Materials Today: Proceedings xxx (xxxx) xxx

Contents lists available at ScienceDirect

Materials Today: Proceedings

journal homepage: www.elsevier.com/locate/matpr

Image processing application in character recognition

Pradeep K. Nanda a, Laxmi Goswami b
a
Department of Electronics and Communication Engineering, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha, India
b
Department of Electronics and Communication Engineering, Sanskriti University, Mathura, Uttar Pradesh, India

a r t i c l e i n f o a b s t r a c t

Article history: Recently, extraction of text or characters from images has become a wide field of research in the field of
Available online xxxx image analysis, neural networks, artificial intelligence and machine vision. The aim of character recogni-
tion is to extract and recognize the text from images and can be thus used to analyze contents, docu-
Keywords: ments, or retrieve certain information etc. Image analysis has various processes and algorithms
Character recognition available for text extraction like edge detection, point detection, corner detection etc. for extracting
FAST the text from images. Many steps are involved in text recognition including pre-processing, segmenta-
Image processing
tion, feature extraction, and classification. This paper addresses FAST algorithm for corner detection
Image analysis
Text extraction
applied on the images for text recognition. The FAST algorithm detects and determines the presence of
a corner by testing a small area around the potential center of the corner with the help of intensity value.
This approach of text recognition can be efficiently used even with images with noise content or blurs.
Ó 2021 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Virtual
Conference on Sustainable Materials (IVCSM-2k20).

1. Introduction OCR follows certain steps in the processing which includes

scanning of input image, pre-processing, segmentation, feature
In the world of digitalization, the documents in paper form need extraction, classification, recognition and post-processing. The
to be converted to digital form for easy search and storing of doc- scanned images are noisy due to extra details incorporated in the
uments for which recognition of the text or characters in images is images thus the pre-processing step is a requirement for removal
done. Character recognition is the method for detecting and recog- of noise by using various filters like Wiener Filter [3], Gabor filter
nizing the characters from raw input data thus identifying the text [4] etc. the pre-processing also includes conversion of RGB image
in the images. It contributes to the advanced automation processes into binary or a gray scale image for better results. Segmentation
and acts as a bridge between machine and man in applications of of image is done to simply the image so that meaningful informa-
various fields including artificial intelligence, neural networks, tion can be extracted for further analysis. Feature extraction is the
computer vision, biometric etc. Character recognition is the pro- step to extract the relevant and required information or the text
cess in which the input data is classified as per the predefined from the images which is further classified by using the classifica-
character class [1]. Optical character recognition [2] is used for rec- tion in image analysis with a predefined set and thus the recogni-
ognizing the text or characters from the image and converts it to a tion of text or characters is performed which is further followed by
form that can be manipulated by machine. Both online and offline post processing of the output if necessary. The steps followed in
processes are available in optical character recognition. optical character recognition are as shown in Fig. 1.
Character recognition is the technique to detect, identify and
segment the characters present in the images. This technique of
recognizing characters from the images, converts them to a 1.1. Optical scanning
machine readable form. The recognition process is utilized in the
field of automation and interfacing between human and machines. An optical scanner is used to identify the characters in the
This technique being advantageous in the areas of pattern recogni- image. The process of scanning is performed to convert the texts
tion is being researched over years due of its wide range of in an image to a text file. The images from the documents to be
applications. processed are scanned by the use of optical scanner. The scanned
image is converted to a gray scale image from colored image for

https://doi.org/10.1016/j.matpr.2021.03.697
2214-7853/Ó 2021 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Virtual Conference on Sustainable Materials (IVCSM-2k20).

Please cite this article as: P.K. Nanda and L. Goswami, Image processing application in character recognition, Materials Today: Proceedings, https://doi.org/
10.1016/j.matpr.2021.03.697
P.K. Nanda and L. Goswami Materials Today: Proceedings xxx (xxxx) xxx

template matching [5] the extracted data is matched with the pre-
loaded data in the system. The maximum matching template deci-
des the character present in the image.

1.5. Classification and recognition

Classification is a step for sorting the text extracted from the

images on the basis of certain classes which is followed by recog-
nition. Thus the term classification is used to classify an image
from its visual content. Image recognition [6] is a combination of
image detection and classification which individually has the abil-
ity to detect the text and classify it further ending with recognition.
The recognition is used in many applications including the
automation, security, surveillance, etc.

