Skew detection using wavelet decomposition and projection

prole analysis
Shutao Li
a,
*
, Qinghua Shen
a
, Jun Sun
b
a
College of Electrical and Information Engineering, Hunan University, Changsha 410082, China
b
Fujitsu R&D Center Co., Ltd., Eagle Run Plaza B1003, Xiaoyun Road No. 26, Chaoyang District, Beijing 100084, China
Received 22 February 2006; received in revised form 22 September 2006
Available online 28 November 2006
Communicated by A.M. Alimi
Abstract
In this paper, a novel document skew detection algorithm based on wavelet decompositions and projection prole analysis is pro-
posed. First, the skewed document images are decomposed by the wavelet transform. The matrix containing the absolute values of
the horizontal sub-band coecients, which preserves the texts horizontal structure, is then rotated through a range of angles. A projec-
tion prole is computed at each angle, and the angle that maximizes a criterion function is regarded as the skew angle. Experimental
results show that this algorithm performs well on document images of various layouts and is also robust to dierent languages. The
eects of various wavelet basis, number of decomposition levels, and parameters of the criterion function are investigated too.
2006 Elsevier B.V. All rights reserved.
Keywords: Skew detection; Document analysis; Projection prole analysis; Wavelet transform
1. Introduction
Document skew detection is necessary for most docu-
ment analysis system and many methods have been devel-
oped. Existing methods typically use: (1) projection
proles analysis (Bloomberg and Kopec, 1993; Bloomberg
et al., 1995; Ishitani, 1993; Liolios et al., 2002; Postl, 1986);
(2) nearest neighbors (Jiang et al., 1999; Liolios et al., 2001;
Lu and Tan, 2003); (3) Hough transform (Amin and
Fischer, 2000; Yu and Jain, 1996; Ham et al., 1994); (4)
mathematical morphology (Das and Chanda, 2001; Naj-
man, 2004); (5) cross-correlations (Akiyama and Hagita,
1990; Yan, 1993; Chaudhuri and Chaudhuri, 1997; Chen
and Ding, 1999; Gatos et al., 1997).
The traditional projection prole (PJ) based approach
for skew detection was proposed by Postl (1986). First,
the input document is rotated through a range of angles
and a projection prole is calculated at each angle. Fea-
tures are then extracted from each projection prole to
determine the skew angle. This is computationally expen-
sive as it is performed directly on the original document
image. Moreover, it is sensitive to the layout of the docu-
ment image.
An improved projection prole based approach was
proposed by Bloomberg and Kopec (1993). The original
document image is down-sampled before the projection
prole is computed. The following operations are based
on the sampled image. Therefore, the image data to be pro-
cessed is reduced and the computational cost is reduced sig-
nicantly. However, a major weakness is that its detection
accuracy is inuenced by the document image layout. It
often fails on document images with multiple font styles/
sizes or those that contain a large amount of non-text
regions (such as pictures, tables or graphics).
The second class of the skew detection methods is based
on the nearest neighbors (Jiang et al., 1999; Liolios et al.,
2001; Lu and Tan, 2003). Here, the angle between each
0167-8655/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.patrec.2006.10.002
*
Corresponding author. Tel.: +86 731 8672916; fax: +86 731 8822224.
E-mail addresses: shutao_li@hnu.cn, shutao_li@yahoo.com.cn (S. Li).
www.elsevier.com/locate/patrec
Pattern Recognition Letters 28 (2007) 555562
http://www.paper.edu.cn

