Iris recognition model based on Principal Component analysis and 2 level Haar wavelet transform: Case study CUHK and UTIRIS iris databases

A biometric recognition system provide automatic identification of human being based on some special and unique physical or behavioral features of the individual. One of the most reliable identification system is iris recognition system. This work aim to recognize and identify iris among many of images that have been save in databases. Each one of database that used manipulating in many steps starting with enhance the details of iris and segment the iris and pupil then extract the raw features based on 2D Haar wavelet transform to capture both global and local features of iris image. After that by Appling reduction step to select only the useful and unique features that belong to each person. In this work PCA used as a reduction method. Finally the minimum distance are used to check the similarity between the database’s features training set and input image, also three similarity techniques are used between input iris image and the template that save in database. Weighted Euclidean distance, Hamming distance and Cosine distance. Cosine method achieve good result than other method by using reduction and without reduction.


‫ك‬ ‫مجلــــــــة‬ ‫التربيــــــة‬ ‫ليــــــــة‬ ‫العشرون‬ ‫و‬ ‫السابع‬ ‫العـــــــــــــــدد‬
that device as an interested device belong to the user, then the user can enter user name and password for login. Common methods include entering a code sent to a user's mobile phone through a text message contain the passcode, entering a code using the Google Authenticator smartphone app. Remembering password or keeping smart card or physical security key from unauthorized people are difficult, compare with biometrics based authentication is highly preferred system. Some companies start adding biometric authentication iris, fingerprint and face to their product devices like apple and Samsung companies [1]. Pattern recognition based on iris of human eye for authentication is unique and more suitable method, but face various challenges in the recognition phase's viz. acquisition of high quality image, segmentation of the concentric boundaries, select the proper features that belong to each person and matching [2]. In this paper, a new scheme has been proposed with some addition features for iris recognition consist of multi-layer haar wavelet transform, used the coefficient after reduction by using PCA.

2-iris databases
In this paper work two databases are used to test the recognition method first is:

3-Proposed frame work
The proposed schema of iris recognition system consist of enrollment and verify phases. Each one of them contain four steps starting with resize and enhance the iris image, then iris localization and normalization, select set of features and check the similarity with the features dataset.

3-1 Enrollment phase
The first part of iris recognition system involves five steps: preprocessing, iris localization and normalization, features extraction and selection then features transformation by create features vector dictionary [5] [6]. The figure number 2 show the steps of enrollment part.  In the first step the preprocessing are applies on all images in both databases by uniformed the image's size and dimensions then convert the images into gray scale after that adaptive histogram equalization are applied to enhance the images and to be ready to detect the edge by applying canny edge detection algorithm to detect the boundary of all eye. Then iris localization is applied to detect the edge of iris and pupil from the input image as shown in figure number three. The iris region can be approximated by two circles first one for pupil and second for iris as shown in figure three. The eyelids and eyelashes normally occlude the upper and lower

‫ك‬ ‫مجلــــــــة‬ ‫التربيــــــة‬ ‫ليــــــــة‬ ‫العشرون‬ ‫و‬ ‫السابع‬ ‫العـــــــــــــــدد‬
parts of the iris region [7]. There are many techniques that used for iris segmentation the most knowing techniques is Hough transform is a standard technique in computer vision algorithm that have been used to detect the simple geometric objects such as circle as in iris case or line in image. An automatic isolation algorithm based on the circular Hough transform is employed by many researcher. The segmentation method starting firstly, an edge map of the iris image is generated by calculating the first derivatives of intensity values in an input image then thresholding the result [8]. The parameters that used in this method to determine the circle geometric are the radius r and the center coordinates x c and y c .Which are able to define any circle according to the equation number 1.
.. (1) To extract iris region from the images and normalized into rectangular shape will applied Daugman's rubber sheet model as shown in figure 4. The rectangular form proper for features vector generator. This method implemented by remaps the homogenous of images by remapping each pixel within the iris region to a pair of polar coordinates (r, θ ) where r is on the interval [0,1] and θ is angle [0,2π].

‫ك‬ ‫مجلــــــــة‬ ‫التربيــــــة‬ ‫ليــــــــة‬ ‫العشرون‬ ‫و‬ ‫السابع‬ ‫العـــــــــــــــدد‬
The represented of the iris region from (x,y) Cartesian coordinates to the normalized non-concentric polar representation is modelled as (2) Where Where (x,y) are the original Cartesian coordinates, I(x,y) is the iris region image, (r, ) are the corresponding normalized polar coordinates, and and are the coordinates of the pupil and iris boundaries along the , direction. This method takes into consideration the pupil size to produce a normalize representation with constant dimensions.

