1 A Novel Technique for Principal Lines Extraction in Palmprint using Morphological Top-Hat filtering Zia-ud-din
[email protected]
Zahoor Jan
[email protected]
Jamil Ahmad
[email protected]
Department of Computer Science, Islamia College Peshawar (Chartered University) University Campus, Peshawar, Khyber Pakhtunkhwa Abstract – Palmprint is one of the main biometric features, which is gaining a lot of attention lately due to its commercial applications. Palmprint is more flexible in identifying an individual based on wrinkles, ridges and Principal lines. These features do not change during the lifetime of a person and no two individuals have similar palmprints. Most of the conventional methods use edge detection operators to detect lines in the palm which produce too many trivial lines. In this paper, we have developed a navel technique to extract principal lines in two phases without using edge detection. First, image is enhanced, blurred and converted to negative. In the second phase Top-Hat filter, closing and binerazation are applied. Noise and undesirable areas are removed with connected component labeling. Key Words: Principal lines extraction, Palmprint recognition, Biometrics, Image filtering, Top-Hat filter. 1. Introduction The palm area of hand contains Principal lines, wrinkles and ridges. The major lines are developed between 3rd and 5th months of pregnancy [1] and the remaining lines subsequently. Palmprints are not heredity [2] and even twins don’t possess identical Palmprint. Palmprint recognition is receiving more attention of the researchers because of simplicity in images acquisition using inexpensive imaging devices, uniqueness, permanence, user approval and cost efficiency of the system [5]. Palmprint recognition uses both high resolution and low resolution images. The farmer images are used for legal application by law enforcing organizations [3] and the later are suitable for public and commercial application. High resolution images are more suitable for the extraction of minutia points, singular points and ridges, while low resolution images are ideal for principal lines, wrinkles and textures extraction [4]. Initial research work focused on legal based application using high resolution images but recently the trend is changing and most of the research is dedicated to commercial application using low resolution images. The success of a Biometrics system depends on five
factors: accuracy, computation speed, security, cost and user acceptance and environment constraints. Palmprint recognition algorithms include line, texture/subspace and statistic based [4]. The line approach depends on information of Principal lines like edge point detection, length and orientation [6]. Texture algorithms use Principal lines, ridges, and wrinkles and invoke transform like Gabor, Discrete Cosine Transform (DCT), Fast Fourier Transform (FFT), Wavelet, etc. independently [7] or combined using statistical operators [4]. While, the statistical based algorithms utilize statistical quantities like mean, standard deviation, median, etc for the feature vector extraction [8]. This study is organized as follow: Section No.2 Related work, Section. No.3 Proposed Methodology, Section No.4 Experimental discussion and Section No.5 Conclusion Section No.6 Experimental Results. 2. Related Work Sakdanupab et al. [15] first normalized image to reduce the effects of different brightness. The gray level profiles in a small window of size 3x3 at 0o, 45o, 90o and 135o degree were used for principal line detection. Noise is removed using morphological closing and smoothing is applied to the resultant image. Zhu et al. [9] extracted coinciding Principal lines by calculating gradient image of four directions. Filters were applied to the resultant image. Using AND operator, filtered image and edges extracted using canny edge function were combined. In [6] Tunkpien et al, first images were blurred using a smoothing filter before any edge detection. Horizontal and vertical lines were detected and combined from a 2x2 gradient operator which was calculated in two directions. Principal lines were extracted by cascading different morphological methods. The feature vector is calculated using shape matching with the help of nearest neighbour. The final recognition was accomplished using Cosine similarity. Han et al. [10] extracted palm lines from region of
2 interest by using Sobel edge operators. Lines were enhanced using different morphological procedures. The resultant image was split into three different shape sizes. Feature vector was obtained by calculating mean of pixels in all blocks. Wu et al. [11] extracted Principal lines using canny edge operator. The scale and direction of edges were determined. Four groups were formed using edges, each indicating edges in a specific direction. Feature vector was calculated from derivation of fuzzy energy in each direction. Leung et al. [12] lines were extracted using sobel edge function. Trivial lines were removed using thresholding and feature vector was extracted depend on the remaining lines.
In preprocessing phase image is normalized by applying various filters, then image is converted to negative i. Image Normalization After obtaining ROI, images with uneven contrast are first enhanced with the enhancement technique in equation (a) and (b) proposed by L. Hong et al. [11]. Fig.2 show low contract image. I(i,j) =
d + ; d ; =
if I(i,j) > otherwise
d ( I (i, j ) )2
(a)
(b)
where ϕd is standard deviation and ρd is the average of the output image and I′(i, j) is the normalized image. ϕ is standard deviation and ρ is the average of the input image.
3. Proposed Methodology Most of the researchers proposed edge detection operator such as Sobel, Canny, Prewitt etc. for Principal lines detection, which produce too many trivial lines. The proposed line extraction method does not use conventional edge detection approach but it uses two phases approach as shown below in the fig.1 Preprocessing Phase Orignal Image
after normalization Figure 2
Region of Interest (ROI)
i. Normalization ii. Smoothing & Median Filter iii. Negative Image
i. ii. iii. iv.