connected component and its nearest neighbor is com-

puted, and a histogram of the angles is formed to determine
the skew angle.
The Hough transform has also been used for document
image skew detection (Amin and Fischer, 2000; Yu and
Jain, 1996; Ham et al., 1994). The idea is that the highest
number of co-linear pixels is on lines that are co-incident
with the baseline of the text. Its disadvantage is that one
has to rst extract text regions in the document image
before the Hough transform is applied. This is not trivial
for document images whose layouts are complex. More-
over, both the nearest neighborhood and Hough trans-
form-based methods are computationally expensive.
On the other hand, mathematical morphology based
methods represent the skewed document images with a more
compact representation of its components, particularly text
objects, such as words, lines, and paragraphs, where they
can be represented by their rectilinear bounding boxes
(Das and Chanda, 2001; Najman, 2004). This kind of algo-
rithm is fast as well as independent of script forms. How-
ever, it is dicult to choose the morphological operators.
Cross-correlations (CC) (Akiyama and Hagita, 1990;
Yan, 1993; Chaudhuri and Chaudhuri, 1997; Chen and
Ding, 1999; Gatos et al., 1997) and the maximum variance
of transition counts (TC) (Ishitani, 1993; Chen and Wang,
2000) are also used for document image skew detection.
Chou et al. proposed a skew estimation method based
on piecewise covering of objects (PCP) such as textlines,
gures, forms, and tables (Chou et al., 2007). First, the
document image is divided into a number of non-overlap-
ping slabs in which each object is covered by parallelo-
grams. Then, the skew angle is estimated based on these
parallelograms or, equivalently, their complementary
regions. Chou et al. demonstrate that their method is faster
and more robust than PJ, TC and CC.
Wavelet transforms have been widely used in image pro-
cessing because of its inherent advantages such as sensitiv-
ity to abrupt variations, multi-resolution (Mallat, 1989;
Daubechies, 1992). After one level of decomposition, the
original image is decomposed into four sub-images which
are approximations of the source image and details in the
vertical, horizontal and diagonal directions.
In this paper, we present a document skew detection
algorithm using wavelet decompositions and projection
prole analysis. Using the wavelet decomposition, the hor-
izontal structure of the document image can be extracted
and emphasized, which will improve the accuracy of prole
projection method. On the other hand, the computational
cost is less than one tenth of that of the original after 2-
level wavelet decomposition. Experimental results show
that the proposed algorithm can deal with document
images with complex layouts and multiple font size. It is
also language independent. Moreover, there is no need to
select text lines from the input document image before
the projection prole is computed.
This paper is organized as follows. Section 2 introduces
wavelet transform. Section 3 briey describes the tradi-
tional projection prole based approach to skew estima-
tion. In Section 4 the proposed algorithm is described in
detail. Experimental results of the proposed algorithm are
presented in Section 5. Discussions are given in Section 6.
Finally, conclusion of this paper is presented in Section 7.
2. Two-dimensional wavelet decomposition
In our algorithm, a two-dimensional discrete wavelet
transform (2D-DWT) is used. The roles of the highpass l-
ters, low-pass lters and downsamplers in 2D-DWT are
shown in Fig. 1. Here, G(z) is a highpass lter, H(z) is a
lowpass lter, and #2 for downsampling. The lowpass
Fig. 1. Illustration of two-dimensional discrete wavelet transform.
Fig. 2. An example of wavelet decomposition. (a) Original document image, (b) one-level decomposition using symlets wavelets.
556 S. Li et al. / Pattern Recognition Letters 28 (2007) 555562
http://www.paper.edu.cn

lters smooth the image while the highpass lters look for
detailed information in the image.
As shown in Fig. 1, when 2D-DWT is implemented on
an image, four frequency bands (LL, LH, HL and HH)
are obtained. Among these four sub-bands, the LL sub-
band corresponds to an approximation of the original
document image, the LH sub-band provides details in the
horizontal direction, the HL sub-band provides details in
the vertical direction, while the HH sub-band provides
details in the diagonal direction. Fig. 2 gives a document
image and its rst level decomposition result using the
symlets wavelet.
3. Projection prole analysis
A popular method for skew detection uses horizontal
projection prole because the texts in most document
images are aligned along horizontal lines. When the hori-
zontal projection prole is applied on an M N image, a
column vector of size M 1 is obtained. Elements of this
column vector are the sum of pixel values in each row of
the document image. An example of the projection proles
of an unskewed and skewed image are shown in Fig. 3. As
can be seen, peaks in Fig. 3(c), which correspond to the
horizontal projection prole of the unskewed image, are
taller than those in Fig. 3(d), which correspond to the hor-
izontal projection prole of the skewed image. In fact,
peaks in Fig. 3(c) average around 170 while peaks in
Fig. 3(d) average around 80. Based on this signicant dif-
ference, the skew angle can be estimated.
4. The proposed algorithm
The proposed algorithm is based on the wavelet trans-
form and horizontal projection prole (Fig. 4):
1. If the input document image is not a gray one, transform
it into a gray-scale image, denoted I
g
.
2. Decompose I
g
with 2D-DWT. Then, four frequency
sub-bands (LL, LH, HL and HH) are obtained. Here,
the LH sub-band is selected because it preserves the
horizontal structure of the document image.
Fig. 3. Projection proles of unskewed and skewed document images. (a) Unskewed document image, (b) document image rotated by 6, (c) horizontal
projection prole of (a), (d) horizontal projection prole of (b).
S. Li et al. / Pattern Recognition Letters 28 (2007) 555562 557
http://www.paper.edu.cn

