IJBSCHS (2009-14-01-2

)

[Original article]

Biomedical Soft Computing and Human Sciences, Vol.14, No.1, pp.11-19 (2009) Copyright©1995 Biomedical Fuzzy Systems Association

(Accepted on 2008.10.3)

Palmprint Based Biometric System: A Comparative Study on Discrete Cosine Transform Energy, Wavelet Transform Energy and SobelCode Methods
Edward WONG Kie Yih* G. Sainarayanan** Ali Chekima*
* School of Engineering and Information Technology, Universiti Malaysia Sabah, Sabah, Malaysia **Department of Electrical and Electronics, New Horizon College of Engineering, Bangalore, India E-mail: drsai@ieee.org The paper was received on Nov. 20, 2007. Abstract: Palmprint based biometric identification has gradually attracted the attention of researchers due to its richness in amount of features. Palmprint contains geometry features, line features, point features, texture features and statistical features. In this paper, simple and effective methodology for palmprint based identification system is proposed. The right hand image is captured using a digital camera without pegging or illumination arrangements. The captured image is aligned using identified key points in the hand and the palmprint region is selected for enhancement and resizing. Different feature extraction methods, namely Discrete Cosine Transform energy features, Wavelet Transform energy features and SobelCode are applied to the resized image to obtain feature vectors. The extracted feature vectors are matched using similarity measurement and feedforward backpropagation neural network. The proposed schemes are tested with hand images from 101 individuals. Keywords Palmprint Identification, Discrete Cosine Transform, Wavelet Transform, SobelCode, Similarity Measurement, Neural Network was introduced a decade ago [2]. It has gradually attracted the attention of various researchers due to its richness in amount of features. Palmprint contains geometry features, line features, point features, texture features and statistical features that can be used to differentiate individuals. Palmprint geometry features include palm area, palm length and palm width. Since these geometry features are not distinctive enough to differentiate individuals, they are usually used in hierarchical palmprint biometric or combined with finger geometry features to form hand geometry biometric system [3]. The line features are unique for every individual. The extractions of the palm lines using stack filter [4], Sobel and morphological operation [5], derivative of Gaussian [6] have been applied in earlier works. The difficulty faced in line features extraction is some people have unclear palm lines or strong wrinkles. Strong wrinkles are the wrinkles that have approximately the same width as the principle lines. Thus, extraction of perfect palm lines is the challenging part. The palmprint point features require high-resolution image. The high-resolution image can be obtained through scanner. In direct scanning method, scanner used several seconds to scan the whole hand. This is very unlikely for the user, which is in hurry or required to use the system several times in a day. In indirect scanning method, the inked palmprint is applied on a

1. Introduction
Biometric is the science of measuring human’s characteristics for the purpose of authenticating or identifying the identity of an individual. Two types of characteristics are measured in biometric technology namely, physiological characteristics and behavioral characteristics [1]. Physiological characteristics measure human body parts while behavioral characteristics measure the actions produced by human such as sound, signature, or posture. Behavioral characteristics are more vulnerable to change than the physiological characteristics. Several types of physiological characteristics used in biometric are appearance of face, hand geometry, fingerprint, iris and palmprint. Palmprint biometric has advantages over other types of biometric system. The palmprint acquisition device costs lesser than the irisscanning device. Palmprint is harder to imitate than fingerprint. Palmprint biometric system can achieve higher accuracy than hand geometry biometric system. It is also more acceptable than face recognition system that may cause privacy issues. The first palmprint biometric identification system
Department of Electrical and Electronics New Horizon College of Engineering Bangalore, India drsai@ieee.org

