HIGH ACCURACY RECOGNITION BIOMETRICS BASED ON FINGER VEIN SCREENING SENSOR

With the great development in information technology, the commonly used biometrics have become very easy to hack such as face, voice and fingerprint etc. This has led to the discovery of new horizons in this field. One of the safest biometrics of today is finger veinBut this technique faces some unique difficulties, the most commonly used being that the vein pattern is difficult to remove because finger vein images are always low in quality, significantly hampered the feature extraction and classification stages. For this purpose, professional algorithms must be considered with the conventional hardware for capturing fingervein images using Red Surface Mounted Diode ( SMD) led. For capturing images, Canon 750D camera with micro lens is used. For high quality images the integrated micro lens is used, and with some adjustments it can also obtain finger print. Features extraction was used by a combination of Hierarchical Centroid and Histogram of Gradients. Results were evaluated with K Nearest Neighbor and Deep Neural Networks using 6 fold stratified cross validation. Results displayed improvement as compared to three latest benchmarks in this field that used 6fold validation ( 5samples for training and one sample for test) for SDUMLAHMT. The work novelty is owing to the hardware design of the sensor within the fingervein recognition system to obtain, simultaneously, highly secured recognition with low computation time, finger vein and finger print at low cost, unlimited users for one device and open source.


I. INTRODUCTION
Today the Information security is being controlled by the use of biometric technology-the technique of identifying a person based on biological features such as physical or behavioral characteristics [1] Physical biometrics include iris, hand geometry, face, fingerprint and behavioral biometrics include gait, voice, signature, keystroke pattern etc. Most of these features could be vulnerable to hack attempts [2], [3] and for this reason had led to the suggestion of more secure biometrics such as vein patterns within finger [4] hand [5] and palm [6] .These are nearly no way to reach without the user's consent and hence harder to fake. Finger vein recognition is considered as one of the harder tasks because of low image quality, uneven-illumination and differences in temperature. Finger vein identification systems are also secure against spoof attacks [7], [8], [9] and [10]. So, FVR continue to need more fast and proficient methods. The typical framework of the biometric finger vein recognition ( FVR) system is presented in Fig. 1.

II. DATABASE
It is a fact commonly agreed [46], [13] that the hardest finger vein database to preprocess is SDUMLA-HMT database as shown in Fig. 2. The reasons for this non-ideality are many such as low quality of images, Finger area is very small as compared to total image size [14] and Images vary in rotation, translation and shift [15]. In getting of images, a number of people who participate is one hundred and six ( 106). The format of images of vein is bmp of each person, images of both two hands were taken. Data of three fingers were collected i.e. index, middle and ring fingers for each hand [16] and for each finger, 6 samples were obtained. For this purpose, Joint Lab for Intelligent Computing and Intelligent Systems of Wuhan University made a device that was used. In this work finger-vein recognition system is modified by developing a new method for image pick up in the hardware of the device. The images are processed by some of enhancement stages, these stages not only have low computational complexity but also high distinguishing power. To evaluate this work, the collected database by proposed device have been used. A number of people who take part is one hundred ( 100). The in Fig. 3 and Table I.

III. FINGER VEIN SENSOR
In this work the proposed hardware for getting finger-vein images is modified by using red Surface Mounted Diode ( SMD) led on both sides, instead of near-infrared led. For getting images, Canon 750D camera is used with micro lens. To obtain better quality images, the integrated micro lens was used with some adjustments it can also capture finger print. Mobile camera can be used instead of Canon camera for image capturing. The SMD led that are used, have better illumination. The proposed device was used to collect the database of this work. The goal of this work is to increase accuracy and efficiency of finger-vein recognition by reducing the processing time. To achieve this goal, the hardware of finger-vein recognition device is modified by using SMD led instead of near-infrared led, the modified system improves images quality by generating high quality images. Finger vein sensor as shown in Fig. 4.

IV. PROCESSING AND ENHANCEMENT IMAGES
The image acquired is preprocessed in different stages. The proposed method is evaluated as following:

A. Region of interest
Detect the region of interest ( ROI) of the finger vein image must be identified and separated from background of image.
The mask filter is used to do this localization, which follows the procedure as [17], [18]. The database image is then

B. Gamma correction
To enhance the brightness of the cropped finger, gamma correction is used [19], [20].

C. Fuzzy histogram equalization
To get higher quality images, fuzzy histogram equalization is used to increase the enhancement of veins illumination.
Dynamic fuzzy histogram equalization is used because it does not affect the brightness and it introduces the histogram image in a manner that in mapping of peaks does not appear again. The values of histogram are spread in a uniform manner between valley similar areas of two serial peaks [21]. In the Fig. 7 The image after fuzzy histogram equalization and image histogram. To remove the impulsive and salt-and-pepper noise, Median filtering is used on the image. It smoothly removes noise and keeps the edge-information. The strong point of Median filter depends upon window size [22], [23], [24]. shown that in Fig. 8.