3. Denote the matrix formed by the absolute values of the

LH sub-band coecients by I
H
. I
H
is rotated through
[a, a] and a horizontal projection prole is computed
at each angle.
4. The skew angle is estimated by using the criterion func-
tion. A proper criterion cannot only reduce computation
time but also obtains more accurate detection result. In
our algorithm, peaks in each of the projection prole are
selected as follows.
Let H be one of the horizontal projection proles.
Therefore H is a column vector. Divide H into M pieces
each size of N 1 (denoted v
1
, v
2
, . . . , v
M
). Then the K
largest values of v
m
(1 6 m 6 M) are selected (denoted
v
m1
, . . . , v
mk
). The criterion function is dened as
sum
X
M
m1
X
K
k1
v
mk
1
The angle that maximizes this criterion function is regarded
as the estimated skew angle.
To speedup the search process, our method is performed
in a coarse-to-ne mode. First, the search step size is 2 to
get a coarse estimate. Denote the founding optimal angle
by I. Then the search space is changed from I 1 to
I + 1 with a step size of 0.5. Denote the next optimal
angle obtained by J. Finally, it searches for the best skew
angle L within the range J 0.5 to J + 0.5 with a ner
step size of 0.1.
5. Experimental results
5.1. Test dataset
The proposed method is evaluated on the open dataset
provide by Chou et al. (http://dar.iis.sinica.edu.tw/Down-
load%20area/skew.htm). It contains 500 images, which
are generated by scanning a collection of dierent docu-
ments, including newspapers, books, magazines, and jour-
nals, with a resolution of 300 dpi. The range of the skew
angle is [15, 15].
These images are divided into ve categories (Table 1):
(1) English documents dominated by text in the horizontal
direction. They are either single-column or double-col-
umns; (2) Chinese or Japanese documents. These are also
dominated by text. But text-lines in these images are either
in the horizontal or vertical direction, or both horizontal
and vertical directions; (3) Documents with large amount
of non-text regions, such as gures, graphics; (4) Docu-
ments dominated by tables and forms; (5) Documents in
multiple languages, such as Arabia, Hindi, Greek. Figures
and graphics are also contained in these images. Some
examples are shown in Fig. 5.
5.2. Experimental results
To evaluate our proposed method, we compare it with
some standard methods such as PJ (projection prole),
TC (transition-counts), CC (cross-correlations) and PCP
(piecewise covering by parallelograms). Results of these
methods on the same set of test images are available from
Fig. 4. Schema of the proposed method.
Table 1
Test samples
Document type Number of test
images
1st Category English documents 100
2nd Category Chinese and Japanese documents 100
3rd Category Documents containing large-scale
gure
100
4th Category Documents containing forms or
tables
100
5th Category Multilingual documents 100
558 S. Li et al. / Pattern Recognition Letters 28 (2007) 555562
http://www.paper.edu.cn