11

Independent Component Analysis [10] and etc. Image pre-processing has two important parts. which can be used to represent the palmprint effectively is investigated. The complete overview of the proposed palmprint biometric system is shown in Fig. In this work. The palmprint feature is stored as feature vectors and matched using similarity measurement and neural network to identify the identity of the individuals. In CCD camera setup. Palmprint statistical features are sometimes grouped into palmprint texture features because it transforms the palmprint image into statistical value. Vol. Key point may be the gaps between two fingers [7] or the tips of the fingers. DCT energy feature and WT energy feature methods were brief in Section 8 and 9 respectively. A Typical Palmprint Biometric System Hand image can be acquired using scanner. The extracted feature is usually represented and stored as a feature vector. Section 10 explains the SobelCode methods and some improvement to further increase the overall accuracy. Pegged hand image reduced the rotational variation. Three different feature extraction methods. The feature extraction depends on the types of feature targeted. the invisible statistical features are extracted. Palmprint texture features are the representation of palmprint image in different transform space so that the targeted feature is emphasized in its transform space.Biomedical Soft Computing and Human Sciences. namely. In feature matching. Wavelet Transform (WT) energy feature [15] and SobelCode [16]. a special lighting is required to capture a brighter hand image. The results and discussion among different type of methods are done in Section 12. Section 11 describes the similarity measurement and the neural network used. Since the CCD camera focusing area is limited. The hand image is usually taken in front of a dark intensity background to ease the image segmentation process. For example. Fig. peg-less acquisition is proposed in this work. Complete Overview of Proposed Palmprint Biometric System Fig. 2. feature extraction and feature matching. Some of palmprint statistical features methods are Principal Component Analysis [10]. circle or custom in shape. Circled ROI mask and custom ROI mask [12] are used if the feature extractor required specific types of ROI region. different features types are combined to form a new feature types. No1 (2009) piece of white paper before scanned to obtain a highresolution palmprint image. Palmprint image selection and extraction are explained in Section 5 and 6 respectively. The ROI mask is square. but not eliminate them. Digital camera can acquire highresolution image in longer distance compared to CCD camera. The right hand image of 101 individuals is acquired using a digital camera. the hand is pegged during the acquisition process. Square ROI mask is used in most of the earlier work. They are image acquisition. 12 . The key point in the hand image is determined before the palmprint image is selected and extracted. 1 shows a typical palmprint biometric system. Palmprint statistical features use low-resolution image. image alignment using ellipse-fitting method [11] is also proposed in earlier work. CCD camera as suggested in [7] can capture lowresolution image in real time. The rest of the paper is organized as follows: Section 3 explains the condition during hand image acquisition. the combination of line feature and geometry feature to form line geometry feature are suggested in [13]. 1. a palmprint based biometric system that can be work in peg-less environment is proposed. Fig. This method can provide more detailed palmprint but it is timely and cannot be applied to the real time application. 2. feature vectors can be compared using similarity measurement or neural network. In some cases. Some of the extracted texture features are using Fourier transform [7]. 2. Since image alignment is still required.14. Palmprint Biometric Methodology A typical palmprint biometric system has four stages. Section 4 determines the key points in the hand image. Region-of-interest (ROI) selection is the cropping of palmprint image from the hand image. The palmprint image is normalized and its features are extracted. Discrete Cosine Transform [8] and Wavelet Transform [9]. image preprocessing. Image alignment is usually done by referring to the key points. image alignment and region-of-interest selection. This is because key points do not vary when the hand rotates. Section 7 gives the overview of feature extraction and representation. Besides that. Discrete Cosine Transform (DCT) energy feature [14]. CCD camera or digital camera. Hand image is taken either in pegged or peg-less environment. Section 13 summarizes the conclusion obtained from this work. to have more user friendliness. Instead of extracting palmprint visible features. Scanner can acquires highresolution hand images but required more time to scan. It can also be connected to computer as the real time acquisition device.