3-1-2 Features extraction
The second major part in enrollment phase is features extraction and transformation. After segment iris image and detect boundary of iris and pupil based on Hough transform and normalization the iris region. Now the iris images are ready to extract the features. By applying Haar Wavelet packets as well as the energy of the packets sub images to extract the features of texture. In this work the 2 levels wavelet packet decomposition of Haar wavelet transform are employed to extract the texture of unwrapped iris region image [9]. "In the Haar wavelet transformation method, low-pass filtering is calculated by averaging two adjacent pixel values, whereas the difference between two adjacent pixel values is signed for high-pass filtering. The Haar wavelet applies a pair of high pass and low-pass filters to image decomposition first in image columns and then in image rows independently" [9]. As a first result the image divided into four sub bands as the

3-1-3 features reduction and transformation
When the vector of raw features are extracted and select only the low low part of the haar wavelet transform in this step the reduction are ready to applied to minimize the number of redundancy features to increase the efficiency of recognition by choice the useful characterize of each iris image in database. In this phase PCA are applied as a reduction method [10]. The selected features will be used to generate new features of each iris image to each person in the database. The new features that will generated are mean, median, standard deviation, variance, skewness , kurtosis ,min and max. To generate this features will used the following equations: Figure

3-2 Verify phase
The second part of the iris recognition system is verifying phase, in this phase the unknown entering iris image will be process to check the similar pattern from the dataset that contain all most known iris image. The verify stages can be shown in figure number 6. The first step in this phase is preprocessing to enhance the query image to be ready to extract the features by using 2 levels of Haar wavelet transform and using the coefficients as a set of features and by using PCA to reduce the unwanted features. In this steps the features are transform into another pattern by select only the useful features that can be help in recognition part. Then when all the features of the unknown iris are generated, now the matching phase is ready to found the similar or closer result with the dataset. In this work will used two methods of similarity measurement: Weighted Euclidean distance and Hamming distance, and compare the result of them in the experiment part.

3-2-1 Matching measurement
To check the similarity between the query unknown iris image and the pattern in the dataset, will use two methods to found the best matching result.

1-Weighted Euclidean distance
It is a technique can be used to compare between two vector of features with same size represented as a vector, the WED gives a measure of how set of values are Figure. 6 verifying phase of iris recognition system

‫ك‬ ‫مجلــــــــة‬ ‫التربيــــــة‬ ‫ليــــــــة‬ ‫العشرون‬ ‫و‬ ‫السابع‬ ‫العـــــــــــــــدد‬
closer or similar, between two vectors of features [11] [10]. The Weighted Euclidean distance can be calculated as an equation number (7)   (7) Where f i is the i th feature of the query input iris.
is the i th feature of the iris template that save in dataset. is standard deviation of the i th feature of the iris template k. The unknown iris found the matching template k , when WED is a minimum at k.

2-The Hamming distance
It is one of the similarity measurement methods gives a measure of how many bits are similar between two vectors of bits. Clearly the Hamming distance cannot be negative, and if it is zero, then the vectors are identical. The distance does not depend on which of two vectors we consider first, the Hamming distance, HD, is "defined as the sum of the exclusive-OR between X and Y bits over N, the total number of bits in the bit vector" [11]. " (8)" In some cases the Hamming distance measure value between the two patterns equal to half. This case occurs because independence implies the two bit patterns will be totally random, so there is half chance of setting any bit to 1, and vice versa [12].

3-Cosine distance
The Cosine distance measure the similarity between two vector with non-zero vector integer or Boolean component, in the space a point may be through of as a direction. The cosine measurement distance between two vectors is an angle between point and vector that make. This angle's degree between 0 to 180, depend on how many dimensions the space has. Where the Cosine of 0 is 1 and it is less than one for any other angle in positive space. If we have two vectors with the same values a cosine similarity of 1,and two vectors at 90° have a similarity of 0, and two vectors diametrically opposed have a similarity of -1 [11]. The equation number (9) show the formula of Cosine distance.

4-Experiment result
The entire proposed frame work of iris recognition system was implanted by using Matlab R2015a. The experiment of this proposed work was investigated with two different databases CUHK and UTIRIS. The first dataset consist of 254 iris images for left and right eye from 36 person. And the other contain 1540 eye image collected from 79 person for left and right eye. Different experiment were performed for those two databases. For CUHK database seven iris's image belong to each person select four images from them for training set and the remain used for testing.
The first experiment was implemented to determine the pupil and iris circle for both databases by using Hough transform with different radius and center. And the accuracy of the detect pupil and iris shown in table number 1.

5-Conclusion
This work are presented iris segmentation and recognition system, which used two iris databases CUHK and UTIRIS gray scale eye image .In this paper work the automatic segmentation are achieved by using Circular Hough transform method for iris segmentation and pupil detection and to enhance the iris images in both databases adaptive histogram equalization are applied on all images. In the second part the features are extracted based on Haar wavelet transform to extract the raw features of iris gray scale image for many reasons it's simple and fast to compute and need less memory compare with another methods of wavelet. By using PCA as a reduction method to select only useful features the recognition rate are increasing. Depend on the minimum similarity measurement are used the experiment result show the Cosine distance score good result more weighted Euclidean distance and Hamming distance .in conclusion Haar wavelet transform is efficient method for recognition and sensitive to noise under different cases and conditions with Cosine distance.