A
u
t
h
o
r
'
s



p
e
r
s
o
n
a
l



c
o
p
y
papers by Chou et al. (2007). For PJ, method proposed by
Postl (1986) is implemented. For TC, method proposed by
Chen and Wang (2000) is implemented. CC stands for
method proposed by Chaudhuri and Chaudhuri (1997),
and PCP stands for the method proposed by Chou et al.
(2007).
The estimation error is usually used to evaluate the eec-
tiveness of the skew detection method. It is dened as the
Fig. 5. Examples of the ve categories of documents. (a) 1st Category, (b) 2nd category, (c) 3rd category, (d) 4th category, (e) 5th category.
S. Li et al. / Pattern Recognition Letters 28 (2007) 555562 559
http://www.paper.edu.cn

dierence between the estimated and actual angles. In our

experiment, both the average and variance of the errors
are computed. The wavelets basis lter is db1 and number
of decomposition levels is 2. N is xed to 25 and the num-
ber of peaks K is xed to 5.
Results are shown in Tables 2 and 3. Table 2 presents
the average and variance of errors on all test images while
Table 3 gives the average and variance of the top 80% min-
imum errors.
From these two tables, we can see that our proposed
method can achieve the best estimation results for the
2nd and 5th categories. For images of the 1st and 4th cat-
egories, the estimation result of our method is similar to
those of the other methods. However, for the images in
the 3rd category, the proposed method is worse than
PCP, TC and CC methods. The main reason is that these
images are dominated by gures. But compared with PJ,
which is the worst, our method performs much better
because of the introduction of the wavelet transform.
Another advantage of the proposed method is that its var-
iance of errors is much smaller than the other methods
(also shown in Tables 2 and 3).
The proposed method takes less than one tenth of the
computation time of PJ because of the use of 2-level wave-
let decomposition. For example, using Matlab 6.5 on a
AMD Sempron 512 MHz PC running Windows XP, for
images with size of 2480
*
3508, our method takes 55 s
while PJ takes 766 s. TC and CC take much more time than
PJ and PCP runs a little faster than PJ (Chou et al., 2007).
The proposed algorithm can run faster with higher decom-
position level.
6. Discussions
6.1. Eect of wavelet basis
To evaluate the eect of wavelet basis lters on the
detection result, dierent wavelet basis lters containing
Daubechies, Symlets and biorthogonal are tested.
In the experiments, second-level wavelet decomposition is
performed. The vector length N is xed to 25, and the num-
ber of peaks K is xed to 5. Table 4 gives the mean and var-
iance of the errors using dierent wavelets basis lters.
From the table, it can be concluded that the db1 is gen-
erally the best.
6.2. Eect of wavelet decomposition level
Eects of the number of decomposition levels are shown
in Table 6. The wavelet basis used are db1 and sym2.
N and K are the same as in Section 6.1. As shown in Table
5, dierent categories require dierent decomposition levels
Table 2
Comparisons of the error rates
Methods 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
Our method 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
PCP 0.149 0.129 0.139 0.143 0.231 0.135 0.111 0.127 0.077 0.075
PJ 0.230 0.206 0.496 0.591 7.787 9.049 0.160 0.163 2.050 5.816
TC 0.185 0.180 0.171 0.155 0.249 0.223 0.150 0.18 0.176 0.240
CC 0.166 0.144 0.180 0.192 0.345 0.325 0.139 0.146 0.197 0.230
Table 3
Comparisons of the top 80% error rates
Methods 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
Our method 0.208 0.015 0.068 0.005 0.450 0.011 0.071 0.008 0.040 0.002
PCP 0.102 0.096 0.088 0.070 0.178 0.102 0.062 0.073 0.051 0.050
PJ 0.153 0.140 0.254 0.263 3.419 4.934 0.096 0.105 0.208 0.264
TC 0.148 0.131 0.108 0.091 0.183 0.144 0.078 0.084 0.105 0.072
CC 0.115 0.109 0.132 0.096 0.223 0.186 0.075 0.078 0.129 0.125
Table 4
Eect of the wavelet basis on the performance (level = 2, N = 25, K = 5)
Wavelet basis 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
db1 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
db4 0.254 0.086 0.173 0.044 0.504 0.020 0.140 0.025 0.089 0.015
bior2.2 0.269 0.092 0.126 0.035 0.498 0.019 0.152 0.027 0.072 0.010
sym2 0.255 0.086 0.153 0.064 0.460 0.020 0.196 0.021 0.075 0.012
560 S. Li et al. / Pattern Recognition Letters 28 (2007) 555562
http://www.paper.edu.cn

