ARTICLE International Journal of Advanced Robotic Systems
Image Quality Enhancement Using the Direction and Thickness of Vein Lines for Finger-Vein Recognition Regular Paper
Young Ho Park1 and Kang Ryoung Park1,* 1 Division of Electronics and Electrical Engineering, Dongguk University, Seoul, Republic of Korea * Corresponding author E-mail:
[email protected]
Received 4 Jun 2012; Accepted 17 Aug 2012 DOI: 10.5772/53474 © 2012 Park and Park; licensee InTech. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract On the basis of the increased emphasis placed on the protection of privacy, biometric recognition systems using physical or behavioural characteristics such as fingerprints, facial characteristics, iris and finger‐vein patterns or the voice have been introduced in applications including door access control, personal certification, Internet banking and ATM machines. Among these, finger‐vein recognition is advantageous in that it involves the use of inexpensive and small devices that are difficult to counterfeit. In general, finger‐vein recognition systems capture images by using near infrared (NIR) illumination in conjunction with a camera. However, such systems can face operational difficulties, since the scattering of light from the skin can make capturing a clear image difficult. To solve this problem, we proposed new image quality enhancement method that measures the direction and thickness of vein lines. This effort represents novel research in four respects. First, since vein lines are detected in input images based on eight directional profiles of a grey image instead of binarized images, the detection error owing to the non‐ uniform illumination of the finger area can be reduced. Second, our method adaptively determines a Gabor filter www.intechopen.com
for the optimal direction and width on the basis of the estimated direction and thickness of a detected vein line. Third, by applying this optimized Gabor filter, a clear vein image can be obtained. Finally, the further processing of the morphological operation is applied in the Gabor filtered image and the resulting image is combined with the original one, through which finger‐ vein image of a higher quality is obtained. Experimental results from application of our proposed image enhancement method show that the equal error rate (EER) of finger‐vein recognition decreases to approximately 0.4% in the case of a local binary pattern‐ based recognition and to approximately 0.3% in the case of a wavelet transform‐based recognition. Keywords Image Enhancement; Finger‐vein Recognition; Adaptive Gabor Filter
1. Introduction With the arrival of an information‐oriented society, the importance of information security has increased Int J Park: Adv Robotic Sy, 2012, Vol. 9, 154:2012 Young Ho Park and Kang Ryoung Image Quality Enhancement Using the Direction and Thickness of Vein Lines for Finger-Vein Recognition
1
significantly. Traditional technologies for user authentication depend upon personal identification numbers, passwords or tokens; however, these face certain disadvantages in that passwords can be divulged and tokens can be stolen. Biometric recognition has, therefore, become an alternative means of solving security problems. Biometrics is the technology of recognizing individuals by using human physiological and/or behavioural characteristics, including facial features, fingerprints, iris patterns, voice prints, vein patterns and gait. Biometric solutions need to satisfy the conditions of universality, distinctiveness, permanence and collectability. Other factors to be considered include performance, acceptability and circumvention [1]. Biometric finger‐vein recognition involves the scanning of vein patterns within a finger, a method that is advantageous in that it utilizes a small and inexpensive capturing device, is robust to differences in finger skin condition (e.g., dry, wet or scarred skin) and is difficult to falsify [2‐8]. However, recognition performance can be affected by variations in vein thickness owing to temperature and physical conditions [4] as well as by the quality of the captured finger‐vein image. The quality of this image can be degraded by variations in illumination and camera sensor conditions, optical blurring and other blurring caused by light scattering on the skin. To address these limitations, studies on image enhancement have been previously conducted. These previous studies can be classified as research into software algorithm‐based methods [6‐11, 17] and hardware‐ based methods [12]. In turn, the software algorithm‐based methods can be classified as non‐restoration‐based methods [6, 7, 10, 11, 17] and restoration‐based methods [8, 9]. As a non‐restoration‐based method, Zhang et al. proposed a finger‐vein image enhancement method combining grey‐level grouping and the use of a circular Gabor filter [6]. However, in applying the filter, they did not consider blood vessel