V. FEATURE EXTRACTION
To extract the information and features from images, histogram of oriented gradient ( HOG) and hierarchical centroid shape descriptor methods are used.

1) Histogram of Oriented Gradient Feature extraction
In tough visual situation, Histogram of Oriented Gradient ( HOG) feature extraction is extensively used due to its ability to do well in those situations [24]. It extracts the information about target edge distribution and is a useful tool for shape identification [25], [26], [27], [28]. Shift and rotations of object do not affect this low-level feature descriptor [29]. Its algorithm works as follows: • Image is separated into small sections which are called " cells".  [30] are calculated for each cell ( for ' signed gradients' , histogram channels are spread evenly over 0-360 and for ' unsigned' over 0-180) .
• For scaling histogram counts, lighting effect is fixed. To do so, add energy measure for each local histogram over the larger connected sections.
• Feature vector, in the end, is accumulated by seeing all orientation counts in the calculated dominant direction of gradients.
The above stated steps therefore normalize the block-cells whose joint histograms establish the HOG feature vector of the complete image. Scaling and rotation do not affect the algorithm because only the key points are used for scalespace extreme a detection. The gradient directions for a sample image patch as shown in Fig. 9 Thus, HOG descriptor primarily comprises of features about histograms of slopes of object edges of an image. HOG feature descriptor can be defined by the equation [31]: Where: K = the total number of overlapping blocks in an image F i is the normalized block vector of i-th block: Where: C = total number of cells in a block. B = total number of bins in a cell.
Bins create a cell and cells make up a block. For a sample image of size 32 × 32, this procedure can be demonstrated as shown in Fig. 10 2) Hierarchical Centroid Shape Descriptor ( HCSD) HCSD [32], [33] is fixed in kd-tree procedure decomposition [34]. It is a feature descriptor mainly for shapes.Centroid coordinates extracted from a binary image create this feature descriptor. It works by iteratively disintegrating an image into sub blocks of pixels by means of the Kd-tree algorithm. It comprises of data regarding centroid coordinates of local sections [35]. If d denotes the depth of the features extraction procedure, then the length of this descriptor is specified by 2 × (2d − 2). Assume I is the M × N binary image. Its foreground is denoted by I f g and background by I bg . HCSD is then created with the following stages: • For input I, find its transpose: I T , • Find centroid C(x c , y c ) at the root level for every individual input as equations below: Raw moment ( also recognized as moment of order (p + q)) for a 2D continuous function f (x, y) is specified by Eq. 5 The raw moments m pq of a digital image with pixel intensities I(i, j) are specified by Eq. 6 • According to their centers of gravity (x = x c , y = y c ) are Iteratively decomposed the image into two sub images until desired depth of breakdown is achieved. At each next stage, it is needed to change the axis of coordinates secured.
• As next step, attained vector is needed to scale over a range of -0.5 to 0.5. Here, 0 is for the root-level centroid.
Negative range represents tree decomposition of left sides and positive range represents right side.
• Lastly, extracted features of the images I and I T are combined. K-tree decomposition structure as shown in Fig. 11 Both of SDUMLA-HMT Database Image and collected database by proposed device are evaluated with K Nearest Neighbor and Deep Neural Networks using 6 fold stratified cross validation. The more details for mechanical operation of KNN and DNN in [36], [37], [38], [39]. K-nearest neighbor classifier, also called KNN [40] is a primitive but immensely favored classification method. The label of K closest samples is assigned to a test feature vector [41], [42]. It is a non-parametric passive algorithm [43].
• It is termed " non-parametric" as it does not cause any inherent assumptions about the placement of data into classes.
• It is termed " passive" because it does not utilize training data for generalization. It forecasts test data based on all the training data.
2) Deep neural networks The process of learning data representations with many levels of abstraction through a set of computational models based on numerous processing layers is called " Deep Learning" ( see Fig. 12) [44], [45]. To evaluate the proposed system the mix finger type and hand type was tested for each feature extraction method and also for each classifier. In the experiment results, it was found that the middle finger is the best of all for the considered database for both SDUMLA-HMT database and collected data by proposed device. Results of the proposed algorithm for recognition and verification for middle finger of right hand, left hand and both hands are collected in the Tables ( II, III, IV) . For KNN, results are given with HOG, HC and with the feature vector of HOG and HC. For DNN, dimensions of combined HOG and HC were first reduced ( to be equal to 15) using Generalized Discriminant Analysis ( GDA) [46] , before using them for classification. The cause for this is that data samples per class were limited and DNN must work best on wide-ranging of data whose feature length is not more than the total available samples.

VIII. CONCLUSIONS AND FUTURE WORK
In this work, it introduces new methods of finger vein enhancement for recognition and verification purposes using two type of database, the first one is the SDUMLA database and the second is the data that collected by proposed hardware device. An extensive evaluation of results is done with KNN. KNN proved to be much more accurate in its predictions on test data. The results obtained from this research to compare the accuracy and reduced processing time for finger-vein