IRJET- Text Line Detection in Camera Caputerd Images using Matlab GUI

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 06 | June 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 206
TEXT LINE DETECTION IN CAMERA CAPUTERD
IMAGES USING MATLAB GUI
Safvana Sajeev1, Ojus Thomas Lee2
1Post Graduation Student, Dept. of CSE, College of Engineering Kidangoor, Kerala, India
2Associate Professor & HOD, Dept. of CSE, College of Engineering Kidangoor, Kerala, India
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In today’s digital world, the camera based text
pre-processing get an important task that finds applicationsin
many fields like Text2Speech conversion, informationretrieval
etc... So, this led to the development of many pre-processing
methods. This paper proposes a system that combines text line
identification fromimagescontainingprintedandhandwritten
text.The Maximally Stable Extremel Region (MSER)algorithm
is used to determine the character regions in the image. Based
on the geometric property of the region, unwanted or
dissimilar regions are removed. The Optical Character
Recognition (OCR) is used to recognize each character and
detect the word region. After detection, the text can be used to
search for information on theinternet.Inaddition,thedetected
text can be converted to text to assistvisuallyimpairedpeoples.
Our proposed method are tested with data sets based on scene
text,street view data set and handwritten text.Thepercentage
of exact text identification ourmethod in scene dataset ismore
than 50% in any situation (Indoor, Outdoor, Book Cover,
Light, Shadow etc..) , in case of street view dataset is 74% and
that for handwritten dataset is 68%.
Keywords: Text Detection, Text2Speech Conversion,
Information Gathering, MSER, OCR
1. INTRODUCTION
The importance of image processing has increased day by
day due to the availability of cost effective devices for
capturing photos and videos. In today’s digital world, people
captured photos and videos and passed to the social media.
Text is the most expansive means of communication. Text is
embedded into documents or scenes for communication. It
can be observable and readable by others. The narrative in
image contain valuable information and can be used for
numerous image and video based applications like image
search web site, information recovery, reviewing text and
identify text in mobile devices. The size, font, style of textand
the back ground with snow, rock make text detection in
camera captured images more difficult.
Text detection is the major task in text line detection. It is
used in Optical Character Recognition (OCR) and
preprocessing algorithms. The English language text in the
images will be captured.
The main text detection challenges are divided into
three groups:
1) Natural image properties: The characters in natural
images are in different font style, differentsize,uniquecolor,
and unique font alignment.
2) Backgrounds of an image: The background includes
grasses bricks, rocks, are leads to more complexity for
identifying text.
3) Interference factors: The main inference factor is
blurring. From a blurred image text can’t be detected.
The Main Contributions of our work are
1) We use MSER algorithm to determine character
regions in images.
2) OCR is used to recognize characters.
3) Provide methods to convert the detected text to
voice.
4) Add methods to do internetsearchofidentifiedtext.
5) We use scene text, streetviewandhandwrittendata
sets to test the method.
The rest of the document is organized as follows,
Section 2 describes the related works done in this paper,
Section 3 and 4 explain the system architecture and its
working. The Section 5 describes the implementation steps
and Section 6 describes the results obtained. Finally Section
7 is conclusion.
2. RELATED WORK
The emergence of smart phones and digital cameras has
made the text line detection from images in demand. In this
scenario we have made a study on available methods for the
text line identification from images. The indent of the study
was identify the existing methods and to suggest improved
methods for the text line identification. After reviewing the
literature in the existing domain we have found that there
exists the following methods for solving the problem.
1) MSER with Ada boost
2) State Estimation
3) Texture Based SVM
4) Chain Level Classification
5) CMSER Algorithm

