AN IMPROVED THINNING ALGORITHM FOR FINGERPRINT RECOGNITION

: Thinning is the important steps in pre-processing phase of fingerprint recognition.It explains the visual quality of skeleton with 1-pixel unit width. This paper presents the implementation work of famous thinning algorithms and enhances the Zhang-Suen algorithm in the facet of removal of pixel criteria for preserving the connectivity of pattern, remove noisy points, and for sensitivity of the binary image. The performances of the implemented algorithms are evaluated using Mean Square Error, Peak Signal Noise Ratio, and computational time measurement standard. The implementation is done on fingerprint databases FVC2000 and FingerDos using java platform


INTRODUCTION
Thinning process is useful for decreasing an object of theimage for collecting the most useful information which is useful for further analysis and recognitionprocess [1].It is applied on binarized image and produces another binarized image as an output image [2][3] or it is called as skeleton.Thinning process usedin many applications like OCR, document image analysis, fingerprint identification, biometric authenticationusing images of retina,signature authentication and so on [4].
The thinning process gives some benefits like:  Itdecreases the width size of the ridge pattern up to a single pixel.  Produces a thin line image of fingerprint image [5].  Maintain connectivity between the pixels of an object [6].  The computer can process the data faster and reduce the processing time.  It reduces the usage of the memory because it generates only essential information and the size of the image becomessmaller [7][8].  The thinned image can be used for the classification process of fingerprint recognition [9]  The skeleton image contains information such as end points, bifurcation and connection points.  It reducesundesirable features and undesirable noise.
Thinning is the main process for fingerprint recognition [10] because post-processing steps like minutiae extraction and core point detection is depended on thinning process. The binarized imagesare not used for extracting the minutiae because the ridges are wide in it. For that make the all ridges of the image thick up to 1 pixel. To find out the global and local features of the fingerprint image which are useful in classification, recognition and in matching [11,[41][42][43] process is become very easy from thinned image because it does not change the structure of ridge and valley.

RELATED WORK
Based on the literature review thinning algorithms either directly wok on grayscale image without binarization [12][13][14][15][16] or maximum algorithms required binary image as an input images [8,[17][18][19][20][21][22][23][24][25]. Three types of algorithm are worked for binary images: 1) sequential algorithms, 2) parallel algorithms, and 3) medial axis transform algorithms [8]. Sequential and parallel algorithms are iterative thinning algorithms while medial axis algorithms are non-iterative thinning algorithms. The iterative thinning algorithms inspect each single pixel in binary image and remove the boundary pixel of the imageunit pixel-width thinned image remains while in noniterative thinning algorithm do not examine single pixels one by one, instead of that they produce some centerline of the pattern and then take a decision whether to remove that particular boundary pixel or not [26][27]. The non-iterative algorithms are efficient in computation time but are responsible for creating noisy skeleton while in iterative thinning algorithm the parallel thinning algorithms arefrequently discussed because they are fastand efficient. The Zhang-Suen et al [28],Guo-Hall et al [29], Abdulla et al [30], R. W. Hall et al [31], etc. are the famous algorithm of thinning. The [32] implemented and compare this all four algorithms and prove that Zhang-suen's algorithm gives the minimum computational time, while Guo-Hall's algorithm constructs the best qualitative skeleton. In that paper proposed effective, faster and better preserves structure outcome on thinning using modified Zhang-Suen's algorithm.
In [33] Guo Z et. al., and Zhang T.Y et. al., are implemented and provethat Zhang-Suenalgorithm give the better result. In [34] proposed Hilditch algorithm to acquire more suitable skeletons obtained and give better results in Chinese characters and numbers as well as [35] and [36] implemented Hilditch algorithm for fingerprint images. The following sections discuss the famous iterative parallel thinning algorithms; Hilditch's algorithm [37], Zhange-Suen's algorithm [28] and proposed enhanced Zhang-Suen's algorithm.

