• Vol 8, No 5 (2017)
  • Electrical, Electronics, and Computer Engineering

Histogram Equalization Implementation in the Preprocessing Phase on Optical Character Recognition

Peter Pangestu, Dennis Gunawan, Seng Hansun


Publish at : 31 Oct 2017 - 00:00
IJtech : IJtech Vol 8, No 5 (2017)
DOI : https://doi.org/10.14716/ijtech.v8i5.877

Cite this article as:
Pangestu, P.., & Gunawan, D..& Hansun, S. 2017. Histogram Equalization Implementation in the Preprocessing Phase on Optical Character Recognition. International Journal of Technology. Volume 8(5), pp.947-956
37
Downloads
Peter Pangestu Computer Science Study Program, Faculty of Engineering and Informatics Universitas Multimedia Nusantara, Jl. Scientia Boulevard, Gading Serpong, Tangerang, Banten 15811, Indonesia
Dennis Gunawan Computer Science Study Program, Faculty of Engineering and Informatics Universitas Multimedia Nusantara, Jl. Scientia Boulevard, Gading Serpong, Tangerang, Banten 15811, Indonesia
Seng Hansun Computer Science Study Program, Faculty of Engineering and Informatics Universitas Multimedia Nusantara, Jl. Scientia Boulevard, Gading Serpong, Tangerang, Banten 15811, Indonesia
Email to Corresponding Author

Abstract

A 2014 report from Digital Marketing Philippines stated that the number of web applications with visual content as their main product has increased significantly. Image processing technology has also undergone significant growth. One example of this is optical character recognition (OCR), which can convert the text on an image to plain text. However, a problem occurs when the image has low contrast and low exposure, which potentially results in information being hidden in the image. To address this problem, histogram equalization is used to enhance the image’s contrast so the hidden information can be shown. Similar to X-ray scanning used in the medical field, histogram equalization processes scanned images that have low brightness and low contrast. In this study, histogram equalization was successfully implemented using OCR preprocessing. The test was done with a dataset that contains dark background images with low light text; the successful outcome resulted in the ability to show 74.95% of the information hidden in the image.

Contrast enhancement; Histogram equalization; Image processing; Information hiding; Optical character recognition

Introduction

In 2014, Digital Marketing Philippines, a digital consultant, issued an analysis of a survey conducted on visuals used in web applications. The analysis explained that the use of visual content in web applications has developed very rapidly (Digitalmarketingphilippines.com, 2014). This suggests that visual message delivery methods, such as infographics and charts, can provide users with more information and draw greater attention. These developments have also been followed by the development of image processing technologies, such as face detection, object detection, and optical character recognition (OCR) (MathWorks.com, 2016).

Using OCR, images can be produced through a process that includes a series of arrangements, background separation, and matching characters (Sánchez et al., 2012). OCR can be achieved using machines that are currently popular. An OCR readeris not versatile; it must be supported by good conditions and it must meet suitable image matching criteria. Light-dark settings and resolution also affect the performance of an OCR reader. The success of machine translation is also affected by the engine as well as the techniques used in the OCR (Abbyy-developers.eu, 2015). Therefore, OCR cannot be run optimally if the image inserted does not support the desired conditions. Thus, preprocessing is very important in order to create an image that is  


ready to be processed. One important preprocessing step is the improvement of the image with dark and light colors.

No OCR reader can read all the conditions of an image perfectly (Abbyy-developers.eu, 2015). Various tests have been conducted with a variety of datasets (iapr-tc11.org, 2015). However, the majority of the images contained in the dataset collection have had fairly good image conditions (quite bright and good contrast), and they have supported the delivery of clear information from the image (no hidden information). Therefore, to ensure that OCR can be used for a variety of image conditions, improvements in the color, light, and dark elements used in an image (color adjustment) are needed. Some of the methods commonly used to improve the condition of an image include histogram equalization, Wiener filtering, median filtering, decorrelation stretch, and unsharp mask filters (MathWorks.com, 2016).

In the present study, histogram equalization was selected as the image enhancement method because it is similar to X-ray scanning that is used in the medical field to scan organs. Histogram equalization was used to clarify the background and the object of regional differences, and to identify information hidden in the images due to low light and low contrast (Akhlis & Sugiyanto, 2011). In addition, this method is considered fairly common and easy to apply to an image (Alginahi, 2010).