direction, which limited the image enhancement they were able to obtain. In addition, although they produced enhanced image samples, they did not document the degree to which their enhancement method improved finger‐vein recognition. Yang et al. proposed an enhancement method using a multi‐channel Gabor filter [7] and an orientation field [10]. In their study [7] the multi‐channel Gabor filter, which used various directions and frequencies, enhanced the finger‐vein image without determining the specific direction and width of the vein upon which it was applied. Although image enhancement was carried out using an orientation field based on vein line characteristics determined by previous research [10], they did not document how well performance was enhanced in their study. Yu et al. proposed a finger‐vein enhancement method based on a 2
Int J Adv Robotic Sy, 2012, Vol. 9, 154:2012
multi‐threshold fuzzy algorithm [11]. Pi et al. proposed a finger‐vein imaging quality enhancement method using an edge‐preserving filter, an elliptical high‐pass filter and histogram equalization [17]. Most of these non‐restoration based methods [6, 7, 10, 11, 17] do not consider the characteristics of vein lines, such as the direction and width of the imaged vein lines, and none of them measured performance enhancements in terms of the accuracy of finger‐vein recognition, even though this is an important final goal in developing an enhancement algorithm. As a restoration‐based method, Yang et al. proposed the restoration method of a finger‐vein image based on the skin layer structure [8] in which scattering blur was removed [9]. In their studies, blurred finger‐vein images were efficiently restored using a model of estimated blurring; however, and as with some of the non‐ restoration‐based studies above, vein line direction and width were not determined during restoration and image enhancement was limited. Additionally, they also did not document any enhancement of accuracy achieved through their method. As a hardware‐based method, Crisan et al. proposed image enhancement using a polarizing filter [12]. Again, the direction and width of the vein lines were not analysed and recognition accuracy enhancement was not documented. Nguyen et al. proposed a quality assessment method of finger‐vein imaging; however, this did not involve the use of an image enhancement method [18]. To improve upon and augment the previous work, therefore, we propose a novel image quality enhancement method that determines both the vein line direction and its thickness. Since vein lines are detected in an input image based on eight directional profiles of a grey image (as opposed to a binarized image), the detection error caused by the non‐uniform illumination of the finger area can be reduced. The optimal Gabor filter direction is adaptively determined on the basis of the estimated direction of the detected vein line, and the optimal Gabor filter width is adaptively determined on the basis of the vein line thickness. By applying a Gabor filter with direction and width optimized in terms of the detected vein line, a clear vein image can be obtained. Additional processing through a morphological operation is applied to the Gabor filtered image and the resulting image is combined with the original one to obtain a higher quality finger‐ vein image. We were able to use our proposed image enhancement method experimentally to reduce the equal error rate (EER) in finger‐vein recognition. Table 1 summarizes the comparison between the previously studied methods and our proposed quality enhancement method.
www.intechopen.com
Category
Non‐ restoration ‐ based Software‐ based method
Restoration ‐based
Hardware‐based Method
Methods ‐ Combination of grey‐level grouping and a circular Gabor filter [6] ‐ Use of a multi‐channel Gabor filter [7] ‐ Use of an orientation field [10] ‐ Use of a multi‐threshold fuzzy algorithm [11] ‐ Use of an edge preserving filter, an elliptic high pass filter and the processing of histogram equalization [17] Application of one Gabor filter where the direction and the width are optimized to the detected vein‐line (The method proposed in this paper) ‐ Restoration of the finger‐vein image by examining the skin layer structure [8] ‐ Restoration of the finger‐vein image by removing the scattering blur [9]
Strengths
Weakness
‐ Did not establish both the direction and the width of ‐ Performance is not the vein line. affected by the vein line ‐ Did not document the detection error performance enhancement owing to the proposed methods
‐ Establishes both the direction and width of the vein line ‐ Documentation of the enhancement of performance ‐ The performance of restoration is good because it considers the skin layer features or light propagation through biological tissue
‐ By using special hardware, such as a ‐ Image enhancement using a polarizing filter, the polarizing filter [12] enhanced images can be obtained quickly.
