Limits...
Nonintrusive Finger-Vein Recognition System Using NIR Image Sensor and Accuracy Analyses According to Various Factors.

Pham TD, Park YH, Nguyen DT, Kwon SY, Park KR - Sensors (Basel) (2015)

Bottom Line: Therefore, fingerprint and iris recognitions are preferred alternatives.Therefore, we propose a nonintrusive finger-vein recognition system using a near infrared (NIR) image sensor and analyze its accuracies considering various factors.The experimental results obtained with three databases showed that our system can be operated in real applications with high accuracy; and the dissimilarity of the finger-veins of different people is larger than that of the finger types and hands.

View Article: PubMed Central - PubMed

Affiliation: Division of Electronics and Electrical Engineering, Dongguk University, 26 Pil-dong 3-ga, Jung-gu, Seoul 100-715, Korea. phamdanhtuyen@gmail.com.

ABSTRACT
Biometrics is a technology that enables an individual person to be identified based on human physiological and behavioral characteristics. Among biometrics technologies, face recognition has been widely used because of its advantages in terms of convenience and non-contact operation. However, its performance is affected by factors such as variation in the illumination, facial expression, and head pose. Therefore, fingerprint and iris recognitions are preferred alternatives. However, the performance of the former can be adversely affected by the skin condition, including scarring and dryness. In addition, the latter has the disadvantages of high cost, large system size, and inconvenience to the user, who has to align their eyes with the iris camera. In an attempt to overcome these problems, finger-vein recognition has been vigorously researched, but an analysis of its accuracies according to various factors has not received much attention. Therefore, we propose a nonintrusive finger-vein recognition system using a near infrared (NIR) image sensor and analyze its accuracies considering various factors. The experimental results obtained with three databases showed that our system can be operated in real applications with high accuracy; and the dissimilarity of the finger-veins of different people is larger than that of the finger types and hands.

Show MeSH

Related in: MedlinePlus

Gabor filtering results of the 50 × 20 pixel sub-sampled images from the three databases: Sub-sampled image from the (a) good-quality; (c) mid-quality; and (e) open databases with their respective Gabor filtered images shown in (b,d,f).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541912&req=5

sensors-15-16866-f005: Gabor filtering results of the 50 × 20 pixel sub-sampled images from the three databases: Sub-sampled image from the (a) good-quality; (c) mid-quality; and (e) open databases with their respective Gabor filtered images shown in (b,d,f).

Mentions: Gabor filtering has been popularly used in finger-vein recognition for enhancing the image quality [6,7,8]. In this research, we apply a four-directional Gabor filter to the 50 × 20 pixel sub-sampled image prior to extracting the finger-vein code to enhance the distinctiveness of the vein image. Gabor filtering of the sub-sampled image could also be helpful to reduce the processing time compared to that of the original finger-vein image [7,8]. A two-dimensional Gabor filter can be represented as follows [6,7,8]:(1)G(x,y)=12πσxσyexp{−12(xθ2σx2+yθ2σy2)}exp(j2πf0xθ)with[xθyθ]=[cosθsinθ−sinθcosθ][xy]where , θ is the direction, and f0 is the central frequency of the Gabor kernel. The two coordinates (x, y) are rotated to xθ and yθ, respectively, and on each coordinate, the spatial envelopes of the Gaussian function are represented by σx and σy, respectively. By eliminating the imaginary part of the Gabor filter, the real part, namely the even-symmetric Gabor filter, is used in this research because of the effectiveness with which it processes time. An even-symmetric Gabor filter is represented as Equation (2) as follows [6,7,8]:(2)GkE(x,y)=12πσxσyexp{−12(xθk2σx2+yθk2σy2)}cos(2πfkxθk)withθk=kπ/4; k = 1, 2, 3, 4where k is the index of the directional channel, and θk and fk represent the orientation and spatial frequency of the kth channel, respectively. Based on previous research [6], the optimal parameters of fk, σx, and σy, are determined to be 0.2, 2.38, and 2.38, respectively, for the four channels in the 0°, 45°, 90°, and 135° directions of the Gabor filter applied to the sub-sampled image of 50 × 20 pixels. A convolution operation is applied to an input finger-vein image with the Gabor filter of the four channels to obtain the filtered image in the form of four separated convolution result images. These images are then combined by selecting, at each pixel position, the pixel with the lowest gray-level value among the four pixels of the four result images to be the final result of Gabor filtering, because, generally, the vein line is darker than the skin region [7]. Figure 5 provides example results of four-directional Gabor filtering on the sub-sampled images in Figure 4c,f,i.