Experimental Methods

The research methodology used in the present study was implemented in the following stages.

3.1.    Learning and Consulting

A literature review was conducted to identify previous research on this topic. This phase was done by collecting supporting data associated with the present research study. The data collection was done by reviewing various types of scientific work, such as books, journals, and articles. This phase is carried out so that the research study can be conducted in accordance with the provisions presented in previous studies in order to produce a valid conclusion.

3.2.    Designing the Application and Identifying the Analysis Requirements

After collecting the supporting data, we conducted a needs analysis to determine the standard to be used in research. In addition to the analysis, we designed the application that will be used as media in the study. This design resulted in several documents, namely flowcharts and the structure of the storage table. Figure 6 shows the procedures we undertook to implement the histogram equalization method used in this research.

 

Figure 6 Flowchart of the histogram equalization method used in the study

 

3.3.    Programming

After establishing the design and determining the analysis requirements, we programmed the application. The application served as a media liaison between the users and the systems used for implementation. The application was designed so users can enter their own samples and obtain rapid analysis of them.

3.4.    Testing and Debugging

After completing the programming stage, the application was tested. The tests were conducted using all the functions that were made in the programming stage (Stage 3). In addition to testing all the functions, we also performed tests on all the possibilities that could occur in the application when it is used by a user.

3.5.    Collecting Samples

After the testing was completed, the application was deemed ready to be used in real-time by users. Therefore, we collected the samples. The sample collection was done by looking for a random sample in accordance with the needs generated in the designing the application and identifying the analysis requirements (Stage 2).

3.6.    Analysis of the Test Results

After obtaining a variety of samples from users, we further analyzed the data to determine the impact of the implementation results and conclusions from the application of the histogram equalization method used in the study.

3.7.    Report and Documentation

To complete the research activities, we wrote a report to document the procedures and findings of the research study.

 

Results and Discussion

After designing and implementing the program, further testing was done using the samples that were collected. The samples collected by searching for dark background samples tended to be black, have low contrast, and were not crowded. The analysis aimed to determine if the image conditions were in accordance with the application of histogram equalization in general. The tests were performed offline using a sample dataset, which contained images with a dark background picture in dominant black (RGB # 000000). Then, some texts were added on each image. Selection of the color of the text was done by adding a hexadecimal value 10 for each component of red, green, and blue (RGB) of the background; for example, RGB #000000 to RGB #101010. The test results are presented in Table 1.

 

Table 1 The testing results

No.

Title

Text

Target

Success (%)

 

Before histogram equalization

After histogram equalization

 

Before histogram equalization

After histogram equalization

1

sugar.png

 

lele jumbo

lele jumbo

0

100

 

2

manus.png

 

I will kill you

I will kill you

0

100

 

3

phantom-assassin-dota-2-dota-2740(1).png

 

assasin

assasin

0

100

 

4

cool_dark_wallpaper.png

 

cool

cool

0

100

 

5

12June2012-Low-light-focusing-lrg

 

LION KING

LION KING

0

100

 

6

images_(5).jpg

 

the dragon

the dragon

0

100

 

7

images.jpg

 

S N THE DARKEST PLACES

THERE IS LIGHT EVEN IN THE DARKEST PLACES

0

48.6486486

 

8

Normal21.bmp

POLITICS H appuintment he

POLITICS H Appointment he general manage

63.1578947

0

 

9

Normal22.bmp

TAIW

TAIW Taipei

40

0

 

10

Normal23.bmp

Not to

not to

100

0

 

11

Normal25.bmp

Chand Said

e arguments change said

47.3684211

0

 

12

Norhe mal26.bmp

the World

the world

100

0

 

13

Normal27.bmp

mm mm mbigs she

news feature abies lishe

20.8333333

0

 

14

Normal28.bmp

atlc exerci of detention tlso reneged

atlc exerci of detention lse reneged

97.14286

0

 

15

Normal29.bmp

in

0

0

 

16

Shadow1.bmp

THURSDAY

zrea ysterda

THURSDAY

100

0

 

17

Shadow2.bmp

massacre and mothers tell it like it

tell it like

a massacre and mothers tell it like it is

93.93939

30.3030303

 

18

Shadow4.bmp

Conclusion

Histogram equalization was successfully implemented during the OCR preprocessing phase by using a web-based program (PHP). The research results demonstrate that implementing histogram equalization during OCR preprocessing improved the OCR performance. The dataset that contained a collection of predominantly black and dark images inserted with dark texts increased the percentage of success to approximately 74.95%.

