jeong et al. [9] in 2010 proposed a technique that aims to decrease error rate in a non-cooperative environment. In combination with AdaBoost, this technique was used to detect pupil and iris borders by two circular edge detectors and this method estimates true eyelid and eyelash boundaries. It is also there are other ways to reduce noise before iris boundaries detection for increasing the accurateness of localization [10]. All these localization methods require high-quality images and an ultimate environment of imaging; consequently, those approaches get less precise or can result with higher error rates in non-ideal situations.
Other methods consist of pixel-based localization approaches. These approaches depend upon defining the border between iris and other pixels in the neighborhood by means of a particular color texture and lighting gradient. Depending on the distinction features an iris designation is made for the identification of iris and non-iris pixels. Khan et al. Khan et al., in 2011 [11] suggested a way to localize iris. They used the two-dimensional profile lines between the iris and the sclera and measured the gradient pixels by pixels in which the iris limit is the most important change. Ashwaq and Duaa, [12] proposed an approach for localizing iris area by using mixture of image processing techniques. The formal features of the iris, like the noise region in several areas of the eye image, are kept in mind (i.e., specular reflections, focus and small visible iris part). The iris occurred and as the iris borders became clear as the boundary area of these two boundaries. However, the overall performance of the iris localization methods is reduced by various constraints such as dark skin, eyelashes, and hair or where the true boundaries of the iris are not detected.
On the other hand, the external iris boundaries appear to be non-circular and non-elliptical in nonzero images such as iris images off-axis. Furthermore, the region of iris localization is not effective since eyelashes and eyelids are incompletely blurred. It is notice that important to use a non-circular method to iris segmentation for overcoming these encounters. For this reason, the active contour is a perfect candidate [13].
Nevertheless, the presence of vulnerabilities in previous active contour models does not make them effective to segment the iris. For example, the active snake contour [14], is sensitized to the initial curve and cannot discover non-axial objects. Though the difficulties related with snake active contour have been mitigated by Xu and Prince [15] in the Active Gradient Vector Folw (GVF), the GVF contour continues to have minimum energy points and stop at the incorrect edges. The standard snake and GVF versions are also built to shrink to the target mark. Therefore, if the primary mask is within the target object the contour reduces and disappears [16].
To address these issues concerning the existing approaches, we propose an approach to localize the iris of the eye where a mixture of gamma transform and contrast enhancement mechanisms are employed to ensure an accurate translation of eye image to become an iris area easy to isolate. The next step is relayed on calculating the image's statistical parameters which are employed as a feature to detect outer iris border. The rest of this article is structured accordingly. The proposed system is described in detail in Section 2. Section 3 discusses the experimental results and discussion. Section 4 includes the conclusions.

PROPOSED System
The localization of iris from the eye image is a significant stage in iris recognition system; it goals to allocate of iris borders accurately. Iris of human is a part between the pupil region and the white sclera region which are represented the inner circle and outer circle respectively as depicted in Figure (  (2) Segmentation of iris which is a technique to detect of the outside border of the iris region.

PUPIL SEGMENTATION
Daugmans integro differential operator (IDO) [4] is one of the primary strategies of iris segmentation, however it struggles to locate the iris center's rough location in noisy environments. A pupil segmentation algorithm is proposed to address this issue. First, many image enhancement techniques, such as contrast stretching and histogram equalization were used to improve the input image before applying the Daugman's operator. Second, the pupil circle is localized using the IDO. By applying the IDO without preprocessing enhancement techniques it is negatively detect pupil boundary which is affected of a noise such as small visible pupil region with light reflections. To eliminate difficulties on the pupil detection, algorithm (1) is applied to the image before applying the Daugman's operator. it was suggested that the areas of light affected will filled by the average pixel intensity in the area around it. Daugman's operator is performed on the eye image processed after algorithm (1)  Step3: Clean image to remove small details using smoothing filter with 3×3 window.
Step4: Convert to binary with threshold equal to 0.35. Srep5: Cut boundary from four direction left, right, top and bottom with ratio 30% of image dimensions. Step6: Collect reflex pixels region coordinates Step7: The light reflex area is filled with the mean of pixels intensities from the neighboring to it.
Step8: Adjust the eye image after filling the light affected region by using Histogram equalization.
Step9: Gamma transform is performed on the I image with α=4 using equation 2.
Step10: Establish the pupil's center (Xc, Yc) and hence its radius Rb was using IDO.

IRIS SEGMENTATION
In this paper an approach is introduced to localize the iris of the eye, it relays on calculating image statistical parameters of eye image. To detect iris pixels, the mean m and standard deviation st values of the 5 × 5 mask are calculated. Next, compare the st and m values for each window with the predefined thresholds. An IrisReg binary image is created where the detection iris pixel is set to zero otherwise it is set to one. If m of the selected window greater than a predefined threshold T1 and st value is greater than threshold T2 the point of the center is considered as iris point and set the IrisReg to zero otherwise it is considered as non-iris point and set to one. Algorithm (2) lists the steps of proposed iris localization. Step8: Perform the region filling to fill light reflex in the iris region Step9: Rotate the resulted image S 180o then put the rotated image in S1 Step10: Add the two image S and S1 before and after rotation IrisReg = S + S1 Step11: Collect the coordinates of the resulted region

EXPERIMENTAL RESULTS AND DISCUSSIONS
The experiment of the suggested iris localization method is tested on an iris image dataset, UBIRIS.v1 [19]. This dataset consists of a set of iris of visible wavelength taken from close range through user collaboration. The UBIRIS.V1 consists of 1877 iris images and the size of each image is reduced to 200 ×150. Figure 6 shows samples of UBIRIS dataset   Table 1 depicts the results of pupil and iris localization after applying the proposed method on the UBIRIS dataset. Table3 shows the time required for iris localization to one image using three methods. As we see the proposed method speed up iris localizing compared with other methods.

CONCLUSIONS
In this article, we present an approach for iris segmentation. The IDO is one of the dominant iris segmentation techniques which is employed for pupil localization after some enhancement techniques is used to eliminate the various types of noise. The developed enhancement process makes the pupil region brightness close to be bi-modal and have distribution for purpose of segmentation. The proposed of outer boundary method relays on calculating statistical parameters for image. The statistical parameters values for each window are compared with Predefined thresholds. This method is robust in light and larger reflection areas because the noise detection is based on these statistical parameters for determining the iris region.