‐ Performance can be affected by the vein line detection error ‐ Did not establish both the direction and the width of the vein line ‐ Did not document any performance enhancement owing to the proposed methods ‐ The proposed method cannot be incorporated for use in conventional vein acquisition devices ‐ Did not establish both the direction and the width of the vein line. ‐ Did not document the performance enhancement owing to the proposed methods
Table 1. Summary of comparisons between the previous works and the proposed method
2. Proposed finger‐vein image enhancement method 2.1 Overall procedure An overview of our proposed finger‐vein image enhancement method is shown in Figure 1. Figure 2 shows a flow diagram of the method. Our proposed method can be divided into five steps. First, an input image is processed using four of directional Gabor filters. Second, vein lines are detected on the basis of eight directional grey profiles of the image. Third, the image is enhanced using an optimal Gabor filter, with width and direction determined using the detected vein line. Fourth, the additional processing of a grey morphological operation is executed to reduce noise. Finally, the original and processed images are combined using a weighted SUM rule to obtain an enhanced final image. www.intechopen.com
Figure 1. Overview of the proposed finger‐vein image enhancement method. Young Ho Park and Kang Ryoung Park: Image Quality Enhancement Using the Direction and Thickness of Vein Lines for Finger-Vein Recognition
3
wheere n (=1, 2, 3,, 4) is the chaannel index, n (= n / 4 ) iss the orientation an nd fn denotess the centre freequency of an n even n‐symmetric Gabor G filter in n the nth chan nnel. By using g the convolution o of the input im mage I ( x , y) with the even‐ w ‐ sym mmetric Gabo or filter Gne ( x, y) , an output imagee On ( x , y) was obtaained as follow ws [7]:
On ( x , y) Gne ( x , y) * I ( x , y) (4))
Figure 2. Flow w chart of the pro oposed finger‐v vein image enhancement m method.
2.2 Finger‐veiin image enhanccement methodd nhancement usiing four of direcctional Gabor fi filters 2.2.1 Image en
wheere * denotes the convolutiion operator. In this study,, fourr Gabor filterss with directio ons of ‐45°, 0°°, 45° and 90°° werre used. At each e point, th he minimum value of thee obta ained amplitu ude was seleccted as the best b matching g resu ult of Gabor filltering. This w was done beca ause vein liness have a lower grrey value thaan the neighb bouring skin;; a therrefore, the grrey profile off a vein line represents a refleection of the Gabor filter p profile around d the verticall axiss. The final ou utput image w was obtained o on the basis off this filtering. Figu ure 3 shows th he results of p processing thee he four of direcctional Gabor filters. image through th
A finger‐vein n image is com mposed of veiin lines of varrious widths. In th he study descrribed in this p paper, grey prrofile analysis wass used to deteermine the veein line direcction. This is in co ontrast with th he previous rresearch, in which w binarization‐‐based method ds for detectin ng vein lines w were used [19]; however, h thee binarization n performancce is affected by the selection n of an optiimal thresholld, a mination variaation process that is made diffficult by illum over the fing ger area. Prio or to the deteection of the vein lines in our method, m four of directional Gabor filterss are applied in order o to enhaance the finger‐vein imagee, as shown in Fig gure 2. A Gabor filter is a a two‐dimensiional filter that can n be representeed using [7, 133]:
(a)
(d d)
2 y2 1 x exp exp ˆj 2 f0 x (1) 2 2 2 x y 2 x y
(b)
(e))
(c)
(f))
G( x , y )
1
where: x cos y sin
sin x cos y
(2)
ˆ , ntation of the Gabor filter, f0 is and j 1 , θ is the orien the centre freequency of the filter and x and y aree the sigma valuess of the Gausssian envelopess along the x‐‐ and y‐axes, respeectively [7]. In n order to redu uce the processsing time, an even‐symmetric Gabor filter ‐‐ which used only the real part of the Gabor filter output ‐ was used in n this study [7]: Gne ( x , y )
4
2 y2 1 x exp cos 2 fn x (3) 2 2 n 2 x y 2 x y 1
Int J Adv Robotic Sy, 2012, Vol. 9, 154:2012
Figu ure 3. Results frrom the four o of directional Gabor G filters: (a)) inpu ut image, (b) im mage result using g a ‐45° Gabor filter, (c) imagee result using a 0° Gaabor filter, (d) im mage result usiing a 45° Gaborr filterr, (e) image reesult using a 990° Gabor filterr, and (f) finall outp put image from the four of directional Gabor fiilters.