In the future, additional studies can be conducting using other advanced methods to improve the image contrast without causing a lot of noise, such as adaptive histogram equalization and contrast-limited adaptive histogram equalization. We believe that more advance methods could result in better image contrast than is possible using a normal histogram equalization method.

 

Acknowledgement


References

Abbyy-developers.eu, 2015. Image Processing and Binarisation for Camera OCR. Available online at https://abbyy.technology/en:features:ocr:cameraocr-preprocessing-binarisation

Ahmad, N., Hadinegoro, A., 2012. Metode Histogram Equalization untuk Perbaikan Citra Digital. In: Proceedings of Seminar Nasional Teknologi Informasi & Komunikasi Terapan 2012, Semarang: Indonesia, INFRM, pp. 439–445

Akhlis, I., Sugiyanto, 2011. Implementasi Metode Histogram Equalization untuk Meningkatkan Kualitas Citra Digital. Jurnal Fisika, Volume 1(2), pp. 70–74

Alginahi, Y., 2010. Preprocessing Techniques in Character Recognition, Character Recognition, Minoru Mori (Ed.), ISBN: 978-953-307-105-3, InTech. Available online at http://cdn.intechopen.com/pdfs-wm/11405.pdf

Digitalmarketingphilippines.com. 2014. Amazing Facts and Statistics about Visual Web. Available online at http://digitalmarketingphilippines.com/wp-content/uploads/2014/01/Amazing-Facts-and-Statistics-about-Visual-Web.jpg

Gonzalez, R., Woods, R., 2008. Digital Image Processing (3rd ed). New Jersey: Prentice-Hall

Holley, R., 2009. How Good Can It Get? Analysing and Improving OCR Accuracy in Large Scale Historic Newspaper Digitisation Programs. Available online at http://www.dlib.org/dlib/march09/holley/03holley.html

iapr-tc11.org, 2015. Datasets List - TC11. Available online at http://www.iapr-tc11.org/mediawiki/index.php/Datasets_List

Krutsch, R., Tenorio, D., 2011. Histogram Equalization. Guadalajara: Freescale Semiconductor Application Note Number AN4318, Rev 0

MathWorks.com, 2016. Image Processing and Computer Vision Examples. Available online at http://www.mathworks.com/examples/product-group/matlab-image-processing-and-computer-vision

Mithe, R., Indalkar, S., Divekar, N., 2013. Optical Character Recognition. International Journal of Recent Technology and Engineering (IJRTE), Volume 2 (1), pp. 72–75

Rachman, E.M.B.P., 2014. Histogram Equalisation. Available online at http://ilmukomputer.org/wp-content/uploads/2014/02/Histogram-Equalisation-Pengolahan-Citra-Digital.odt

Rice, S.V., Jenkins, F.R., Nartker, T.A., 1995. The Fourth Annual Test of OCR Accuracy. Available online at http://www.expervision.com/wp-content/uploads/2012/12/1995.The_Fourth_Annual_Test_of_OCR_Accuracy.pdf

Sánchez, J., Perronnin, F., de Campos, T., 2012. Modeling the Spatial Layout of Images Beyond Spatial Pyramids. Pattern Recognition Letters, Volume 33(16), pp. 2216–2223

Xcitex, Inc., 2010. Image Processing: Brightness, Contrast, Gamma, and Exponential/Logarithmic Settings in ProAnalyst. Available online at http://www.xcitex.com/Resource%20Center/ProAnalyst/Application%20Notes/App%20Note%20151%20-%20Image%20Processing%20Brightness,%20Contrast,%20Gamma%20and%20Exponential.pdf

Zybert, C., 2014. How does Optical Character Recognition Work. Available online at http://nedocs.com/how-does-optical-character-recognition-work/

Copyright © 2017 Faculty of Engineering
International Journal of Technology
IJTech secretariat, Engineering Center Bld., 2nd Fl.
Faculty of Engineering, Universitas Indonesia
Depok 16424, Indonesia.