 Open Access
 Total Downloads : 844
 Authors : I.Muthulakshmi, Dr.D. Gnanadurai
 Paper ID : IJERTV1IS4158
 Volume & Issue : Volume 01, Issue 04 (June 2012)
 Published (First Online): 30062012
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Time Complexity Analysis of Mband Wavelet Inpainting Technique for Distorted Digital Images
I.Muthulakshmi
Assistant Professor / HOD, CSE Department.
VV College of Engineering , VV Nagar, Tisaiyanvilai627 657 Tuticorin District.
Dr.D. Gnanadurai,
Principal, J.P College of Engineering,Auykudy, Tenkasi627 852. Tirunelveli District.
Abstract: Image inpainting is an image processing technique to restore damaged or lost portions in an image. This paper presents a new M – band wavelet based image inpainting technique for digital images and its time complexity analysis for the suitability of energy aware computing applications. The proposed technique will decompose the given original into sub bands using Mband dual tree wavelet. The damaged or lost portions of an image are then identified in different sub bands and it is filled with neighborhood frequency values. The experiment shows that the proposed technique retains the same level of time complexity with respect to Complex Wavelet Transform technique and it is around 50 seconds at an average for the input image of size 512×512. At the same time, it failed to improve the time complexity with respect to other well known techniques in the literature Discrete Wavelet Transform, Haar and Daubechies. The proposed scheme time complexity is reduced 16% at an average for an input grey scale image of size 512×512. The reasons are use of M band wavelet decomposition and reconstruction in the proposed technique. It is concluded that the proposed scheme can be applied in energy aware computing applications after fine tuning the time complexity of the scheme.
Keywords: In painting, wavelets, DWT, Haar, Daubechies, CWT, 2D Dual tree Complex Wavelet transform

Introduction
Image Inpainting is the process of modifying an image in an undetectable form to recover the damaged or lost portions in an image. The applications of image inpainting falls into broad spectrum that ranges from the restoration of damaged paintings and photographs to the removal/replacement of selected objects [7]. Image inpainting provides an interface to restore damaged region of an image completely to make natural look
after the inpainting [1 – 5]. Applications of image inpainting range from restoration of photographs, films and paintings, to removal of occlusions, such as text, subtitles, stamps and publicity from images. In addition, inpainting can also be used to produce special effects [8]. Traditional image inpainting is widely replaced by image processing tools such as Photoshop. Bertalmio [6, 8] have introduced a technique for digital inpainting of still images that produces very impressive results. Digital techniques are starting to be a widespread way of performing inpainting, ranging from attempts to fully automatic detection and removal of scratches in film [3, 4, 19].
Image inpainting uses interpolation to forecast the pixel values for the damaged or lost portions in an image using its neighborhood pixels [2, 5, 9, 1517]. The other techniques such as neural network, wavelet transform, grid cell analysis (GCA), genetic algorithm (GA), artificial life (AL), fuzzy set theory, texture classification and more has also been employed in image inpainting [10]. Wavelet transformation has been identified as a promising technique for all kinds of image inpainting, since it decomposes the image in multi resolution. The damaged or lost portions of an image are considered as either low frequency or high frequency components; it can easily be identified in various bands of wavelet. Therefore, the wavelet is playing significant role in image inpainting [13].
The 2D discrete wavelet transformation is mainly applied to the model of digital image data in order to find the locality and length of the crack [11, 18]. Other purposes include frequency analysis, selection of region of interest (ROI) and transform data [12, 20].
This paper is organized as follows. Section I presents introduction to image inpainting and significance of wavelet in it. Section II explains the related work in the direction of the proposed technique. Section III presents the proposed technique. Section IV discusses the experimental
results and analysis. The conclusion is given in Section V.

Related work
Through the literature survey on Image Inpainting using wavelets, it is found that there are few works in the same direction of proposed technique that gave significant remarks.
Gunamani Jena [21] has presented an inpainting algorithm, which implements the filling of damaged region with impressive results. Many algorithms usually required several minutes on current personal computers for the inpainting of relatively small areas. Such a time is unacceptable for interactive sessions and motivated us to design a simpler and faster algorithm capable of producing similar results in just a few seconds. The results produced by the algorithm are two to three orders of magnitude faster to the existing.