2. Methodology

The detection of feature points in images is essential for many

tasks in machine vision, including optical flow computation, struc-
ture from motion, object tracking, 3D scene reconstruction from
stereo image pairs, etc. The reason that this approach is so popular
Fig. 1. Optical character recognition. is that feature points provide a sufficient constraint to compute
image displacements reliably, and that by processing feature
points the data is reduced by orders of magnitude compared to
further processing so as to save the storage and computational
the original image data, which is particularly important for appli-
work.
cation that must run in real time. The most intuitive types of fea-
ture point is the comer. Comers are image points that show a
1.2. Pre-processing strong two- dimensional intensity change, and are therefore well
distinguished from the neighbouring points. Comer detectors have
The pre-processing step involves conversion of colored image to been widely used as feature point detectors because comers corre-
a binary or gray scale image by binary conversion. This step is fol- spond to image locations with a high information content, and they
lowed only if the input image is not converted by the optical scan- can be matched between images reliably. These matched feature
ner. If the scanner converts the image then preprocessing includes point locations are then taken as an input to high level computer
the removal of noise that might occur due to the scanning of the vision tasks.
image or discrepancy in the image caused due to thresholding of To be useful for feature point matching a comer detector should
the image. Preprocessing is an important step to eliminate all kind satisfy the criteria of consistency where the detected positions
of discrepancies present in the image to get a better output. The should be insensitive to the variation of noise and they should
pre-processing of the images perform the task of noise elimination not move when multiple images are acquired of the same scene.
by filters such as Gaussian filter, skew recognition by Hough trans- The other criteria to be followed is accuracy where the comers
form cluster and binarization by thresholding. should be detected as close as possible to the correct position.
The last criteria being speed which is necessary as even the best
1.3. Segmentation comer detector is useless for the real time tasks if it is not fast
enough.
Segmentation is a step in the process of image analysis where
the regions with useful information are located for separating 3. Fast corner detection
them from irrelevant data, figures and noise. The more precise
use of segmentation [2] is to assign labels to each pixel of the The approach of corner detection is used for feature extraction
image where pixels with same label have certain characteristics. step in the image analysis. Corner detection is an approach used
Edge detection can be used to extract or segment the image by cov- in computer vision systems to extract the relevant information
ering the entire image. The edge detection is applied on the images and features from an image. The application of corner detection
to extract the boundaries of the objects present in the images. This is found in motion detection, tracking images in videos, 3D model-
is done by detecting the difference in brightness in the images i.e. ing, registration of images, object recognition etc. FAST is an algo-
the discontinuities in brightness and giving the pixel values where rithm used to identify certain points in an image which is a pixel
the brightness increases or decreases sharply. value with a well defines location in the image and can be effi-
ciently detected. The method of corner detection is performed
1.4. Feature extraction using features-from-accelerated-segment test(FAST) algorithm
[7]. Corner detection [8] can be performed by using many other
Feature extraction is the process of dimensionality reduction by algorithms like Harris Corner detection, SUSAN corner detection
which an initial set of raw data is reduced to more manageable etc. as shown in Fig. 2 but due to being more efficient FAST algo-
groups for processing. A characteristic of these large data sets is rithm is more preferred for corner detection.
a large number of variables that require a lot of computing This algorithm uses the method where the presence of a corner
resources to process. The image after being pre-processed is used is detected by considering a circular area for testing it out around
as an input for the step of feature extraction. The relevant data is the potential center of the corner. This testing of the corner detects
extracted from the images by using various methods like point dis- the presence of the corner if a portion of the image has high inten-
tribution, series expansion and structural analysis. Further using sity pixels than that of the addition of center and predefined
2
P.K. Nanda and L. Goswami Materials Today: Proceedings xxx (xxxx) xxx

 Pixel ‘‘p” is selected in the input image as shown in figure with

an intensity assumed to be ‘‘I”. This is the pixel that is identified
as an interest point.
 Threshold intensity ‘‘T” is set.
 A circle surrounding the pixel ‘‘p” is set consisting of 16 pixels.
 ‘‘N” no of pixels are considered to be either higher or lower than
the value of ‘‘I” with an addition to threshold ‘‘T”. These N pixels
determine the point of interest.
 At least 3 to 4 pixels must meet the threshold criteria as dis-
cussed in step 4, so that the point of interest can be detected.
If the pixel values do not meet the threshold criteria then the
pint of interest is not present.
 The process is repeated until threshold criteria meets.

4. Applications

Character recognition has many applications in various fields of

technology [10] and has enabled the scanning of documents to
convert them into machine readable format. OCR has reduced the
manual work to a great extent by extracting the relevant informa-
Fig. 2. FAST algorithm.
tion from the documents and automatically storing them in the
system thus providing an accurate and efficient processing system.
Few of the application of character recognition technology have
threshold value or if the portion has low intensity pixels than that been addressed below.
of the sum of center and the predefined threshold value. The algo-
rithm is meant for real time applications and has the sole purpose 4.1. Healthcare
of identifying the points of interest in the given input image where
the points are those pixels whose position is well defined and can The processing of papers in medical field require the technology
be effectively detected. These points of interest hold valuable infor- of optical character recognition to deal with a large volume of data
mation and can be repeatedly used in other images as well. The like patient records, insurance forms, health forms, treatment fee
detection of point of interests in an image can be applied in the forms etc. To keep the database updated, it becomes a necessity
fields like object recognition [9], tracking of videos, matching of to store relevant data in the electronic database so that accessing
images, etc. becomes easy and can be done efficiently with less time consump-
The algorithm followed in FAST as shown in Fig. 3 is as below: tion. By using OCR, extraction of relevant data from the documents
can be done easily and stored in the database.