The key point is circled in Fig. 6. Key Point 1 (KP1). α is considered long if it is more than 100 pixels from the edge of the image. Selection of different background color is to test the robustness of the algorithm to differentiate the hand image from the background when the background changes. parallel or down). it is shown that the maxima points in the graphs represent the tips of the finger. By referring to the boundary pixels index. No pegs are used to align the hand and no lighting arrangements are made in this setup compared to the earlier works in [7]. 6 shows the distances between boundary pixels in clockwise direction from the Pw versus the index of boundary pixels. Two different types of background namely. 4 shows the extracted binary image using Otsu’s Method. 5: Boundary of the Hand Image hand image. Fig. Key Point Distance Differences β Up Parallel Down α Long 10 7 4 Short 5 4 4 Fig. Pw is selected to find the key points for image alignment because it gives shorter distance at the gaps between the fingers while longer distance at the tip of the fingers. The connectivity between the boundary pixels is set at 8connected neighborhood to get a smoother boundary 13 . The red channel of the hand images is separated from its background using Otsu’s method [17]. the location of the wrist is determined. the first and third valley of the graph is located. Fig. Fig. the KP1 and KP2 in hand image are determined. The distance between the boundary pixels with Pw is calculated. Key Point 2 (KP2). 5 shows the boundary of the hand image. Using a tracking algorithm. Since the skin of the hand contains reddish color. 6. black and dark blue color backgrounds are used in this work. The fingers are required to spread apart and the hand is lean against the background. 1. and the minima points are the gaps between two fingers. et. × Pw Fig. Tab. 3. 7 shows the definition of α and β in graphical form. The boundary pixels of the hand image are traced using boundary-tracking algorithm [18] from the Pw in the clockwise direction. The first valley in the graph is the gaps between little finger and ring finger. al: Palmprint Based Biometric System: A Comparative Study 3. Fig. Binary Hand Image By referring to the border of the binary image. 3 shows one of the right hand images. Key Point Determination The hand image is represented using Red-GreenBlue (RGB) format. The center of the wrist is defined as Pw. Table 1 shows the orientations of the hand and key point distance differences. β is the orientation of hand respective to the orientation of wrist (up. Fig. Hand Image Acquisition The right-hand images of 101 different users are taken in front of a uniform dark intensity background using Canon PowerShot A430 digital camera. WONG. Let α be the length of the wrist. One of the classes is the background while the other one is the hand image. Captured Hand Image 4. 4. the red channel is selected for image segmentation. The third valley in the graph is the gaps between middle finger and index finger. 6. Fig. Distance of Boundary Pixels from Pw versus Boundary Pixels Index From Fig. Fig.E. Otsu’s method calculates the suitable global thresholding value for every hand image according to the variances between two classes.

2 times of A. 11 and 12 show the extracted palmprint image for same individual and different individuals respectively. Palmprint Extraction Since the size of the original image is large.3). as in Fig. A is calculated using following equation: A = ( KP1x − KP 2 x ) 2 + ( KP1y − KP 2 y ) 2 Fig. the location of the palmprint area is estimated. A consists of two finger roots and some gaps between fingers. To avoid the ROI mask inserted the background between thumb and index finger. Calculation of θ degrees The rotated hand image has the value of zeros in the non-image location. θ is calculated using the key point as shown in the Fig. R (l . it is noticeable that the maximum difference between the extracted key points and the exact key points are only 10 pixels. (1) where (x.1). The width of the fingers is approximately 0.2 to ensure that the entire ROI mask will be situated within the palmprint area. The length for each side of the square is calculated using ROI _ Length = ( 1 + B + B ) × A (2) Fig. Palmprint Selection A variable size of mask is created to crop the Region-of-Interest (ROI).2). 12.Biomedical Soft Computing and Human Sciences. where R is the rotated hand image with size of l x l. it is shown that the ROI selection for hand image acquired without pegging is unique.5 times of A. Vol. From Fig. α and β From Table 1. 14 . R (3. only half of the finger width is considered. LineKP is 0. 11.  . 7. out of the hand image (768 x 1024 pixels). Thus. the max extension at the end of the line connecting KP1 and KP2. The index of the first non-zero pixels (Idx) is located.14. 9 shows the proposed image alignment and ROI selection method. 10. No1 (2009) Fig. 9. In this work. Fig. Let S is the diagonal pixels of the rotated hand image or pixels where its x and y coordinate are equal. B is defines as 0. By referring to the distance between the two key points. The ROI mask is lowered P pixels parallel with the LineKP where P is 0. a smaller hand image is cropped out from the original hand image before image alignment using key points and palmprint image extraction. y) is the coordinate for the key points KP1 or KP2 in the image. 8. l )] Since the palm lines are located below the finger roots. 8 shows the selection of palmprint area using ROI mask according to key points. Fig. P = B× A (3) (4) Fig. The rotated hand image is cropped Idx pixels from its border to obtain the palmprint image. The palmprint image for different individuals is also distinguishable.25 times of A. R (2. Proposed Image Alignment and ROI Selection Method The cropped hand image is rotated by θ degrees. 10. 6. S = [R (1. Distance between two key points. Fig. The hand images are rotate to align the rotational variant hand images into a predefined direction. Selection of Palmprint Area 5.