6) Graph Model
In the case of MSER with Ada boost, [1] [2] [3] propose a
connected component procedure that extraction the
connected component and form a group of individual
components that contain text. [4] [5] are suggesting
procedure to detecting character content from natural
images. Under the State Estimation category, [6] [7] [8]
initiate state estimation procedure for the extraction of text
in natural images. Here algorithm has two phases. First
phase using the connected components, interline spacing is
calculated. The next phase of the procedure identifies the
text block using the estimated states. [9][10][11] Proposed
texture based SVM method. Texture classification is like a
score image. Each pixel is denoted as the possibility of the
part of region which contain text. [12] Propose a connected
component with SVM method to identifythecharactersfrom
the image. [13] Propose arbitrary text detection in natural
images using chain level classifier. Chain Level Classifier is a
system that produces robust result for detecting text. [14]
Suggest a method to detect road sign text detection. In
addition, [15] [16] Proposed anidentificationmechanism for
traffic sign boards.
3. SYSTEM ARCHITECTURE
The purpose of this paper istorecognizetextfromimages.
The images are scene images at shadow, light, indoor,
outdoor, book cover condition. Next street view images that
contain the shop name, posters. Next oneishandwrittentext
images real time. Here we are using an OCR classifier for the
purpose of text detection. The detected text is converted to
speechforvisuallyimpairedpeoples.Andalsothetextispassed
to Google for gather information about the text. The MSER is
usedtodetecttheregionsintheimage.Afterdetecttheregions
unwanted regions are removed based on their geometric
property. Using OCR texts/words are detected.
Fig -1: System Architecture
Initially upload the image and by using MSER regions are
identified. The detected MSER regions are taken and remove
thenon-textregionbasedontheirgeometricalsimilarity. Using
boundingpropertyeachcharacterinthetextregionisbounded.
Thenexpandtheboundingboxtogetthewords/text. TheOCR
detects text in the bounding box. After the detection detected
textisconvertingtospeechanditisalsosenttoGooglebrowser
for gather information about the text content.
In the case of handwritten text, the detected regions are
taken and the unwanted regions are removed. Here also the
detected region contain characters are bounded. And using
OCR the text is detected. Then detected text is converting to
speech and sent to Google browser.
4. WORKING
Initially login to the system by using username and
password. After login a main window is displayed. The
project works in two streams, for the extraction of two
general categories of text extraction.
The two text extractions are,
1) Text Extraction-1
2) Text Extraction-2
The Text Extraction-1 is for scene and street view image
database. The Text Extraction-2 handle handwrittenimage
database.
In the case of Text Extraction-1 after we input theimage,
the process involved is as follows.
1) Convert image intoMSERregions:Converttheimage
into regions by using MSER algorithm.
2) Remove Non-text region based on their geometric
property: Based on the geometric property of the regions
remove the unwanted/dissimilar region.
3) Characters are bounded using bounding box: Using
bounding box property of the MSER regions, characters are
bounded. The bounding boxes are represented by using
rectangle.
4) Expand the bounding box: Increase sizes of the
bounding box for represent full word/texts. The size is
increased by taking length and breadth values.
5) Text Detection: The bounding box contains
texts/words are detected by using OCR. The OCR read the
entire bounding box and recognizes text content and shown
it.
6) Text is display in file and converts it to speech: The
detected text content in the bounding boxes is read and
writes to a file. Then file content is read and convert it into
speech.
7) Text is sent to Google browser for searching/gather
information: Here also read the file content and it sent to a
browser for collect the information about the text content.
For Text Extraction-2 after we input the image, the
process involved is explained below,
1) Convert image intoMSERregions:Converttheimage
into regions by using MSER algorithm.