2.2.2 2 Image enhanccement using on ne optimal Gabor filter and a grey morphollogical operation next step, an optimal G Gabor filter with w direction n In the t and width deterrmined by th he detected vein v line wass www.intechopen.com
applied and the further processing of a grey morphological operation was executed, as shown in Figure 2. This step was divided into six sub‐steps: (Step 1) Extracting eight grey profiles which are orthogonal to the vein lines of eight directions (‐67.5°, ‐45°, ‐22.5°, 0°, 22.5°, 45°, 67.5° and 90°), respectively, at each position while scanning in the horizontal direction. (Step 2) An average filter was used to smooth the extracted profiles. (Step 3) Each profile was checked to see whether it corresponded to a specific vein line. (Step 4) Determining the direction of the vein line whose depth of profile is deepest and selecting the corresponding direction of the Gabor filter. (Step 5) Selecting one optimal Gabor filter whose filtered value is minimized (best matching) among the Gabor filters of various widths and obtaining the result image based on the minimum filtered value at each position.
(Step 6) Applying a grey morphological operation to the result image and combining it with an input image as the final enhanced image. First, grey profiles orthogonal to the vein lines in eight directions were extracted by scanning each image point in a horizontal direction. Because the grey level of the vein line is lower than that of the non‐vein area, such as skin, a “vein section” of an image will form valleys in the grey level function; this is illustrated in Figure 4. The grey profile closest to being orthogonal to the direction of the vein will produce the deepest valley, as shown in Figure 4 (d). On the basis of this comparison, the direction of the vein can be estimated. Figure 4 shows an example of the profiles extracted at a point within a finger‐vein image. As shown in Figure 4, the profiles produced some noise, making analysis difficult. To remove the profile noise, an average filter was applied, as shown in Figure 5.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
Figure 4. Examples of grey profiles orthogonal to the vein lines in eight directions. In each figure, the horizontal and vertical axes represent the pixel position and the grey value, respectively. The dotted lines show the direction of the measured grey profiles and the solid lines represent the vein line directions orthogonal to the grey profiles. (a) is the original finger‐vein image from which the direction of the vein line is to be measured (red‐coloured box). The vein directions are (b) ‐67.5°, (c) ‐45°, (d) ‐22.5°, (e) 0°, (f) 22.5°, (g) 45°, (h) 67.5° and (i) 90°. www.intechopen.com
Young Ho Park and Kang Ryoung Park: Image Quality Enhancement Using the Direction and Thickness of Vein Lines for Finger-Vein Recognition
5
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
Figure 5. Example of the smoothing of profiles from Figure 4 using an average filter. In each figure, the horizontal and vertical axes denote the pixel position and the grey value, respectively. The vein directions are (a) ‐67.5°, (b) ‐45°, (c) ‐22.5°, (d) 0°, (e) 22.5°, (f) 45°, (g) 67.5° and (h) 90°.
Each smoothed profile was then checked to see if it belonged to a vein line on the basis of the following conditions:
abs ( Lmax R max ) Vavg (5) abs (horizontal position of Vmin − C)