A. Ismail et al. [22] have proposed an integrated technique for the recognition and purging of cracks on digitized images. Using steepest descent algorithm (SDA), initially the cracks have been identified. Then, the identified cracks have been purged using either a gradient Function (GRF) and processed data or a semiautomatic procedure based on region growing. Lastly, crack filling has been performed using the steepest descent method. The proposed
specimen during a fatigue test. The technique has allowed a quick scanning of the entire surface with all possible (pixelwise) locations of micro crack centers and the detection of cracks containing a subpixel opening. An experimental test case has been presented as a design of the method and a comparison has been conducted with a replica technique.
YANG Jianbin et al [14] used dualtree complex wavelet transform tool in signal and image processing. From the above discussed survey, it is found that complex wavelet transform outperformed in the sense of shiftinvariance, direction and anti aliasing.
Hence, a dual tree wavelet based image inpainting techniques using the approaches proposed by Cai, Chan, and Shen is proposed in this paperwork towards optimizing the overall performance of the technique on level of recovery and time.
III.Proposed Image Inpainting
Technique
Let a be an image in the original image domain
D
a { aij ;1 i P,1 i Q }
(1)
And the a' be known, observed region and D is the
technique has been implemented using Matlab, Surfer
and Visual Fortran programming. Experimental results have shown that their technique has performed
inpainting domain. The intensity value
(ai ) 0 (i) (i)
(2)
effectively on digitized images suffering from cracks. Dayal R. Parhi and Sasanka Choudhury [23] have conducted a comprehensive review of several techniques in the field of crack detection in Beam
in the domain D where is the noise term. The proposed system finds an image b that matches
0 in D and have meaningful content in the domain
Like Structure. Sensibility analysis of experimentally measured frequencies as a decisive factor for crack
D since the value of
(ai )
is arbitrary when
iD .
identification has been employed widely in the last few decades because of its straightforwardness. The determination of crack parameters such as depth and location is complicated. Several techniques have been discussed on the basis of dynamic analysis of Crack. The techniques mostly used for crck detection were fuzzy logic neural network, fuzzy system, hybrid neuro genetic algorithm, artificial neural network, artificial intelligence.
K.N.Sivabalan and D.Gananadurai [24] have utilized Gabor filter and Gaussian filter in order to remove the texture elements in the digital image by separating the defected area. Then, a fast searching algorithm which uses feature extraction parameters has been proposed to find the defected pixels and to robustly segment it. Their proposed method was appropriate for both texture and non texture images. Consequently, the algorithm has successfully detected the damage in the digital texture image using non
The proposed system consists of the following steps
Initial value assignment, converting to frequency domain, coefficients thresholding, Reconstruction and Iterative image inpainting.

Initial value assignment using nearest neighbor algorithm
Initially the closest entries of a' are identified
and replaced using nearest neighbor algorithm. The selection of closest entries can be realized in two methods, first, as is, on the set of entities, and, second by considering only entities with non missing entries in the attribute corresponding to that of targets missing entry. The proposed system uses the second approach for initial assignment of the damaged portion. The nearest neighbor algorithm is given below.
Step 1: Read an initial value a' .
Step 2: Find K neighbors of a' .
texture methods.


Rupil et al. [25] have introduced a digital image correlation technique for recognizing and calculating automatically the micro cracks on the surface of a
Step 3: Find the data matrix a'
K neighbors.
consisting of a' and
Step 4: Apply an imputation algorithm to a' and impute missing values in a' .
Initially the diagonal matrix follows.
D is obtained as
Step 5: Repeat the above steps until a' is filled.