the palmprint image is analyzed in single resolution. Discrete Cosine Transform Discrete Cosine Transform (DCT) is a Fourierrelated transforms that is equivalent to roughly twice the length of Discrete Fourier Transform but operating on real data with even symmetry. Wavelet Transform In Discrete Cosine Transform. Palmprint Image for Same Individual image is resized to 256 x 256 pixels to normalize the size of all the palmprint images. 12. 12 (b) is plotted in dark gray color. Palmprint Image for Different Individuals 7. the histogram of the image is adjusted to the full 256 bins. Fig. DCT is applied. The feature representation must be short but contains vital information that can differentiate different individuals. 13. 14 shows the feature representation method discussed earlier. DCT coefficients in each sub-region is squared and summed to obtain the DCT energy. Let D represents horizontal (H). The palmprint image is in RGB format. the wavelet coefficients are squared and summed to obtain its energy value. multi resolution analysis using Wavelet Transform is proposed. the DCT energy features for the Fig. SobelCode that extract line features in specific direction is also tested in this study. 12 (a) is plotted in light gray color while the DCT energy features for the Fig. WT energy feature analyze the palmprint in multi-resolution level. When the decomposition level increases. The palmprint image is in RGB color format. Wavelet Transform (WT) energy feature and SobelCode are investigated in this work. Three types of feature extraction and representation. For each of the detail coefficient. Since the line features on the palmprint image are unique. The enhanced 15 . the wavelet coefficient image is separated into 4 x 4 blocks. In [21]. The extracted features are represented in feature vector for easy comparison in later stage. 13. 12. Fig. Fig. For every coefficient blocks. Multi resolution wavelet transform can extract different types of line in different resolution level. for each of the color channel of the palmprint image. et. The DCT coefficient obtained in every block is separated into four subregions. Level one decomposition allows the extraction of ridges information. SobelCode represented the palmprint image pixels by pixels using a resized version of the palmprint image. Thus. The enhanced image is decomposed into six level of Haar Wavelet decomposition. the palmprint image is enhanced to improve its contrast. 11. Then. wrinkles and ridges can only be acquire in different resolution. For every cD. 16 shows the wavelet energy features for interclass palmprint image respectively. DCT energy feature analyze the texture information of the palmprint image in 4 x 4 blocks where each block is 16 x 16 pixels. The DCT energy feature is arranged to form feature vector. Fig. 4 x 4 blocks are applied and its energy feature is calculated. Fig. Since the palm lines.E. 13 shows the DCT energy features for the Fig. 9. energy features in sixdecomposition level are used to represent every single block. For every 16 x 16 pixels block of normalized palmprint images. Feature Extraction and Representation Palmprint image contains various types of features. In [20]. vertical (V) or diagonal (D) details coefficients. al: Palmprint Based Biometric System: A Comparative Study Fig. these features are investigated in this study. 8. where every different resolution level aims at specific types of palm lines. It is observable that the DCT energy features for different individuals are different in both magnitude and location. The wavelet energies in different decomposition levels are combined and normalized before arranging it to form the feature vector. the palmprint image is converted to grayscale intensity image before its histogram is adjusted to full 256 bins. WONG. Since texture features and line features required lowresolution image and can distinguish people effectively. DCT Energy Feature for Different Individuals In Fig. larger palm lines such as wrinkles and principal lines are extracted. such as principal lines. After the palmprint image is extracted. the palmprint image is converted to grayscale intensity image. Discrete Cosine Transform (DCT) energy feature.