IMPLEMENTED WORK
First, implement Hilditch's Algorithm [37]. It follow the below steps. Hilditch's algorithm 3x3 windows is used. It involves performing various passes on the pattern. On every pass, it check the below four conditions if it satisfied then it change the pixel value from black to white.
Stop when nothing changes (no more pixels can be removed)  Here N(p1) = number of non-zero neighbors of p1 and  M(p1) = number of 0,1 patterns in the sequence p2,p3,p4,p5,p6,p7,p8,p9,p2 Then after implement Zhang-Suen algorithm [28]. This algorithm used 3X3 matrix as mask for binary image move down an entire image and go through the all pixels of the image. The below Fig. 1 show the 3X3 matrix. P9 (i-1,j-1) This techniqueeliminatesentire contour points of the imageexcluding those which reside to the skeleton [28]. The difficulty in maintaining the image connectivity with 3x3 operator is restricted in fully parallel thinning algorithms [29]. Up to the image is not reach at stable, and for maintaining the image connectivity many parallel algorithms divide the iteration in two sub iteration [28][29][30][31]. It follow the below steps.  Here M(P1) is the number of 0 to 1 transitions from P2 to P9 in a clockwise direction  N(P1) is the number of non-zero neighbors of P1 that is N(P1) = P2 + P3 + P4 + … + P8 + P9.  Do until image is in steady state  In the first sub-iteration, the contour point P1 is deleted from the digital pattern if it satisfies the following conditions: See in Appendix-I, that some new noisy points are created after implementing Zhang-Suen's thinning algorithm. To remove this noisy point followed fix ridge algorithm after thinning process. This algorithm checks the each pixel of thinned image using 3X3 matrix like following Fig. 2.

matrix implementation of fix ridge algorithm]
See in below e.g. If p is foreground and all its neighbors are background, then remove p. If p is background and check its four neighbors, if there are more than 3 neighbors is foreground and then mark p is foreground.
The fix ridge operationapplied two times for deleting the most of noisy point.

PERFORMANCE METRICS
There are three measurement standard are used to compare the performance of implemented algorithm. 1) PSNR and MSE value 2) Execution Time 3) Quality of an Image 1) PSNR and MSE value [38] PSNR:-PSNR is defined as peak signal-to-noise ratio between two images using following equation (6). The quality is measured among original and recreated image using it. The quality of recreated images is given by high PSNR and with low MSE value. Where, L is the number of discrete gray level.

2) Execution Time
The execution time is used to count the complete time for the whole process.The calculation is consider as the time between the first line and last line of code.

3) Quality of an image
The standard quality of an image is used to measure the spurious branches, noise, and thickness of an image.

EXPERIMENTAL RESULTS AND ANALYSIS
The implementation of thinning algorithms are done using java language and for the experiments the databases FVC 2000 DB1, DB2, DB3, DB4 released on the web [39], and FingerDOS database [40] are used. The details of the databases are given in below Table 1 and Table 2 The below Figure 4 and Figure 5 give clear idea using graphical representation of MSE and PSNR value comparison for three algorithms The following Table 4 show the comparison of algorithms based on execution time.It is clear that the execution time of Zhang-Suen's compare to Hilditch's algorithm take lower time but the proposed enhanced algorithm take little bit few more milliseconds to compute. But that can be neglect because the quality is important for the next minutiae extraction phase.
[ The below Table 5 in Appendix-I show that the quality of an images are improved using the proposed enhanced thinning algorithm. The mark on the images show that removal of spurious branches, noise, and thickness of an image.

CONCLUSION AND FUTURE WORK
This paper presents the role of thinning in fingerprint recognition. It shows theimplementation of famous algorithm Hilditch and Zhang-Suen's algorithm and also enhances the Zhang-Suen's algorithm. The performance is evaluated on FVC2000 and FingerDOS databases using MSE and PSNR measurement parameter. The experimental result prove that the enhanced Zhang-Suen's algorithm is better than the existing Hilditch and Zhang-Suen's algorithm in above all the mentioned aspect. In future, we try to work on post-processing phase like minutiae extraction, remove false point, core point detection and minutiae matching and try to enhance the accuracy level of fingerprint recognition.