to get the best results. To have a balance between compu-

tation time and detection accuracy, level two is the best
choice.
6.3. Eect of N and K
Tables 6 and 7 show the eects of vector length N and
number of peaks K on the results. The wavelet basis is
db1 and the number of decomposition levels is 2. In Table
6, K is xed to 5 while in Table 7, N is xed to 25. From
Table 6, we can see that the parameter N has small inu-
ence on the detection accuracy. And the smaller N is, the
faster the algorithm works. As is shown in the Table 7,
changing K has no inuence on the result and the speed
of the proposed algorithm.
6.4. Detection range
To investigate how large a skew angle the proposed
method can tolerate, all the zero-angle images are also
rotated by 15, 30 and 45 in both clockwise and anti-
clockwise directions. From the detection results shown in
Table 8, we can conclude that the proposed method works
well in the range of [45, 45] and the skew angle has little
eect on the accuracy. This demonstrates that the proposed
method is eective since the skew angle is usually very
small in practice. For angles in the range of (90, 45)
or (45, 90), the detection result depends on the layout of
the document image. If text-lines in the images are horizon-
tal, the estimated angle is close to its actual skew angle. But
for images whose text-lines are vertical or both vertical and
Table 5
Eect of the number of wavelet decomposition levels on the performance (N = 25, K = 5)
Wavelet basis level 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
db1 1 0.260 0.086 0.139 0.035 0.505 0.017 0.135 0.023 0.066 0.009
2 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
3 0.209 0.074 0.271 0.231 0.496 0.028 0.121 0.022 0.095 0.018
sym2 1 0.257 0.086 0.126 0.035 0.505 0.017 0.127 0.023 0.071 0.011
2 0.255 0.084 0.153 0.064 0.460 0.020 0.125 0.021 0.075 0.117
3 0.183 0.061 0.450 0.970 0.473 0.062 0.134 0.027 0.117 0.034
Table 6
Eect of the vector length N on the performance (db1, level = 2, K = 5)
Vector length N 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
15 0.256 0.088 0.132 0.031 0.524 0.027 0.127 0.021 0.073 0.009
20 0.255 0.088 0.140 0.064 0.505 0.024 0.122 0.021 0.079 0.010
25 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
Table 7
Eect of the number of peaks K on the performance (db1, level = 2, N = 25)
Number of peaks K 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
5 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
10 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
15 0.256 0.088 0.126 0.035 0.499 0.019 0.125 0.021 0.071 0.006
Table 8
Detection range of the proposed method
Skew angles 1st Category 2nd Category 3rd Category 4th Category 5th Category
Mean Var Mean Var Mean Var Mean Var Mean Var
45 0.255 0.088 0.129 0.031 0.502 0.024 0.122 0.021 0.075 0.008
30 0.256 0.087 0.127 0.030 0.505 0.022 0.125 0.024 0.073 0.009
15 0.256 0.088 0.126 0.035 0.500 0.024 0.124 0.021 0.072 0.006
15 0.256 0.088 0.126 0.035 0.500 0.024 0.124 0.021 0.072 0.006
30 0.255 0.088 0.127 0.032 0.504 0.022 0.125 0.023 0.074 0.009
45 0.255 0.087 0.129 0.031 0.502 0.024 0.122 0.021 0.075 0.008
S. Li et al. / Pattern Recognition Letters 28 (2007) 555562 561
http://www.paper.edu.cn

