外文翻譯--一種對(duì)指紋預(yù)處理和特征提取的高效算法（英文）

1、 Procedia Computer Science Procedia Computer Science 00 (2009) 000?000 www.elsevier.com/locate/procediaICEBT 2010 An efficient Algorithm for fingerprint preprocessing and feature extraction P.Gnanasivam*, S. Muttan

2、Centre for Medical Electroinics, CEG Campus, Anna University Chennai, Chennai-600041, India Abstract In practice, the placement of finger on the scanner for authentication is not done with the utmost care as when placed

3、 during the enrollment and this result in rejections of genuine users. Moreover, user behaviour and envionmental conditions decrease the geniune acceptance rate (GAR) for authentication of fingerprints. To overcome the

4、se limitations, an efficient preprocessing algorithm is proposed to achieve good vertical orientation and high ridge curvature area around the core point for fingerprint authentication and analysis. The algorithm is imp

5、lemented in two stages. The process of obtaining the vertical oriented fingerprint image is carried out in the first step. This is followed by core point detection of a fingerprint. Core point detection is efficiently

6、identified for any type of fingerprints. The developed algorithm is tested using a line based feature extraction algorithm with a large internal database and samples of fingerprint vericication competition (FVC). Only f

7、or the poor quality images, broken ridges are identified which results in a difference in minutiae points. With the proposed algorithm 94% of the tested images were oriented vertically and its genuiness is verified by

8、 comparing the minutiae details of the oriented and unoriented image of the same subject. Keywords: Vertical orientation, high ridge curvature area, Core detection. Feature extraction1. Introduction A fingerprint is

9、the pattern of ridges and furrows on the surface of a fingertip. At a fine level or local level, the characteristic of ridges and valley are known as minutiae. In reality, the fingerprints are not exactly vertically orie

10、nted and it may be ±45º away from the assumed vertical orientation [1]. In this paper, the finger print image rotation is carried out well before image enhancement for the process of extraction of high curvat

11、ure area around the core point. Also, a good orientation model can provide a comprehensive description of the subject that enables discovery of the embedded features. Minutiae around the core point play an important ro

12、le as it is useful in many applications particularly in fingerprint classification and fingerprint analysis. User behaviour and environmental considerations such as angle and placement of the finger on the scanner, dirt

13、 or residue on the sensor, injury or residue on the finger are also necessary for fingerprint matching and analysis. Enrolment of quality fingerprints may be achieved when collection of multiple samples from a same subj

14、ect is possible. There are applications where decision should be made with a single available template. The existing method does not reliably handle poor quality fingerprints when the orientation field is very noisy a

15、nd can be misled by poor structural cues due to the presence of finger cracks. Around the core point rich minutiae information?????????than other region, which is necessary for the fingerprint verification/identificati

16、on [2]. Initially, the enrolled fingerprint image is oriented vertically using rotation algorithm. The orientation of the image is estimated as the angle between the x axis and the major axis of the ellipse. A reasonabl

17、y good quality images with an angular displacement of 1to ±90 degrees are oriented vertically. In the proposed algorithm, the candidate core pixel is selected by considering all the pixel locations where the sum o

18、f all the difference values in the window is around 360, where we use a tolerance of ±3. Then all the candidate pixels present close to each other are bridged together and only the centroid of the region is conside

19、red. After vertical orientation and core detection, the region of interest with core point and high curvature pattern is cropped by selecting n pixels around the core point. The selection of n pixels depends on the app

20、lication and necessity. The proposed processes are verified with the line based feature extraction developed by the authors [8]. * Corresponding author. Tel.: +91 44 2445 2312; fax: +91 44 2743 5769. E-mail address: p

21、gnanasivam@yahoo.com. c ? 2010 Published by Elsevier LtdProcedia Computer Science 2 (2010) 133–142www.elsevier.com/locate/procedia1877-0509 c ? 2010 Published by Elsevier Ltddoi:10.1016/j.procs.2010.11.017Open access und

22、er CC BY-NC-ND license.Open access under CC BY-NC-ND license.P. Gnansivam, S. Muttan/ Procedia Computer Science 00 (2010) 000?0003.1. Image Orientation An input image is read from graphics file. Bilinear interpolation is

23、 adopted to resize the image into a standard or assuming constant size. Let I(x, y) denote a two dimensional grey-scale image, Fig 2 explains the steps involved in the fingerprint rotation process Fig.2. Fingerprint ro

24、tation process (i) Normalization and Segmentation: This process identifies the ridge region of the input image. The given input image is segmented into blocks of size n x n and the standard deviation (STD) in each regi

25、on is evaluated. If the standard deviation is above the threshold it is marked as part of the fingerprint. The image is normalized to have zeroed mean, unit standard deviation prior to performing this process so that

26、the threshold specified is relative to a unit standard deviation. MASK image IM1 is obtained using equation (1), (2) and (3) mean(I) - y) I(x, y) I(x,(1) y)/STD(I) I(x, y) I(x,(2) SQRT(RV) Y) I(X, RM y) (x, I1(3)

27、 Where, RM is the required mean and RV is the required variance. (ii) Morphological Filtering: Morphological filtering is based mainly on some mathematical morphology transformations [17, 18]. Here we perform morpholog

28、ical closing on the normalized and segmented gray scale image IM1 (equation 3) to obtain the closed image, IM2. The structuring element (SE) must be a single structuring element object, as opposed to an array of object

29、s. The morphological close operation is a dilation followed by erosion, using the same structuring element for both operations. This fills the image regions and holes. Dilation of a MASK image IM1(x, y) by a structurin

30、g element SE (m, n) is denoted by IM2(x, y) n)} SE(m, n) - y m, - (x max{I y) SE)(x, , D(I y) (x, I M1 M1 M2(4) Erosion of IM2(x, y) by a structuring element SE (m, n) is denoted by IM3(x, y) } n) SE(m, n) y m, (x min{

31、I y) SE)(x, , E(I y) (x, I M2 M2 M3(5) Orientation angle identification Fingerprint orientation estimation Normalization and segmentation Morphological Filtering Vertical Orientation P. Gnanasivam, S. Muttan / Proced