Nonintrusive Finger-Vein Recognition System Using NIR Image Sensor and Accuracy Analyses According to Various Factors.

Pham TD, Park YH, Nguyen DT, Kwon SY, Park KR - Sensors (Basel) (2015)

Gabor filtering results of the 50 × 20 pixel sub-sampled images from the three databases: Sub-sampled image from the (a) good-quality; (c) mid-quality; and (e) open databases with their respective Gabor filtered images shown in (b,d,f).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541912&req=5

sensors-15-16866-f005: Gabor filtering results of the 50 × 20 pixel sub-sampled images from the three databases: Sub-sampled image from the (a) good-quality; (c) mid-quality; and (e) open databases with their respective Gabor filtered images shown in (b,d,f).
Mentions: Gabor filtering has been popularly used in finger-vein recognition for enhancing the image quality [6,7,8]. In this research, we apply a four-directional Gabor filter to the 50 × 20 pixel sub-sampled image prior to extracting the finger-vein code to enhance the distinctiveness of the vein image. Gabor filtering of the sub-sampled image could also be helpful to reduce the processing time compared to that of the original finger-vein image [7,8]. A two-dimensional Gabor filter can be represented as follows [6,7,8]:(1)G(x,y)=12πσxσyexp{−12(xθ2σx2+yθ2σy2)}exp(j2πf0xθ)with[xθyθ]=[cosθsinθ−sinθcosθ][xy]where , θ is the direction, and f0 is the central frequency of the Gabor kernel. The two coordinates (x, y) are rotated to xθ and yθ, respectively, and on each coordinate, the spatial envelopes of the Gaussian function are represented by σx and σy, respectively. By eliminating the imaginary part of the Gabor filter, the real part, namely the even-symmetric Gabor filter, is used in this research because of the effectiveness with which it processes time. An even-symmetric Gabor filter is represented as Equation (2) as follows [6,7,8]:(2)GkE(x,y)=12πσxσyexp{−12(xθk2σx2+yθk2σy2)}cos(2πfkxθk)withθk=kπ/4; k = 1, 2, 3, 4where k is the index of the directional channel, and θk and fk represent the orientation and spatial frequency of the kth channel, respectively. Based on previous research [6], the optimal parameters of fk, σx, and σy, are determined to be 0.2, 2.38, and 2.38, respectively, for the four channels in the 0°, 45°, 90°, and 135° directions of the Gabor filter applied to the sub-sampled image of 50 × 20 pixels. A convolution operation is applied to an input finger-vein image with the Gabor filter of the four channels to obtain the filtered image in the form of four separated convolution result images. These images are then combined by selecting, at each pixel position, the pixel with the lowest gray-level value among the four pixels of the four result images to be the final result of Gabor filtering, because, generally, the vein line is darker than the skin region [7]. Figure 5 provides example results of four-directional Gabor filtering on the sub-sampled images in Figure 4c,f,i.

Bottom Line: Therefore, fingerprint and iris recognitions are preferred alternatives.Therefore, we propose a nonintrusive finger-vein recognition system using a near infrared (NIR) image sensor and analyze its accuracies considering various factors.The experimental results obtained with three databases showed that our system can be operated in real applications with high accuracy; and the dissimilarity of the finger-veins of different people is larger than that of the finger types and hands.

View Article: PubMed Central - PubMed

Affiliation: Division of Electronics and Electrical Engineering, Dongguk University, 26 Pil-dong 3-ga, Jung-gu, Seoul 100-715, Korea. phamdanhtuyen@gmail.com.

ABSTRACT
Biometrics is a technology that enables an individual person to be identified based on human physiological and behavioral characteristics. Among biometrics technologies, face recognition has been widely used because of its advantages in terms of convenience and non-contact operation. However, its performance is affected by factors such as variation in the illumination, facial expression, and head pose. Therefore, fingerprint and iris recognitions are preferred alternatives. However, the performance of the former can be adversely affected by the skin condition, including scarring and dryness. In addition, the latter has the disadvantages of high cost, large system size, and inconvenience to the user, who has to align their eyes with the iris camera. In an attempt to overcome these problems, finger-vein recognition has been vigorously researched, but an analysis of its accuracies according to various factors has not received much attention. Therefore, we propose a nonintrusive finger-vein recognition system using a near infrared (NIR) image sensor and analyze its accuracies considering various factors. The experimental results obtained with three databases showed that our system can be operated in real applications with high accuracy; and the dissimilarity of the finger-veins of different people is larger than that of the finger types and hands.

Show MeSH
Related in: MedlinePlus