horizontal, the estimated angle is prone to the complemen-

tary angle of the actual angle. This problem can be solved if
prior knowledge about the layout of the image is known.
7. Conclusions
In this paper, a novel algorithm for document skew
detection based on wavelet decomposition and projection
prole analysis is proposed. From the point of wavelet
decomposition, the horizontal sub-band is the best candi-
date for document skew detection. It can save computation
time and improve the estimation accuracy. The experimen-
tal results show that the proposed algorithm works well on
various documents including Chinese, Japanese, English,
diagrams, etc.
Acknowledgements
We would like to give many thanks to the anonymous
reviewers for helpful comments and constructive sugges-
tions. We also thank Prof. Fu Chang for providing the
source data. This paper is supported by the National Nat-
ural Science Foundation of China (No. 6040204), Program
for New Century Excellent Talents in University, and the
Excellent Youth Foundation of Hunan Province.
References
Akiyama, T., Hagita, N., 1990. Automated entry system for printed
documents. Pattern Recognition 23, 11411154.
Amin, A., Fischer, S., 2000. A document skew detection method using the
Hough transform. Pattern Anal. Appl. 3, 243253.
Bloomberg, D.S., Kopec, G.E., 1993. Method and apparatus for
identication and correction of document skew. Xerox Corporation,
US Patent 5,187,753.
Bloomberg, D.S., Kopec, G.E., Dasari, L., 1995. Measuring document
image skew and orientation. Proc. SPIE 2422, 302316.
Chou, C.H., Chu, S.Y., Chang, F., 2007. Estimation of skew angles for
scanned documents based on piecewise covering by parallelograms.
Pattern Recognition 40, 443455.
Chaudhuri, A., Chaudhuri, S., 1997. Robust detection of skew in
document images. IEEE Trans. Image Process. 6, 344349.
Chen, M., Ding, X., 1999. A robust skew detection algorithm for grayscale
document image. In: Proc. 5th Internat. Conf. on Document Analysis
and Recognition, pp. 617620.
Chen, Y.K., Wang, J.F., 2000. Skew detection and reconstruction based
on maximization of variance of transition-counts. Pattern Recognition
33, 195208.
Das, A.K., Chanda, B., 2001. A fast algorithm for skew detection of
document images using morphology. Int. J. Document Anal. Recog-
nition 4, 109114.
Daubechies, I., 1992. Ten Lectures on Wavelets. SIAM, Philadelphia.
Gatos, B., Papamarkos, N., Chamzas, C., 1997. Skew detection and text
line position determination in digitized documents. Pattern Recogni-
tion 30, 15051519.
Ham, Y.K., Chung, H.K., Kim, I.K., Park, R.H., 1994. Automated
analysis of mixed documents consisting of printed Korean alphanu-
meric texts and graphic images. Opt. Eng. 33, 18451853.
Ishitani, Y., 1993. Document skew detection based on local region
complexity. In: Proc. 2nd Internat. Conf. on Document Analysis and
Recognition, Tsukuba Science City, Japan, pp. 4952.
Jiang, X., Bunke, H., Widmer-Kljajo, D., 1999. Skew detection of
document images by focused nearest-neighbor clustering. In: Proc. 5th
Internat. Conf. on Document Analysis and Recognition, Bangalore,
pp. 629632.
Liolios, N., Fakotakis, N., Kokkinakis, G., 2001. Improved document
skew detection based on text line connected component clustering. In:
Proc. Internat. Conf. on Image Processing, Thessaloniki, Greece, vol.
1, pp. 10981101.
Liolios, N., Fakotakis, N., Kokkinakis, G., 2002. On the generalization of
the form identication and skew detection problem. Pattern Recog-
nition 35, 253264.
Lu, Y., Tan, C.L., 2003. A nearest-neighbor chain based approach to skew
estimation in document images. Pattern Recognition Lett. 24, 2315
2323.
Mallat, S., 1989. A theory for multiresolution signal decomposition: The
wavelet representation. IEEE Trans. Pattern Anal. Machine Intell. 11,
674693.
Najman, L., 2004. Using mathematical morphology for document skew
estimation. In: Proc. SPIE, Document Recognition and Retrieval XI,
vol. 5296, pp. 182191.
Postl, W., 1986. Detection of linear oblique structures and skew scan in
digitized documents. In: Proc. 8th Internat. Conf. on Pattern Recog-
nition, Paris, France, pp. 687689.
Yan, H., 1993. Skew correction of document images using interline cross-
correlation. CVGIP: Graph. Models Image Process. 55, 538543.
Yu, B., Jain, A.K., 1996. A robust and fast skew detection algorithm for
generic documents. Pattern Recognition 29, 15991630.
562 S. Li et al. / Pattern Recognition Letters 28 (2007) 555562
http://www.paper.edu.cn