Modified approach to transform arc from text to linear form text a preprocessing stage for ocr

Signal & Image Processing : An International Journal (SIPIJ) Vol.5, No.4, August 2014
MODIFIED APPROACH TO TRANSFORM ARC-FORM-
TEXT TO LINEAR-FORM-TEXT: A
PREPROCESSING STAGE FOR OCR
Vijayashree C S 1, Shruthi C V 2 and Vasudev T 2
1 P.E.T Research Foundation, P.E.S College of Engineering, Mandya-571401, India.
2 Maharaja Research Foundation, Maharaja Institution of Technology,
Mysore-571 438, India.
ABSTRACT
Arc-form-text is an artistic-text which is quite common in several documents such as certificates,
advertisements and history documents. OCRs fail to read such arc-form-text and it is necessary to
transform the same to linear-form-text at preprocessing stage. In this paper, we present a modification to
an existing transformation model for better readability by OCRs. The method takes the segmented arc-form-
text as input. Initially two concentric ellipses are approximated to enclose the arc-form-text and later
the modified transformation model transforms the text in arc-form to linear-form. The proposed method is
implemented on several upper semi-circular arc-form-text inputs and the readability of the transformed text
is analyzed with an OCR.
KEYWORDS
Artistic-text, Arc-form-text, Linear-form-text, OCR.
1. INTRODUCTION
Document image analysis (DIA) is an important research discipline in the area of Image
Processing. Many researchers are working on different problems of document images starting
from image acquisition to image understanding [1,2]. The research in this field is focusing to
come out with generic approaches to accomplish automation in document reading, extracting
contents from documents and these have lead into many vibrating research problems [2]. The
results of the research on the above problems are converging towards the generic solutions to
major issues in DIA.
In spite of considerable research work in the area of DIA, a major issue which is not sufficiently
addressed is, reading or extracting the contents of the text which appear in artistic-form in a
document. Many documents, especially certificates, marks cards, sign boards, logos, etc., have
artistic text. In addition, many official seals on the documents for authentication are also artistic
in nature. The contents of such artistic-text definitely have some valuable information that has to
be processed. Most of the graduation certificates issued by the Universities contain the name of
the university in artistic form. If such document has to be processed by an Optical Character
Reader (OCR), the OCR should be able to read such artistic-text or proper pre-processing is
required to make that text readable by OCR. Few such artistic-texts in documents are, text
appearing in triangular-form, arc-form, circular-form, wave form. Samples of such artistic-texts
are shown in Figure 1. The contents of such text normally convey the identity information like
company’s name, type of document, etc., which is the main source for classification of the
document.
DOI : 10.5121/sipij.2014.5407 67

68
Figure 1. Samples of artistic-form text in document
Documents containing artistic-text, when subjected to reading by OCRs, fail to be read, as the
OCRs are developed to read linear texts. Hence, it is necessary to transform artistic-text to linear
text such that OCRs are able to read the contents efficiently. Approaches developed for general
skew detection and correction are not suitable to transform such artistic-text documents into linear
form. Hence, it is required to come out with different approaches that can transform artistic-form
text into linear form text and make the same suitable for reading by an OCR.
2. RELATED WORK
One of the major problems encountered in DIA is implicit/inherent skew noticed in document
images [3,4]. Inherent skew, is due to the natural inclinations of text lines in the document.
Considerable amount of work is reported in literature on explicit skew detection [5-16]. Each of
the approaches reported in literature on explicit skew detection has its own advantages and
limitations, and these approaches are not extendable for detecting inherent skew. Since artistic
texts also have inherent orientation in the document, artistic-texts are said to have implicit skew.
To the best of our efforts while surveying for literature in the direction of implicit skew detection
and correction, we could find the work of Pal et al in detecting multiple implicit skewed lines
within a document[3], i.e., detecting lines within the document having different orientations and
the work carried out by Vasudev et al to transform arc-form-text to linear-form[17]. Vishwanath
et al[18] have proposed connected component Technique for character extraction from document
image having artistic-form-text. The implicit skew in extracted characters is detected using
Hough Transform and corrected. Further Vishwanath et al[19] have proposed Radon transform
for the detection of implicit skew in the extracted characters and their correction. The work
proposed by Vasudev et al[17] performs transformation to considerable extent but suffers from
tilt deformation and an additional stage is required for tilt corrections in the model. Further, the
readability efficiency after transformation is claimed as 84% in this method. This drawback of the
approach proposed by Vasudev et al[17] has motivated us to continue the work to design an
efficient transformation model that transforms the arc-form-text to linear-form-text without tilt
deformation to produce the output suitable to OCR for better readability.
The proposed work assumes that the arc-form-text has been segmented out from the document
and is free from noise. Further, it is assumed that the arc-form-text in the document is either
circular or elliptical in shape and is limited to only in the upper half circle or ellipse. The
proposed model has two stages. The initial stage is to estimate two concentric imaginary ellipses
to enclose the arc-form-text. This stage is performed as proposed in [17] and the same is briefed
in section 3. In the second stage transformation takes place and section 4 describes the modified
transformation model. Experimental results are discussed in section 5. The conclusion of the work
is given in section 6.
3. ENCLOSURE OF TEXT WITHIN SUITABLE ARCS
The transformation model to transform arc-form-text to linear-form requires two imaginary
elliptical arcs [20] to be searched which encloses the arc-form-text. The procedure developed in
[17] is made use in this work. Figure 2 shows sample of arc-form-text and Figure 3 illustrates the