14. Feature Matching Two types of feature matching. 7 (a) and (b) is the resized enhanced palmprint images for the same individual while (c) is the resized enhanced images for another user. Fig. namely similarity measurement and neural network are used in this work. Fig. 18 shows the SobelCode for the resized enhanced image in Fig. 17. This is just inverse of Euclidean Distance where zeros represent both of the feature vectors are the same (no different) while the greater the Euclidean distance value. 16 shows the four different Sobel operators. 45. Then. Euclidean distance calculates the summation of squared differences between two feature vectors while Hamming Distance calculates the total differences between two feature vectors in terms of pixels. Resized Enhanced Palmprint Image for (a) and (b) Same Individual (c) Different Individuals Fig. the likeliness between two feature vectors is calculated. 18. No1 (2009) (a) (b) (c) Fig. Each set of the hand image contains one hand image from 101 16 . WT Energy Feature Representation Method Fig. 90 and 135 degrees The Sobel results are threshold according to the value sign. 90 or 135 degrees Sobel operator is used to extract the line details of the palmprint image in the selected direction. Fig. In similarity measurement. There are ten sets of hand image in this work. In this paper. 17. the differences between inter-class feature vectors (feature vectors from different individuals) are clear. SobelCode for (a) and (b) Same Individual (c) Different Individuals 11. 15. 15.l ) 2 l =1 k (5) where Fvi. 16. 45. 45. Fig. This is because the SobelCode is consists of only ones and zeroes (logical data). four different Sobel operators. 0.Biomedical Soft Computing and Human Sciences. Vol. DCT energy features and wavelet energy features are matched using Euclidean Distance while the SobelCode is matched using Hamming Distance. Each of the RGB channel in palmprint images are adjusted to the full 256 bins before converted to grayscale intensity image.l − FV j . (a) (b) (c) Fig. the enhanced images are resized to 60 x 60 pixels.l = Feature vector i with length k. palmprint image can also be compared using line features as in SobelCode. 90 and 135 degrees are applied to the resized palmprint images. Hamming Distance tends towards one if both of the feature vectors are approximately the same while tends towards zero if both of the feature vectors are different. In [22]. the more dissimilarity between both of the feature vectors.  ( FVi. 3 x 3 Sobel Operator in 0. Sobel operator of 0. SobelCode Besides analyzing palmprint image in different texture features. 14. The general equation for Euclidean distance is E _ Dist = 10. Some of the similarity measurements are Euclidean distance and Hamming distance. Wavelet Energy Feature for Different Individual From Fig.