Conversion of image to frequency domain by means of wavelet
D 1
ij
0
if aij
ij
if a '
(7)
The proposed system uses the Mband Complex 2 D Dual tree wavelet transform which posses the
Subsequently the initial guess of the original
l
n
image is done. by using the For n=1,2,.
unique geometrical features for frequency domain conversion. This decomposition provides local, multi
f *Shrink ( , )
. By using the shrinkage
scale directional analysis. The wavelet transform is self possessed of cascading Mband filter banks. The
procedures as in [14] are carried out for all the M
bands of 2DCWT coefficients. As follows
Mband trees are obtained by performing two Mband multi resolution analyses in parallel in the real case,
or four in the complex case. The dual tree
0
.l
shrink(u, )  l 
if l 
if l '
(8)
decompositions are shift variant, with each trend keeping the same characteristics when the data is delayed. Different sub bands and two sets of coefficients preferentially capture different directions. The Mband biorthogonal wavelet decomposition
[l]
Where l is the given intensity. And then the iterative algorithm
ln1 Dl (I D) fl
of L2 (R)
is based on the joint use of two sets of
m M m M
(9)
is repeated until the n convergence. Using [25], if
basic functions 0 , m
which l*
(i) 0
for every
satisfy the following scaling equations expressed in the frequency domain.
is the output of (35) then
values I of (1), then it will be the solution of the
m
M1/ 2
(M ) 0( )0
( )
interpolation problem. Otherwise the solution
(3)
l * *Shrink ( ,)
will be the denoising and
l*
M1/ 2
(M ) ( ) ( )
interpolation problem.
m 0 0
(4)
Here 0 is the father wavelet and
m is mother
3.2.3 Reconstruction
f
Let be the vector of image samples, the
wavelets. The mother wavelets are obtained through the Hilbert transform that uses the fourier analysis.
vector of coefficients produced by the primal M band
H
Along with this, the interpolation functions can
decomposition and
be the vector of coefficients
also be used for the separation of the signals.

Direction Extraction in the different sub bands
After the decomposition, the sub bands are combined
produced by dual one. The global decomposition operator is
D : f C D1 f
together in a linear fashion to extract the directions
CH
D f
2
(10)
from the images. Some linear combinations of the primal and dual sub bands are used to extract the
Where
D1 U1F1
and
D2 U 2 F2
F1 and
local directions present in the image. The defined analytic wavelets for direction sub bands are
F 2 being the pre filtering operations and U1 and
k (t) m (t) i H(t) m 21/ 2 m
m
(t)
(5)
U 2 be the orthogonal m band decomposition then
the following can be proved. Assume that
x( p g, q l) g ,lZ 2 is an orthonormal family of
m
k (t)
m
21/ 2
i H(t)
(6)
L2 (R2 ).
Provided that there exist
The above functions are used to extract the
I J I
(R*)3
[ , ]2
e e 0 for almost all x y ,
directions that falling in the first and third quadrant of the frequency plane. Likewise the real part of the tensor product of an analytic wavelet and anti analytic
 x(
(11)
x ,
y ) 
< Ie ,
(
x )  A 0
are used to select the frequency components which are localized in the second /fourth quadrant of the
 x( p 2 y , q
( p,q )(0,0)
2z ) 2 J
2 4
I I
x x 0
(12)
frequency plane. After the direction extraction, the thresholding is applied on the images.

Coefficients thresholding
The D is the frame operator. The dual frame reconstruction operator is given by
I (F1' F1 F 2' F 2)1 (F1'U11 F 2'U 21 H )
(13)
Where
F1' designates the ad joint of an operator
F1 . The formula (33) minimizes the impact of possible errors in the computation of the wavelet coefficients. U11 and U 21 are the inverse of M
band wavelet transforms and F1' , F 2'
and
( (F1' F1 F 2' F 2)1 correspond to filtering with
 (F1*( ) 2  (F 2*(
) 2
1
p
q
frequency responses
1 p q ,
1 p q
and
( F1 (
) 2  F1 (
) 2 )1
respectively.
2
p
q
Thus the proposed technique restores the original image from the damages or lost.