estimation of two imaginary elliptical arcs enclosing the arc-form-text under consideration using
the algorithm given in [17].
69
Figure 2. Arc-form input text Figure 3. Two imaginary arcs drawn to enclose
Arc-form-text
Though the inner arc does not contribute much during the process of transformation, it is useful in
detecting the height of the text and to restrict transformation process to be within the arc-form
region. After enclosing the arc- text between two imaginary suitable arcs, it is required to
transform all the points on this elliptical band into a linear band of points and the same is
explained in the subsequent section.
4. MODIFIED TRANSFORMATION MODEL
The principle adopted to perform transformation is a point processing technique [4]. In this
transformation model, a set of points representing line in one orientation is transformed to
represent a line of points in another orientation. Extending this concept, an arc-form text can be
considered as a set of n consecutive lines in different orientations, where n being the distinct
points on surface of the outer arc. These n lines with different orientations are transformed to n
vertical lines, which results in the text appearing horizontally linear. For comprehension, Figure 4
shows how lines within two arcs having different orientation are represented as n vertical line.
Figure. 4 Representation of n lines in different orientation within two arcs as n vertical lines
A transformation function T can be expressed as,
S = T [F] (1)
where
F = {l1,l2,…,ln} , li i = 1,…,n is the ith line within arcs having m points
S = {lt1,lt2,…ltn } , lti i = 1,…,n is the ith transformed line having m points
T is the transformation function that simply puts the points of li on lti and
li = {p1,p2,…,pm} , pj j = 1,…,m is the jth point on the ith line within the arc

70
lti = {q1,q2,..,qm} , qj j = 1,…,m is the jth point on the ith transformed line
qk = T[pk] k = 1,… m, m is the number of points in the kth line l and transformed line lt
To transform arc-form text to linear, as shown in Figure 5 a series of lines from the centre of the
arc are drawn to each point on the surface of outer arc.
Figure 5. Lines drawn from centre of the ellipse
Algorithm for the transformation is given below:
Input: Text image enclosed with outer and inner arcs.
Output: Linear-form text
1. L is the blank document
2. For each point on the outer arc (xj ,yj)
a. Draw the line from the centre of arc to outer arc (xj,yj)
b. Determine point of intersection on inner arc (xj`,yj`)
c. Consider the line l from (xj,yj) to (xj`,yj`)
d. For each li
Transform m points of li to lti
e. append lti into L.
Figure 6. Transformed text to linear-form
Result of the implemented algorithm for the sample input Figure 2 is shown in Figure 6. It is
evident from Figure 6 that arc-form-text gets transformed to linear form, without tilt
deformations. However variations in height in few characters and distortions are noticed due to
the contributions of aspect ratio and stair case effect in lines.
5. EXPERIMENTAL RESULTS
Experiments are conducted on arc-form-text with different sizes and different arc shapes for
English texts. The results of experiments establish an overall readability between 73% - 100% in
transformed and tilt corrected text which is quite promising and encouraging. The result provides
a better input for OCRs. Experimental results before usage of the proposed approach is shown in