2) Extract each region and shown in figures: The
regions contain characters and they are represented by
using bounding box property. The bounding boxes are
shown in rectangle shape and read all bounding boxes and
display as a figure.
3) Compare the bounding box characterwithpreviously
stored character dataset: Compare the characters with
previously stored character dataset. If any match is found
then character is detected and store to a file. If character is
not detected then take next bounding box character. This
process is iterating until all bounding box characters are
detected. The detected characters are form text/words.
4) Text is display in file and converts to speech: The
detected text/words are read and display in a file and
convert it into speech.
5) Text is sent to Googlebrowser forgatherinformation:
Detected text is sent to a browser for collect information
about text content.
5. IMPLEMENTATION
The project was implemented using MATLAB and was
used for the detection of text from images. MATLAB stands
for Matrix Laboratory it is a programing language. It was
developed by MathWorks. MATLAB provide operations like
matrix manipulations operation, plotting graphs and data,
creating user interface, and interfacing with programs
written in other languages.
The GUI required for project was done by using MATLAB
GUI. GUIs stands for graphical user interfaces and it provide
an easy way to run the program or application without any
type commands. The GUI contains controls like menus,
toolbars, buttons, and sliders [17].
The testing of the proposed method was carried outusing
different kinds of images having text embedded in it. The
category of images used involves scene images with text,
images with street sign boards andimageswithhandwritten
text. The images of these categories have been used by
taking following data sets.
1) KAIST_Scene_Text_Database [18]
2) The_Street_View_Text_Dataset[19]
3) Handwritten Text Database
The scene text database is again subdivided into two,
1) Image taken by using mobile phone
2) Image taken by using digital camera
The image taken by using mobile phone is again
subdivided based on their condition. The conditions are,
1). Indoor
2). Outdoor
3). Book Cover
The image taken by using digital camera is again
subdivided based on their condition. The conditions are,
1). Indoor
2). Outdoor
3). Shadow
4). Light
The Street View database contains images like shopfront,
posters, sign boards, headlines etc… The Handwritten
database contains upper case handwritten text images.
6. RESULTS
Using the implementation plans discussed in the previous
sectionwerunthetestingofthedatasetandobtainsacoupleof
results. Acomparisonoftheresultsisrepresentedaspiechart
ineachofthecase. Weclassifytheresultsascorrect,partialand
wrong results. A result is said to be correct if the proposed
method is able to identify the text in the image with >90%
accuracy. In the case of partial results, the text identified is
>60% and <90%. If the number of characters identified from
theimageis<60%accuratethen,theresultissaidtobewrong.
6.1 Mobile Camera Image - Outdoor
Condition (MCI-OC)
We use the scene dataset here with images taken using
mobile camera. The image is taken under an outdoor
condition is considered. Here some results are shows partial
and wrong output. The partial output is due to the difference
in fonts and styles. For example, in some font ‘l’ is treated as
‘u’. The wrong output is due poor edge detection. The
comparisons of accuracy of the results are given in fig 2.
Table 1 gives a statistics of the images in the data set
considered under this category.
Fig- 2: Graph for Image from mobile camera under
outdoor condition (Scene Text Database)

6.2 Mobile Camera Image - Indoor Condition
(MCI-IC)
We use the indoor images shot by mobile camera in the
scene dataset. Here we get 48% of images correctly
identified. The partial output is due to the difference in font
style and background of the image. Some results are shows
wrong output, because the edge detection is not work
properly. The comparisons of accuracy of the results are
given in fig 3. A statistics of the images in the data set
considered under this category is shown in table 1.
Fig - 3: Graph for Image from mobile camera under
indoor condition (Scene Text Database)
6.3MobileCameraImage-BookCoverCondition
(MCI-BC)
The book cover images captured by mobile camera in the
scene data set are used here. Here some results are shows
partial and wrong output. The partial output is due to the
difference in fonts and styles. For example, in some font ‘o’s
treated as ‘0’. Some results are shows wrong output, because
the edge detection is not work properly. The comparisons of
accuracy of the results are given in fig 4. Table 1 gives the
number of images in the data set considered under this
category.
Fig - 4: Graph for Image from mobile camera under book
cover condition (Scene Text Database)
6.4 Digital Camera Image - Shadow Condition
(DCI-SC)
In this section we considered the images under shadow
condition in the scene data set taken by digital camera. Here
someresults are shows partial and wrong output.Thepartial
output is due to the edge detection. The edge detection is not
work properly. Some results are shows wrong output due to
the background. The comparisons of accuracy of the results
are given in fig 5. A listingofthenumberofimagesinthedata
set for this category is presented in table 1.
Fig - 5: Graph for Image from digital camera under
shadow condition (Scene Text Database)
6.5 Digital Camera Image - Indoor Condition
(DCI-IC)
In thissection wealso use theimages taken bydigitalcamera
in indoor condition and are part of scene data set. Here some
results are shows partial and wrong output. The partial
output is due to the difference in font style and edge
detection. Some results are shows wrong output, because of
the background of the image. The comparisonsofaccuracyof
the results are given in fig 6. The number of images in the
data set considered under this category is presented in table
1.
indoor condition (Scene Text Database)