the image is pre-processed and its feature is extracted and represented. Fig. The minimum hamming distance is then normalized to range 0 and 1 using the following formula: HD = 1 − HD 56 × 56 × 4 Table 2 shows the FAR.68 93. False Rejection Rate (FRR) is the percentage for number of wrongly rejected individuals over the total number of correct matching. Tab. Sliding Window The Hamming distance is defined as There are ten sets of hand image in this study. 135 degrees). Fig. when the minimum gradient of 10-7 has been reached or when the desired performance goal has been reached. From the Euclidean distance calculation for five sets of the hand images. The threshold point acts as a breakpoint to differentiate the genuine users with the imposters. al: Palmprint Based Biometric System: A Comparative Study different individuals. The input layer has the size of the feature vector. In similarity measurement (Euclidean Distance and Hamming Distance). the threshold point will be set as a global threshold. If the feature vector belongs to the genuine individuals. 2.38 Energy WT ED 88. Scaled conjugate gradient-based feedforward backpropagation neural network is also used in this work as classifier. 19. j )) 4 56 56 where h = {0. Three sets of the hand image are used for neural network training. The feature matching for SobelCode using Hamming distance is as follows. 17 . Another three different sets of hand images are used to find the threshold point. the accuracy is calculated as follows: Accuracy = (1 − ( FAR + FRR )) × 100 2 HDy = h = 0 i =1 j =1    ( FV1 (i. a true positive is obtained.30 97. i and j is the row and column of the feature vector. the central 56 x 56 pixels of SobelCodex are cropped out. The hidden layer has equal or double the size of the input data. If it is belongs to others. 12.90 91. Tangent sigmoid activation function is used in the network training. 19 shows the sliding window method used to find the Hamming distance. The cropped SobelCodex is compared with the SobelCodex stored in the database using sliding window method. a threshold point is calculated. 90. which are one hidden layer and one output layer.30 97. a false positive or false acceptance is detected.95 94. The neural network stopping condition are when the maximum epochs reached 20 000.58 94. FRR and accuracy of different types of feature using Euclidean Distance or Hamming Distance. j ) ⊕ FV 2 (i. The minimum of the 25 hamming distance is: HD = min( HD y ) (9) (7) where y = hamming distance from 1 to 25. After the threshold point is obtained. When a hand image is inserted into the system.32 97. A total of 25 hamming distance can be obtained using the sliding window method between two different SobelCode. Results (6) False Acceptance Rate (FAR) is the percentage of wrongly accepted individuals over the total number of wrong matching. The proposed neural network consists of two layers.40 94.71 93. et.84 Energy SobelHD 97.E. The accuracy of the neural network is calculated as the total number of correctly identified template over total number of user template. The output layer has 101 neurons to represent the identity of every user.31 Code Note: SM – Similarity Measurement ED – Euclidean Distance HD – Hamming Distance E0 – Original Grayscale Palmprint Image E1 – Adjusted Palmprint Image E2 – Histogram Equalized Palmprint Image E3 – Individually Adjusted Palmprint Image E4 – Histogram Equalized Individually Adjusted Palmprint Image (8) where 56 x 56 is the size of the cropped SobelCodex and 4 is SobelCode in different directions.30 97. 45.10 91. The feature vector is simulated with the trained network.47 94. WONG. The remaining four sets of data are used for to test the trained network. y = number of hamming distance and ⊕ = exclusive OR operation. Similarity Measurement Results Enhance Image Type Feature SM Type E0 E1 E2 E3 E4 DCT ED 90. Firstly. This threshold point is used in the Euclidean distance for the remaining of five sets hand image to separate the genuine user with the imposter.