Signal Image Processing : An International Journal (SIPIJ) Vol.5, No.4, August 2014
Table 1. Experimental results illustrated in Table 2 indicate that the proposed approach efficiently
transforms arc-form-texts to linear-form-texts which are suitable for OCRs.
Analysis of readability by an OCR of the text before transformation and after transformation is
performed with respect to English text using the OCR “ Readiris Pro 9 ”. In this process, first, the
samples of arc-form-text are taken as input to OCR and subjected to be read by the OCR and the
result is tabulated in Table 1. The transformed text of proposed method is subjected to be read by
OCR and the result is tabulated in Table 2.
71
Table 1. OCR’s readability of arc-form-text
Input arc-form-text OCR Recognition Readability
0%
0%
0%
0%
0%
0%
0%

72
0%
0%
0%
Table 2. OCR’s readability before and after correction
Input arc-form-text
Transformed text and
OCR Recognition of text
Readability
10/10 - 100%
18/18 - 100%
18/18 - 100%
11/13 - 84.61%
20/21 - 95%
14/14 - 100%

73
19/19 - 100%
HANDLE W1TH CARE
14/14-100%
11/15- 73.33%
15/15-100%
Table 1 indicates that an arc-form-text is given as input to OCR, it recognizes the text as picture
rather than text and retains it in the picture form and the readability of text by OCR is obviously
0%. This demands that most of the OCRs do require linear text for reading. The Table 2 shows
that the transformation of arc-form-text to linear-form-text. Numerator indicates the number of
characters recognized and denominator signifies the total number of characters in the text. The
method has been tested with over 300 plus samples and it shows average readability of 95%.
Table 2 also illustrates the reading efficiency of OCR and shows considerably better results than
the method proposed [17], which was claimed as 84%. Analysis of readability by OCR is done at
only two stages – (i) arc-form-text and (ii) transformed text. Further, the method has been
experimented on the arc-form-text of other languages. The output is promising and few sample
outputs are given in Figures 7 and 8.
Figure 7.a Kannada arc-form input text Figure 7.b Transformed text to linear-form
Figure 8.a Hindi arc-form input text Figure 8.b Transformed text to linear-form