6.6 Digital Camera Image - Outdoor Condition
(DCI-OC)
We use the images shot by digital camera under outdoor
condition and are part of the scene data set. The partial
output is due to the difference infont.Someresultsareshows
wrong output, because of the improper edge detection. The
comparisons of accuracy of the results are given in fig 7. The
listing of images in the data set for this category is presented
in table 1.
outdoor condition (Scene Text Database)
6.7 Digital Camera Image - Light Condition (DCI-
LC)
Set of digital camera images shot in the light condition
which are members of scene data set is used in this section.
The results are shows partial and wrong output. The partial
output is due to the edge detection. Some results are shows
wrong output, because of the image is taken nothorizontally.
The comparisons of accuracy of the results are given in fig 8.
The statistics of images in the data set for this category is
presented in table 1.
Fig - 8: Graph for Image from digital camera under light
condition (Scene Text Database)
6.8 Street View Image (SVI)
The street view text dataset isusedinthissection.Herethe
street view images are considered. The street view images
contain sign boards, posters, shop front, placards, mile stone
etc… Theresults are shows partial output, because theimage
is taken from the street so, the quality of the text is less. The
wrong output is due to the poor edge detection. The
comparisons of accuracy of the results are given in fig 9. A
number of images in the data set for this category is
Fig - 9: Graph for Street View Text Database
6.9 Handwritten Image (HI)
Handwrittenuppercaseletters/charactersareusedinthis
section. The real time and existing handwritten characters
are considered and checks the accuracy of getting correct
output. Here the challenging factor is person writing style of
each character. The partial output is due to mismatch in the
person writing character and exiting alphabet database. For
example sometimes ‘c’ is treated as ‘e’. Thewrongresultsdue
to worst region detection. The region detection isnotproper.
The comparisons of accuracy of the resultsaregiveninfig10.
A list of number of images in the data set for this category is
Fig - 10: Graph for Handwritten Text Database