Zhang (1998): “Automated Personal Identification by Palmprint”. Shu and D. Table 3 shows the results of the neural network. 11. SobelCode can achieve an accuracy of 94. Equal Error Rate or ERR is the line where the FAR is equal to the FRR. 20. Then.40 From Table 3. From Fig. vol.89 98. no. Kluwer Academic Publishers. Thus. in terms of similarity measurement. Fig. Fig. References [1] David Zhang (2004): “Palmprint Authentication”.41 96. to the length of the feature vector. For similarity measurement. the SobelCode method is not tested using neural network in this study. An accuracy of 98 percent can be obtained using the WT energy features method using neural network. 20. Neural network classifies the palmprint image better than the similarity measurement. The palmprint images are enhanced and resized to a predefine size. Moreover. it is noticeable that neural network can increase slightly the accuracy of the palmprint biometric system compared to similarity measurement methods. the accuracy of the system using ERR is 93. it is shown that the extracted palmprint image is unique in different individuals. The repetition of the data representing the same block has reduced the accuracy of the system in this study. the time taken to identify an individual is less than 10 seconds. The feature vectors are compared using similarity measurement and neural network. In another hand. From the study.87 96. The histogram equalized methods work better in Wavelet Energy and SobelCode methods. Graph FAR versus FRR for SobelCode Method From Fig. The hand image is taken without pegging or lighting illumination. However. Neural Network Results E1 E2 E3 E4 DCT Energy 96. 8. 14400 (60 x 240 pixels) input nodes are required.84 percent compared to DCT energy feature (94. 20 is normalized to 0. SobelCode who also representing every block with a logical value can provide a better accuracy than the DCT energy feature. FAR equal to FRR at point 0. but it is better when it used neural network to classify. 20 shows the graph of FAR versus FRR for the SobelCode Method [16]. the neural network training is computationally expensive. To put the SobelCode as the input of the neural network. the hand image is segmented and its palmprint area is extracted using a square ROI mask. DCT energy feature represents every block with a single energy value. Histogram equalized palmprint image can achieve higher accuracy than the histogram equalized individually adjusted palmprint image. Accuracy = (1 − ( 2 × ERR )) × 100 (10) The SobelCode is too large to form an input feature vector in neural network. All of the feature extraction method can achieve more than 91 percent of accuracy. palmprint biometric identification system using multiple features will also be suggested to further improve the accuracy of the system. 12 and 17. WT energy feature and SobelCode information is extracted from the resized enhanced palmprint image.62 98.62 percent) and WT energy feature (91.32 percent or calculated using the equation below: The FAR and FRR in Fig. thus the ERR value needs to be multiplied with 2 to get the correct ERR. In the future. pp. [2] W.035. Wavelet energy features may not be as similar as DCT energy features. This is because SobelCode has 60 x 240 = 14400 comparisons to determine the identity of the individual than the DCT energy feature with the feature length of 1024. 37. Vol.41 percent can be obtained using the DCT energy features method using neural network. The hidden neurons are set at the value equal 18 . it is shown that the SobelCode can achieve higher accuracy than DCT energy feature and WT energy feature. DCT energy features can obtain the highest accuracy compared to wavelet energy and SobelCode.52 percent). Optical Engineering. 3. Conclusion The right hand image of 101 different individuals is acquired using a digital camera.5.17 97. DCT energy feature. The enhanced palmprint images (E1 to E4) can achieve higher accuracy compared to the original grayscale palmprint images. 2659-2362. 13. No1 (2009) The results of the Table 2 are obtained from the same database with 1000 hand images (10 right hand images for 100 different individuals). Tab. The program coding will be optimized in the near future.14. An accuracy of 96. similarity measurement is faster and easier to implement since it does not required training. In this case. In this work. From Table 2.Biomedical Soft Computing and Human Sciences.02 95. WT energy feature uses several energy values present the same image blocks in different resolution levels. the DCT energy feature has 1024 hidden neurons while WT energy feature has 288 neurons.02 WT Energy 98.