6. CONCLUSIONS
The proposed approach efficiently transforms an arc-form-text without tilt deformations and
thereby makes better readability by an OCR with a readability of 95%. This approach employs the
principle of ellipse drawing algorithm for transformation. Any image processing application with
the computational complexity O(n2) is considered to be less expensive. Because of considerable
time complexity and simple point processing technique in spatial domain, the proposed method is
claimed as a simple and less complex approach to transform arc-form text to linear-form. The
readability by OCR increases as compared to the earlier approach proposed by Vasudev et al[17]
with slight variation in height of characters and distortions. The method can be extended
comfortably to transform text from other languages. A method to restrict variations in the heights
of the characters due to aspect ratio and eliminate distortions due to stair case effect in final
output is under investigation which can enhance the output more effectively and efficiently.
REFERENCES
[1] O’Gorman, Lawrence, Kasturi, Rangachar. Executive Briefing: Document Image Analysis. IEEE
74
Computer Society Press, 1998.
[2] Nagabhushan, P. Document Image Processing, in : Proc. National Pre-Conf. workshop on Document
Processing, India, pp 114, 2001.
[3] Pal U, Mitra M, Choudhri B B. Multi-Skew Detection of Indian Script Documents, in: Proc. Int. Conf.
on Document Analysis and Recognit ion(ICDAR 2001) , 2001.
[4] Vasudev T, Hemanthkumar G, Nagabhushan P. Detection and Correction of Vertical Skew in
Characters, in: Proc. 3rd Int. Conf. on Innovative Applications of Information Technology for
Developing World(AACC 2005) CD version, Nepal, 2005.
[5] Zheng Zhang. Restoration of Curved Document Images Throug 3D Shape Modeling, in: Conf. on
Computer Vision and Pattern Recognition(CVPR2004), 2004.
[6] Amin A, Fischer A. A Document Skew Detection Method Using the Hough Transform. J. Pattern
Aual. Applicat. 3,243-253, 2000.
[7] Kavallieratou E, Fakotakis N, Kokkinakis G. Skew Angle Estimation for Printed and Handwritten
Documents Using the Wigner-Ville Distribution, Image Vis. Comput. 20, 813-824, 2002.
[8] Liolios N, Fakotakis N, Kokkinakis G. On the Generalization of the Form Identification and Skew
Detection Problem. Pattern Recognition(35), 243-264, 2003.
[9] Murali S, Vasudev T, Hemanthkumar G, Nagabhushan P. Language Independent Skew Detection and
Correction of Printed Text Document Images: A Non-rotational Approach. VIVEK – Int. J. Artif.
Intell. 16(2), 08-15,2006.
[10] Shivakumar P, Nagabhushan P, Hemanthkumar G, Manjunath. Skew Estimation by Improved
Boundary Growing for Text Document in South Indian Languages. VIVEK –Int. J. Artif. Intell. 16(2),
15-21, 2006.
[11] Lu Yue, Tan, Chew Lim. A Nearest Chained Aproach to Skew Estimation in Document Images.
Pattern Recognition Lett. 24, 2315-2323, 2003.
[12] Cao,Yang, Wang, Shuhua, Li, Heng. Skew Detection and Correction in Document Image Based On
Straight Line Fitting, Pattern Recognition Lett. 24(12), 1871-1879, 2003.
[13] Vasudev T, Hemanthkumar G, Nagabhushan, P. Segmentation of Characters in Arc-form-text, in:
Proc. on Cognitive System(ICCS 2005), CD version, India, 2005.
[14] Breuel T. The Future of Document Imaging in the Era of Electronic Documents , in: Proc. Int.
Workshop on Document Analysis, India, pp.275-296, 2005.
[15] Kennedy L M, Basu M. Image Enhancement Using a Human Visual System Model, Pattern
Recognition 30(12), 2001-2014, 1997.
[16] Vasudev T, Hemanthkumar G, Nagabhushan, P. An Elliptical Approximation Model for Removal of
Text-Line Bending Deformation at Page Borders in a Document Image, in: Proc. Int. Conf. on
Cognition and Recognition, India, pp.645-654, 2005.
[17] Vasudev T, Hmanthkumar G, Nagabhushan P. Transformation of Arc-form-text to Linear-form-text
Suitable for OCR, Pattern Recognition Letter 28 (2008) 2343-2351, 2008.
[18] Vishwanath C. Kagawade, Vijayashree C.S., Vasudev T. Transformation of Artistic Form Text to
Linear Form Text for OCR Systems, International Conference on Advances in Computing
(ICAdc2012)

[19] Vishwanath C. Kagawade, Vijayashree C.S., Vasudev T. Transformation of Artistic Form Text to
Linear Form Text for OCR Systems using Radon Transform, International Conference on Emerging
Research in Electronics, Computer Science and Technology(ICERECT-12)
75
[20] Donald Hearn, Baker M P. Computer Graphics, Pearson Education, second ed. 2003, 2003
AUTHORS
Vasudev T is currently Professor in the Department of Computer Applications, at
Maharaja Institute of Technology, Mysore. He obtained his Bachelor of Science and
Post-Graduate Diploma in Computer Programming with two Masters Degrees, one in
Computer Applications and the other one in Computer Science and Technology. He was
awarded Ph. D. Degree in Computer Science from University of Mysore. He has 30
years of experience in academics. He has published over 30 articles in reputed journals
in his area of research Digital Image Processing, specifically Document Image
Processing.
Vijayashree C. S, obtained her B.E. Degree in Computer Science from B.I.T,
Bangalore and M.E. Degree in Computer Science from U.V.C.E, Bangalore. She is
pursuing research towards her Ph. D. Degree in Computer Science of University of
Mysore, Mysore, at P.E.S. College of Engineering, Mandya
Shruthi C. V, obtained her B.E. Degree from VTU, Belgaum. Currently she is a student
in the Department of Computer Science and Engineering, Maharaja Institute of
Technology, Mysore. She is pursuing her PG Degree in Computer Science and
Technology of VTU, Belgaum.

Modified approach to transform arc from text to linear form text a preprocessing stage for ocr

More Related Content

Viewers also liked (19)

Similar to Modified approach to transform arc from text to linear form text a preprocessing stage for ocr (20)

Recently uploaded (20)

Modified approach to transform arc from text to linear form text a preprocessing stage for ocr