Top-Hat Filter Contrast Adjustment Binarization Component Labeling
ii. Image Smoothing Noise and trivial lines can be removed by blurring the image. A smoothing filter of 5x5 standard mask shown in table-(1) is applied to blur the image. Standard median filter is applied to remove any leftover noise. Results show in Figure-3. 1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
5x5 smoothing mask filter
Principal Lines
Extraction Phase Figure 1 Smoothed image 3.1
Preprocessing Phase:
median filter applied Figure 3
(1)
3 iii. Negative of Image After normalization and removal of noise, the resultant enhanced image is converted to negative Fig.4.
visibility of the image. Fig.6 shows the resultant image after contrast adjustment.
Contrast adjustment Figure 6 Negative image Figure 4 3.2
Extraction Phase: In Extraction phase Top-Hat filter, contrast enhancement and closing is applied respectively on the negative image. Noise is removed with connected component labeling after binarazation.
iii. Image Binarization The grayscale image resulted from Top-Hat filter and contrast adjustment is converted to a binary image using a threshold value. The result is shown in fig.7
i. Top-Hat Filter We propose to correct uneven illumination by using Top-Hat filter. Top-hat filtering calculates the morphological opening and then subtracts the resultant image from the source image shown in eq. (c). T =A Where = (A )
(c)
Where A denotes original image, morphological opening, structure element, erosion and dilation respectively
Binary Image Figure 7
Keeping in view, Top-Hat filter with a disk-shaped structuring element with 2 point radius is used to remove the uneven background illumination from an image. The resultant image is show in fig.5.
iv. Noise removal Connected component labeling is used for trivial lines removal. 8-neighborhood is used for identifying connected pixels. Components which have less pixels than threshold value are usually trivial lines and hence are removed. The final image after remove noise is shown in Fig.8.
Top-Hat Filter Figure 5
Final image Figure 8 4.
ii. Contrast Adjustment Linear Contrast adjustment is applied by scaling the intensity values in the original image to improve the
Experimental discussion IIT Palmprint Database version 1.0 [16] is a widely used database for palmprint research. The current database consist of images of 235 users. 100 random images are selected for this research. The experiments
4 were implemented in MATLAB software with Image Processing Toolbox and on a Dell Optiplex 960 machince with an Intel® CoreTM 2 Duo CPU Processor and 2 GB RAM configured with Microsoft Windows 7 Professional. Conclusion This paper proposes a simple and effective technique to extract Principal lines from Palmprint without using conventional edge detection operator.
[11]
[12]
5.
Results of experiments conducted in this study clearly show that the proposed technique produce less trivial lines and as a result create more vibrant principal lines than Tunkpien et al [6 ] , Zhu et al. [9] and Sakdanupab et al. [15].
6. [1]
[2]
[3]
[4]
[5] [6]
[7]
[8]
[9]
[10]
References X. Wu, K. Wang, D. Zhang, Line feature extraction and matching in palmprint, in: Proceeding of the Second International Conference on Image and Graphics,2002, pp. 583–590. A. Kong, D. Zhang, G. Lu, A study of identical twins palmprint for personal verification, Pattern Recognition 39 (11) (2006) 2149–2156. NEC Automated Palmprint Identification System ― http://www.necmalaysia.com/my/Solutions/ PID/products/ppi.html‖. A Kong, D Zhang , M Kamel, A survey of palmprint recognition, Pattern Recognition 42 (2009) 1408—1418. Biometric Technology Application Manual (2008). National Biometric Security Project. P. Tunkpien, S. Panduwadeethorn, S. Phimoltares (2010), Compact extraction of Principal lines in palmprint using consecutive filtering operations. In the proceedings of the second International conference on knowledge and smart technologies, pp 39-44. H. Imtiaz, S. A. Fattah (2010), A DCT- based feature extraction algorithm for palmprint recognition. ICCCCT’10: IEEE, pp 657 - 660 C. C. Han, H. Cheng, C. Lin, K. Fan (2003), Personal authentication using palmprint features. Journal of Pattern recognition, 36(2003), pp 371-381 L. Zhu, R. Xing (2009), Hierarchical palmprint recognition based on major line feature and dual tree complex wavelet texture feature. IEEE International conference on fuzzy systems and knowledge discovery, pp 15-19 C. C. Han, H. Cheng, C. Lin, K. Fan (2003), Personal authentication using palmprint features. Journal of Pattern recognition, 36(2003), pp 371-381
[13]
[14]
[15]
[16]
X. Wu, K. Wang, D. Zhang (2004), An approach to line feature representation and matching for palmprint recognition. Journal of Software 15(6), pp 870-880 M. K.H. Leung, A.C.M. Fong, S. C. Hui (2007), Palmprint Verification for Controlling Access to Shared Computing Resources. Pervasive Computing, pp 40- 47. The Hong Kong Polytechnic University (PolyU) Finger-Knuckle-Print Database, "http://www4.comp.polyu.edu.hk/~biometrics/F KP.htm" Biometrics Ideal Test, "http://biometrics.idealtest.org/dbDetailForUser. do?id=5". M. Sakdanupab, "An Efficient Approach for Automatic Palmprint Classification," presented at the IEEE International Conference on Signal Image Technology and Internet Based Systems, 2008. Ajay Kumar, "Incorporating Cohort Information for Reliable Palmprint Authentication," Proc. ICVGIP, Bhubneshwar, India, pp. 583-590, Dec. 2008
5 7.
Experimental Results
Original Image
Proposed method
. Zhu et al. [9]
Tunkpien et al [6 ]
Sakdanupab et al. [15]