Experimental Results and Analysis
The proposed scheme is simulated on Matlab 2010a using test bed that contains 5 standard test images. They are Image1: Barbara, Image2: Boat, Image3: Fruits, Image4: Peppers and Image5; Lena. All the images were taken in Grey Scale Mode and size of 512×512. For the damaged images, above mentioned images are manipulated in three levels for providing the cracked images.
For the quantitative analysis of the proposed technique, the Peak Signal to Noise Ratio (PSNR), Standard deviation to Mean Ratio(S/M) and Tic/Toc functions in Matlab are taken.
The time analysis is done on the aggregate time of 5 test images for a scheme. For the comparison, Discrte Wavelet Transformation (DWT), Haar, Daubechies, and Complex Wavelet Transformation (CWT) are taken from the literature survey [1520].
IV.1 PSNR and S/M Ration Analysis
The fig.1 to fig.10 shows the screen shot of the proposed techniques PSNR analysis. The PSNR values are listed in the tab.1, tab.2 and tab.3. The entire results are taken for three level cracks in the input images.
From the observation of above mentioned values, it is found that the proposed scheme improves the PSNR 5% at an average for all the input images at all the levels. It shows that the proposed scheme able to replace the damaged or lost pixel values with the values that are very close to the original pixel values.
Figure 1: Proposed Technique Outputs for Image1.
Figure 2: DWT, Haar, Daubechies and CWT Outputs for Image 1.
Figure 3: Proposed Technique Outputs for Image 2.
Figure 4: DWT, Haar, Daubechies and CWT outputs for Image2.
Figure 5: Proposed Technique Outputs for Image 3.
Figure 6: DWT, Haar, Daubechies and CWT Outputs for Image3.
Figure 7: Proposed Technique Outputs for Image 4.
Figure 8: DWT, Haar, Daubechies and CWT Outputs for Image 4.
Figure 9: Proposed Technique Outputs for Image 5.
Figure 10: DWT, Haar, Daubechies and CWT Outputs for Image 5.
Table 1: Performance Comparison of Proposed Techniques for Crack Level 1
Image1 Image2
Image3
Image4
Image5
Total
Average
Standard deviation
S/M
DWT
15.682659 17.38496
17.204759
16.78396
19.2851
86.34143
17.26829
1.307092
0.075693
Haar
14.702661 16.87209
17.003936
16.52693
18.72821
83.83382
16.76676
1.43463
0.085564
Daubechies
14.74743 17.06256
17.133125
16.59035
18.96635
84.49982
16.89996
1.506654
0.089151
CWT
15.528204 17.34496
17.211023
16.71531
19.22632
86.02581
17.20516
1.337609
0.077745
Proposed
15.668087 17.3652
17.218906
16.77732
19.23909
86.26859
17.25372
1.29388
0.074991
Table 2: Performance Comparison of Proposed Techniques for Crack Level 2
Image1
Image2
Image3
Image4
Image5
Total
Average
Standard
deviation
S/M
DWT
12.31413
13.62292
15.3487
14.82145
15.72731
71.83451
14.3669
1.395405
0.097126
Haar
11.84051
13.33268
15.2288
14.66204
15.5623
70.62632
14.12526
1.534543
0.108638
Daubechies
11.86531
13.46868
15.29738
14.73502
15.62703
70.99343
14.19869
1.542118
0.10861
CWT
12.18224
13.58814 15.35559 14.80449 15.70939 71.63985 14.32797 1.444204 0.100796
Proposed
12.29484
13.61145 15.37255 14.82146 15.75517 71.85546 14.37109 1.415034 0.098464
Table 3: Performance Comparison of Proposed Techniques for Crack Level 3
Image1
Image2
Image3
Image4
Image5
Total
Average
Standard
deviation
S/M
DWT
11.38259
13.32602
14.08648
13.54998
12.84451
65.18958
13.03792
1.027402
0.078801
Haar
11.18111
13.05052
13.99956
13.38804
12.70727
64.32649
12.8653
1.055409
0.082035
Daubechies
11.16304
13.16882
14.04156
13.44157
12.81528
64.63029
12.92606
1.082766
0.083766
CWT
11.33501
13.28759
14.09376
13.48232
12.84611
65.04478
13.00896
1.037682
0.079767
Proposed
11.45881
13.30298
14.11342
13.52093
12.89258
65.28872
13.05774
0.99662
0.076324
Comparative Analysis on PSNR
18
16
PSNR Values
14
12
10
8 Crack Level 1
6
4 Crack Level 2
2 Crack Level 3
0
Techniques used for Comparison
Figure 11: Comparative Analysis on PSNR
reconstructions. However, the time analysis is taken for aggregate time of 5 images. Therefore, by considering single image, the performance degradation is about 16%. By considering the optimization, such performance degradation is accepted in the proposed scheme. At the same time, the proposed scheme retains the same level of time complexity with respect to CWT and it is around 50 seconds at an average for the input grey scale image of size 512×512. The results are shown in fig. 12 and fig.13.
Table 4: Time Analysis of Proposed Technique with Existing
Techniques.
Comparative Analysis on S/M Ratio
Name of the
Complexity Level
S / M r a t i o V a lu e s
0.12
0.1
0.08
0.06
0.04
0.02
0
Crack Level 1 Crack Level 2 Crack Level 3
Technique Level 1 Level 2 Level 3
DWT 9.0566 11.2536 10.2459
HAAR 37.6459 37.8556 37.7662
DAUBECHIES 40.9068 40.4051 39.3322
CWT 175.3135 218.7938 267.5719
PROPOSED 174.2567 215.8767 250.5678
All the values are given in seconds. For a scheme time value for any level refers the aggregate time of 5 test images.
Techniques used for Comparison
Time Analysis of Proposed Scheme
Figure 12: Comparative analysis on S/M Ratio.