Table -1: Count and results statistics of the categories of
images used
Image
Details
M
CI-
OC
M
CI-
IC
MCI
-BC
DCI
-SC
DCI
-IC
DCI
-OC
DCI
-LC
SV
I
HI
No. of
Image
s
31 23 32 31 46 71 15 47 32
No. of
correc
t
output
17 11 24 17 60 46 8 35 22
No. of
partial
output
10 8 6 10 10 14 3 7 6
No. of
wrong
output
5 4 2 5 6 11 4 5 4
7. CONCLUSIONS
As technology progress day by day text line detection
become an important task. Here we focus on detecting the
text from image, convert to speech and also have the
provision to gather information the text from internet. The
system is implemented using MATLAB. The text regions are
detecting using MSER region detector. The OCR classifier
detects the text from MSER regions. Here we are using three
types of datasets--Scene Database, Street Database and
Handwritten Database. The results shows that in scene
database we get more than 40% output under any
conditions like if the image is click in a shadow, light, indoor
or outdoor situation. In street database get 74% output as
correct. In handwritten database we get 68% output as
correct.
In future develop an OCR for other languages and apply
some machine learning techniques that convert language
sentences from one language to another.
REFERENCES
1. H. I. Koo, D. H. Kim, Scene text detection via
connected component clustering and non-text
filtering, IEEE transactionsonimageprocessing22
(6) (2013) 2296–2305.
2. X. Chen, A. L. Yuille, Detecting and reading text in
natural scenes, in: Computer Vision and Pattern
Recognition, 2004, CVPR 2004. Proceedings of the
2004 IEEE Computer Society Conference on Vol.2,
IEEE, 2004, pp, II–II.
3. M. S, detecting text in natural scenes with
connected components clustering and non-text
filtering, in: International journal of Engg and
technology (IRJET2016),IRJET,2016,pp.725–728
4. X.-C. Yin, X. Yin, K. Huang, H.-W. Hao,Robusttext
detection in natural scene images, IEEE
transactions on pattern analysis and machine
intelligence 36 (5) (2014) 970–983.
5. X. Yin, X.-C. Yin, H.-W. Hao, K.Iqbal,Effectivetext
localization in natural scene images with mser,
geometry-based grouping and adaboost, in:
Proceedings of the 21st International Conference
on Pattern Recognition (ICPR2012), IEEE, 2012,
pp. 725–728
6. H. I. Koo, N. I. Cho, State estimation in a document
image and its application in text block
identification andtextlineextraction,in: European
Conference on Computer Vision, Springer, 2010,
pp. 421–434.
7. H. I. Koo, N. I. Cho, Text-line extraction in
handwritten chinese documents based on an
energy minimization framework, IEEE
Transactions on Image Processing 21 (3) (2012)
1169–1175.
8. Y.-F. Pan, X. Hou, C.-L. Liu, et al., A hybrid
approach to detect and localize texts in natural
scene images, IEEE Transactions on Image
Processing 20 (3) (2011) 800.
9. K. I. Kim, K. Jung, J. H. Kim, Texture-based
approach for text detection in images using
support vector machines and continuously
adaptive mean shift algorithm, IEEE Transactions
on Pattern Analysis and Machine Intelligence 25
(12) (2003) 1631–1639.
10. K. Jung, K. I. Kim, A. K. Jain, Text information
extraction in images and video: a survey, Pattern
recognition 37 (5) (2004) 977–997.
11. L. G´omez, D.Karatzas,Mser-basedreal-timetext
detection and tracking, in: Pattern Recognition
(ICPR), 2014 22nd International Conference on,
IEEE, 2014, pp. 3110–3115.

12. S. Unar, A. Hussain, M. Shaikh, K. H. Memon, M.
A. Ansari, Z. Memon, A study on text detection
and localization techniques for natural scene
images, IJCSNS 18 (1) (2018) 100.
13. C. Yao, X. Bai, W. Liu, Y. Ma, Z. Tu, Detectingtexts
of arbitrary orientations in natural images, in:
2012 IEEE Conference on Computer Vision and
Pattern Recognition, IEEE, 2012, pp. 1083–109.
14. M. S. Hossain, A. F. Alwan, M. Pervin, Road sign
text detection using contrast intensify maximally
stable extremal regions, in: 2018 IEEESymposium
on Computer Applications & Industrial Electronics
(ISCAIE), IEEE, 2018, pp. 321–325.
15. K. Sumi, M. Arun Kumar, Detection, and
recognition of road traffic signs-a survey,
International Journal of Computer Application
(0975-8887) 160 (3).
16. Y. Tang, X. Wu, Scene text detection and
segmentation based on cascaded convolution
neural networks, IEEE Transactions on Image
Processing 26 (3) (2017) 1509–1520.
17. https://in.mathworks.com/discovery/matlab-
gui.html
18. http://www.iaprtc11.org/mediawiki/index.php?ti
tle = KAIST_Scene_Text_Database
19. http://www.iaprtc11.org/mediawiki/index.php?ti
tle= The_Street_View_Text_Datase
BIOGRAPHIES
Safvana Sajeev, She is a post-
graduation student in College of
Engineering Kidangoor.
Specialization in Computer and
Information Science. She received
the graduation in Computer
Science and Engineering from
Gurudeva Institute of Science and
Technology. Her areas of interest
are Image Processing and
Databases.
Ojus Thomas Lee, is an Associate
Professor in CollegeofEngineering
Kidangoor. His areasofinterestare
Databases, Cloud Computing,
Sensor Networks and Image
Processing.

IRJET- Text Line Detection in Camera Caputerd Images using Matlab GUI

More Related Content

What's hot (19)

Similar to IRJET- Text Line Detection in Camera Caputerd Images using Matlab GUI (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Text Line Detection in Camera Caputerd Images using Matlab GUI