4.2. Banking sector

Capturing of data automatically has simplified the processing of

banking system thus making the system more efficient and
increased computation speed. OCR technology is vastly used in rec-
ognizing the numbers mentioned in the check to make it easily
readable by the system. The technology of OCR facilitates the cus-
tomers to scan the check and deposit those using electronic devices
like mobile phones for reading and processing all the required
parts of the check.

4.3. Digital libraries

Enormous number of libraries and such places where books are

stored is very common everywhere. The huge number of books in
the library need to be maintained for which digitization plays a
vital role and thus OCR is beneficial to reduce the manual work
and thus can keep track of the books easily by scanning and storing
it in database. Similarly digitized libraries are recently found where
all the books are available online for use and this is a great appli-
cation of character recognition by converting the hardcopy in dig-
ital form thus making access easy.

4.4. Entertainment

Optical character recognition is applied in the entertainment

industry in areas like music. OCR is applied to the musical nota-
tions so that the notations can be digitized and thus easily stored
and accessible in the system. The notations of music are converted
Fig. 3. Steps of FAST algorithm. using specific applications to maintain the quality of the music
3
P.K. Nanda and L. Goswami Materials Today: Proceedings xxx (xxxx) xxx

thus allowing the electronic devices to read and play the written extracting the relevant data from images and also addresses the
music notes. FAST algorithm which is advantageous over many other algorithms
and is used for efficient computation and better result. Various
4.5. Invoice imaging application of OCR has been discussed hence concluding that the
technology of OCR in digitized world plays an important role. This
Being used in various business applications for record keeping paper concludes that character recognition performed using FAST
of the finances, OCR plays an important role in companies and corner detection method gives an efficient output which is error
organizations. It simplifies the collection and analysis of the rele- free, thus resulting in increased accuracy in the field of image anal-
vant data for processing. Invoice imaging is used in all sectors like ysis and digitization.
hospitals, banks or other government and private agencies.
Declaration of Competing Interest
4.6. Handwriting recognition
The authors declare that they have no known competing finan-
The ability to interpret the handwritten input from sources such cial interests or personal relationships that could have appeared
as paper, photographs, etc. is termed as handwriting recognition. to influence the work reported in this paper.
The images of the text are sensed by the optical scanner or intelli-
gent word recognition. By using the steps followed in the character References
recognition, the handwritten form of text is scanned and matched
in the database. [1] F. Brucker, F. Benites, E. Sapozhnikova, Multi-label classification and extracting
predicted class hierarchies, Pattern Recognit. 44 (3) (2011) 724–738.
[2] A. Chaudhuri, K. Mandaviya, P. Badelia, S. K. Ghosh, Optical character
4.7. Captcha recognition systems, in Studies in Fuzziness and Soft Computing, 2017.
[3] Y. Wen, Y. Lu, J. Yan, Z. Zhou, K.M. von Deneen, P. Shi, An algorithm for license
plate recognition applied to intelligent transportation system, IEEE Trans.
A method for confirming the use of various platforms by human Intell. Transport. Syst. 12 (3) (2011) 830–845.
being, captcha is used. An image consisting of a series of letters or [4] S.S. Sarwar, P. Panda, K. Roy, Gabor filter assisted energy efficient fast learning
numbers or a combination of both is generated. With increase in convolutional neural networks, Proceedings of the International Symposium
on Low Power Electronics and Design, 2017.
technology, human activities related to education, travel, and [5] P. Hidayatullah, N. Syakrani, I. Suhartini, W. Muhlis, Optical character
many more are carried out via internet. To prevent hacking and recognition improvement for license plate recognition in Indonesia, in
unauthorized access, captcha is used where character recognition Proceedings - UKSim-AMSS 6th European Modelling Symposium, EMS 2012,
2012.
plays an important task.
[6] A. Kashyap, B. Suresh, A. Patil, S. Sharma, A. Jaiswal, Automatic number plate
recognition, in Proceedings - IEEE 2018 International Conference on Advances
5. Conclusion in Computing, Communication Control and Networking, ICACCCN 2018, 2018.
[7] F. Borisyuk, A. Gordo, V. Sivakumar, Rosetta: Large scale system for text
detection and recognition in images, in Proceedings of the ACM SIGKDD
In the digital world, OCR is an important technology for recog- International Conference on Knowledge Discovery and Data Mining, 2018.
nition of data and translating it to machine readable form making [8] E. Mueggler, C. Bartolozzi, D. Scaramuzza, Fast event-based corner detection, in
British Machine Vision Conference 2017, BMVC 2017, 2017.
the access easy. The data in digital form which is in the machine [9] K. Wang, B. Babenko, S. Belongie, End-to-end scene text recognition,
readable form can be easily transferred within electronic devices Proceedings of the IEEE International Conference on Computer Vision, 2011.
thus reducing the manual work and saving time consumption. This [10] Amarjot Singh, Ketan Bacchuwar, Akshay Bhasin, A Survey of OCR
Applications, Int. J. Mach. Learn. Comput. (2012) 314–318, https://doi.org/
paper discusses about the processing of character recognition 10.7763/IJMLC.2012.V2.137.
using image analysis and explaining various steps required for