Time Complexity Analysis of the Proposed Technique
Another experiment is conducted on the proposed technique for measuring its time complexity towards applying the technique for energy aware computing applications such as mobile, WiFi Networks. Such applications use the techniques that
300
Tim e Ta k en f o r E x ecu ti o n in Sec o n d s
250
200
150
100
50
0
Techniques used for Comparison
Complex ity Level Level 1
Complex ity Level Level 2
Complex ity Level Level 3
are expected to utilize the power to the level best minimum for its all operations. In those cases, time complexity is playing major role and it must be reasonable and minimum.
From the table 4, it is clear tha the proposed scheme is not improves the time with respect to the other techniques. The overall time analysis of the proposed scheme decrease 80% at an average with respect to other techniques. The reasons are use of multi band wavelet decomposition and
Figure 12: Comparative analysis of proposed technique on Time Taken for Execution for aggregate of all the Input Image at an Average.
From the observation it is concluded that the proposed scheme is best suitable for performing image inpainting and for the energy aware computing allocations sector, its time complexity need to be further fine tuned.
Time Analysis of Proposed Scheme
T i m e T a k e n f o r E x e c u t i o n in S e c o n d s
60
50
40

Joyeux, L., Buisson, O., Besserer, B., Boukir, S., Detection and Removal of Line Scratches in Motion Picture Films, IEEE Computer Society Conference on Computer Vision and Pattern
30
20
10
0
Techniques used for Comaprison
Complexity Level Level 1
Complexity Level Level 2
Complexity Level Level 3
Recognition, Page No. 548553, 1999.

S. Kim, N. Bose, H. Valenzuela, Recursive reconstruction of high resolution image from noisy undersampled multiframes, IEEE Trans. Acoust., Speech, Signal Process., Vol.No. 38, Issue No. 6, Page No. 10131027, Jun.1990.

N. Kingsbury, Complex wavelets for shift invariant analysis and filtering of signals, Appl. Comput. Harmon. Anal., Vol.No. 10, Issue No.

Figure 13: Comparative analysis of proposed technique on Time Taken for Execution for Individual Input Image at an Average.


Conclusion
In this paper, an Mband wavelet based image inpainting scheme is proposed for digital grey scale images. It uses Mband wavelet to locate the cracks in an image and fill the same. The experiment shows that the proposed technique retains the same level of time complexity with respect to Complex Wavelet Transform technique and it is around 50 seconds at an average for the input image of size 512×512. At the same time, it failed to improve the time complexity with respect to other well known techniques in the literature Discrete Wavelet Transform, Haar and Daubechies. The proposed scheme time complexity is reduced 16% at an average for an input grey scale image of size 512×512. The reasons are use of M band wavelet decomposition and reconstruction in the proposed technique. It is concluded that the proposed scheme can be applied in energy aware computing applications after fine tuning the time complexity of the technique.
References

Chen, Fang, Suter, David, Motion Estimation for Noise Reduction in Historical Films: MPEG Encoding Effects, In the Proceedings of 6th Digital Image Computing: Techniques and Applications (DICTA2002) Conference, Page No. 207212, 2002.

Sachin V, Solanki, A. R. Mahajan, Cracks Inspection and Interpolation in Digitized Artistic Picture using Image Processing Approach, International Journal of Recent Trends in Engineering, Vol.No. 1, Issue 2, May 2009.

A.C. Kokaram, R.D. Morris, W.J. Fitzgerald,
P.J.W. Rayner, Detection of missing data in image sequences, IEEE Transactions on Image Processing, Vol. No. 11, Issue 4, Page No. 1496 1508, 1995.

Y. Huang, M. Ng, Y. Wen, Fast image restoration methods for impulse and Gaussian noise removal, IEEE Signal Process. Lett., Vol.No.16, Issue No. 6, Page No. 457460, Jun. 2009.
3, Page No.234253, May 2001.