Vol. D. Poon. vol. E. Malaysia. New York. Wong. 36. Proceedings of Fourth International Conference on Audio. 12531257. and David Zhang (2006): “Palmprint Texture Analysis Using Derivative of Gaussian Filters”. in 1979 and 1984 respectively. Fan (2003): “Personal Authentication using Palm-print features”.L. Pattern Recognition Letter. vol.C. He has been working as a research assistant at Universiti Malaysia Sabah. [9] Xiang-Qian Wu. Y. [16] K. Antennas. Sainarayanan and Ali Chekima (2007): “Palmprint Identification Using Wavelet Energy”. Sachin Jambawalikar. K. [5] C. E. G. Y. Signal Processing. Cheng. Wong. M. Edward WONG Kie Yih He received the B.M. Wong. Sainarayanan He completed his B. (1979): “A Threshold Selection Method From Gray-Level Histograms. 9. 25th -28th Nov 2007. [4] P. He has published one book. H. 62-66.. New Techniques and Products. 239-248. H. [12] C. Kumar. pp 668-678. Part B 34(6). Intl. G. pp. “bwtraceboundary” (2005): URL: http://www. Man. Shen (2004): “A New Method in Locating and Segmenting Palmprint into Region-of-interest”. Andrew Teoh. Wong. G. Kuanquan Wang.C. Medical Imaging. [6] Xiangqian Wu. David Zhang (2004): “A face and palmprint recognition approaches based on discriminant DCT feature extraction”. Data Compression. conferences. pp. N.S. Y. W. no. and Cybernetics. Pattern Recognition.C. Wu and M. issue 2. C. 9. Sainarayanan and Ali Chekima (2006): “Palmprint Authentication using Relative Geometric Features”.com/access/helpdesk/help/toolbox/images/bw traceboundary. E. Conf. [10] Tee Connie. No. pp 743-748. pp 8591. 3rd International Conference on Artificial Intelligent and Engineering Technology (ICAIET 2006). Jain (2003): “Personal Verification using Palmprint and Hand Geometry”. Wong.E (Control Systems) from PSG College of Technology and Ph. E. Proceedings of the First International Conference on Machine Learning and Cybernetics. Artificial Intelligence and Data Mining. Michael Goh and David Ngo (2003): “Palmprint Recognition with PCA and ICA”. Sainarayanan and Ali Chekima (2007): “Palmprint Identification Using Discrete Cosine Transform”. You. 25. Iss.Sc. Ali Chekima He received his BEngg in Electronics from Ecole Nationale Polytechnique of Algiers in 1976 and his MSc and PhD both in Electrical Engineering from Rensselaer Polytechnic Institute Troy. 1979. 1041-1050. [13] K. [8] Xiao-Yuan Jing. Vol. Conference of Image and Vision Computing New Zealand 2003 (IVCNZ’03). J. 4. Malaysia. World Engineering Congress 2007. He is a graduate member of IEEE. pp.html. Nov. where he was Chairman of the Scientific Committee of the Department as well as in charge of the Postgraduate Program while teaching at both graduate and undergraduate levels. [7] D.K. Han. four book chapters and 85 technical publications in International Journals and Conferences. He joined the Electronics Department at the Ecole Nationale Polytechnique in 1984. mathworks. G. on Pattern Analysis and Machine Intelligence.E. no.D in Image Processing from University Malaysia Sabah in 2002. Accessed on 6 Nov 2007. Beijing. pp 371-381. H. [14] K. pp. IEEE Transactions on Systems. USA in 2005. He is recipient of Sir Thomas Memorial award from Institution of Engineers (India) and Excellent Scientist award from Ministry of Higher learning. Malaysia-Japan International Symposium on Advanced Technology (MJISAT 2007). 1. al: Palmprint Based Biometric System: A Comparative Study [3] Yaroslav Bulatov.and Video-Based Biometric Person Authentication (AVBPA).C. His research outcome won Silver Medal in 32nd International Exhibition of Inventions. Wong (2003): “Online Palmprint Identification”. and Cybernetics. Photograph of the second author 19 . His areas of research are Image processing. Man. Sainarayanan and Ali Chekima (2007): “Palmprint Identification Using SobelCode”. Y.E (Electronics & Instrumentation) from Annamalai University and M. Piyush Kumar and Saurabh Sethia (2002): “Hand recognition using geometric classifiers”. April 2004 Currently he is working as Professor and Head of Department Electrical and Electronics Engineering in New Horizon College of Engineering. pp.C.M. Wong. WONG.C. Manuscript. Li: (1997): “Pyramid edge Detection Based on Stack Filter”.” IEEE Transactions on Systems. G. Geneva. D. pp 751–754. 533–536. degree in microelectronics from Campbell University. Kong. Pattern Recognition. Shen and A. 1. 4. [18] The Mathworks. Bangalore. He was member of several scientific committees at the national level. 2006. Image Processing Toolbox. 18. Lin and K. Proceedings of the 17th International Conference on Pattern Recognition 2004 (ICPR 2004). On Intelligent & Advanced Systems (ICIAS 2007). His research interests include Source Coding. 12th -15th Nov 2007 [17] Otsu. et. [15] K. Switzerland.L. vol. He has published more than 80 papers in refereed journals. Artificial Intelligence and Control System. He has been working as an Associate Professor at the School of Engineering and Information Technology at Universiti Malaysia Sabah since October 1996. Kuan-Quan Wang and David Zhang (2002): "Wavelet Based Palmprint Recognition". Biometrics. book chapters and research reports. Zhang. [11] A. Vol. North Carolina. pp 2405-2415. pp 227-232. International Conference on Computational Intelligence and Security 2006. IEEE Trans. His present research interests include the biometric and image processing.

Sign up to vote on this title
UsefulNot useful