Manuel M. Oliveira, Brian Bowen Richard, McKenna YuSung Chang, Fast Digital Image Inpainting, In the Proceedings of the International Conference on Visualization, Imaging and Image Processing (VIIP 2001), Marbella, Spain. September 35, 2001.

Marcelo Bertalmio, Guillermo Sapiro, Vicent Caselles and Coloma Ballester, Image inpainting, Proceedings of the 27th annual conference on Computer graphics and interactive techniques, 2000.

Yong Hu, Chunxia, Zhao, A Local Binary Pattern Based Methods for Pavement Crack Detection, Journals of pattern recognition research, 2009.

M. Figueiredo, J. BioucasDias, and M. Afonso, Fast framebased image deconvolution using variable splitting and constrained optimization, in Proc. IEEE Workshop Statist. Signal Process. Cardiff, U.K., 2009, Page No.109112.

M. Figueiredo, J. BioucasDias, Restoration of Poissonian images using alternating direction optimization, IEEE Trans. Image Process.,Vol.No. 19, Issue No. 12, Page No. 31333145, Dec. 2010.

Mohammad Ali, LotfollahiYaghin, Mohammad Amin Hesari, Using Wavelet Analysis in Crack Detection at the Arch Concrete Dam under Frequency Analysis with FEM, World Applied Sciences Journal, Vol. No. 3,Issue No.4, Page No.691704, 2008.

YANG Jianbin, Image inpainting using complex 2D dualtree wavelet transform, Applied Mathematics, Vol.No. 26,Issue No.1, Page No.7076, Applied Mathematics – A Journal of Chinese Universities, 2011.

D. Gabay, B. Mercier, A dual algorithm for the solution of nonlinear variational problems via finite element approximation, Comput. Math. Appl., Vol. No. 2, Issue No. 1, Page No. 1740, 1976.

T. Goldstein, S. Osher, The split Bregman method for L1regularized problems, SIAM J. Imag. Sci., Vol.No. 2,Issue No. 2, Page No. 323 343,2009.

S. Hemami, R. Gray, Subbandcoded image reconstruction for lossy packet networks, IEEE Trans. Image Process., Vol.No. 6, Issue No. 4, Page No.523539, Apr. 1997.

I. A. Ismail, E. A. Rakh, S. I. Zaki, M. A. Ashabrawy, M. K. Shaat, Perceptible Grouping of Cracks in nuclear piping using wavelet transformation, International Journal of Computer and Electrical Engineering, Vol.No. 1,
Issue No. 5, December, 2009

A.C. Kokaram, R.D. Morris, W.J. Fitzgerald,
P.J.W. Rayner, Interpolation of missing data in image sequences, IEEE Transactions on Image Processing, Vol.No.11, Issue No.4, Page No. 15091519, 1995.
[20]M. Browne, M. Dorn, R. Ouellette, T. Christaller,S. Shiry, Wavelet Entropybased Feature Extraction for Crack Detection in Sewer Pipes, In the proceeding of the 6th International Conference on Mechatronics Technology, 2002.
[21]Gunamani Jena, Restoration of Still Images using Inpainting techniques, International Journal of Computer Science & Communication, Vol. No.1, Issue No. 2, Page No. 7174, July December 2010. [22]I. A. Ismail, E. A. Rakh, S. I. Zaki, M. A.Ashabrawy, M. K. Shaat, Crack detection and filling, using steepest descent method, International Journal of Computer and Electrical Engineering, Vol.No. 1, Issue No. 4, October, 2009.
[23]Dayal R. Parhi, Sasanka Choudhury, Analysis of smart crack detection methodologies in various structures, Journal of Engineering and Technology Research, Vol.No. 3,Issue No.5,Page No. 139147, May 2011. [24]K.N.Sivabala,D.Gananadurai, Efficient defect detection algorithm for gray level digital images using Gabor wavelet filter and gaussian filter, International Journal of Engineering Science and Technology (IJEST),Vol.No. 3, Issue No. 4,Apr 2011
[25] J. Rupil, S. Roux,F. Hild,and L.Vincent, Fatigue micro crack detection with digital image correlation, The Journal of Strain Analysis for Engineering Design, Vol.No. 46, Issue No. 6, 2011.