International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 05 Issue: 12 | Dec 2018 www.irjet.net p-ISSN: 2395-0072

Text Recognization of Product for Blind Person using MATLAB

Rupali Deshmukh1, Prof. Manoj Kathane2, Prof. Yogesh Sushir3
1Student of M.E. E&TC of Dr. V. B. Kolte College of Engg. Malkapur
2,3Professor, E&TC of Dr. V. B. Kolte College of Engg. Malkapur
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – There is printed text everywhere around us and system cannot read the text from the complex background.
we see it in our day to day life. Like product names, restaurant Reading is essential for every human being. Printed text is
menus, instructions on medicines etc. But the question arises everywhere in the form of reports, receipts, bank statements,
how Visually Impaired or blind people can recognize this text. restaurant menus, product packages, medicine bottles etc.
Thus surely they need some assistance to read the text. In this can help blind users and those with low vision to read text,
project I tried to propose a camera-based assistive text there are few devices that can provide good access to
reading framework to help low visual power or blind person to common hand-held objects such as product packages, and
read text and product label or document from hand-held object sprinted with text such as prescription medication
objects. By using MATLAB coding and camera it is tried to help bottles. The ability of people who are blind or have
blind persons to read information of the products. In this significant visual impairments to read printed labels and
project camera acts as main vision in detecting the label product packages will enhance independent living and foster
image of the product then image is processed internally and economic and social self-sufficiency. Image processing is
separates label from image by using MATLAB program and processing of images using mathematical operations in any
finally identifies the product name and information the optical form of signal processing for which the input is an image.
character recognition. The output of image processing may be either an image or a
set of characteristics or the parameters which is related to
the image. Most image-processing techniques which involve
Key Words: Text reorganisation, camera-based text
treating the image as a two-dimensional signal and applying
assistance, MATLAB algorithm for RGB
the standard signal-processing techniques to the input.
Image processing is usually refers to digital image
1. INTRODUCTION processing, optical and analog image processing also are
possible. The acquisition of images is referred to as imaging.
Recent developments in computer systems, digital cameras, The close related to image processing is computer graphics
and different software like MATLAB make it feasible to assist and computer vision. In computer graphics, images are
low visibility individuals by developing camera-based manually made from physical models of objects, surrounding
products that combine computer vision technology with and lighting, instead of being acquired from natural scenes,
other existing commercial products such optical character and also in most animated movies. Computer vision, on the
recognition (OCR) systems. Million people are visually other hand is often considered a high-level image processing.
impaired worldwide, near about 39 millions are blind. Even In modern sciences and technologies, images also get much
in developing country like India, in 2015 Blind people broader scopes due to the ever growing importance of
Association survey reported that a 12 million people are scientific visualization .The millions of visually impaired
blind. Using system like video magnifiers, screen readers people in worldwide are still blind.
help blind person and those with low vision to access the
documents and text. The ability of people who are blind or 1.1 Problem Review
have low visual impairments to read printed labels and
documents will enhance independent living and social self- For visually Impaired or blind people it is very important
sufficiency. Today, there are many systems that have to be an independent. For that it is important to provide some
promise to portable use, but they cannot provide the product assistance them in reading. So, I tried to propose a camera-
labeling. Such systems are bar code reader which helps to based assistive text reading framework to help low visual
blind person to identify the different product. Database can power or blind person to read text and product label or
gives the permission to access the information for blind document from hand-held objects.
persons about the product through speech .But there is big
limitation for blind person to find the position of bar code on 1.2 Objective
the product. Some assistive systems like pane scanner which
is used in some situations. Such systems integrated with OCR Our Objective is to develop a system for visually challenged
software having functions to scanning and recognition of the person who can help them to identify the various products
text and have integrated voice output. These systems and give some more information about the product. To fulfill
generally design to read the text from simple backgrounds, this objective some sub objectives were formed which are as
standard fonts and also small range of fonts. Some systems following.
need only white background for scanning the text. This

© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1048
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 12 | Dec 2018 www.irjet.net p-ISSN: 2395-0072

i) Identify the common deficiency in most of the character 2.2 Conversion of RGB to Gray Scale
recognition software/tools by calculating the recognition rate
of each character and digit and find out the characters and
To make the system more simple i.e. work for noisy
digits whose recognition rate is very less.
conditions or complicated background, image pre-processing
methods like noise filtering are applied. The processing time
ii) Designing and development of the model to eliminate the of the overall process is long, so to reduce this process timing
common deficiency identified. the input image is converted from RGB to gray scale. This
preprocessing of images in this paper is a technique to
iii) Develop the algorithm to implement the above model. improve the quality of images. The main purpose of this
conversion is to enhance and extracts useful information
iv) Testing and Performance evaluation by analyzing results from the image. Two preprocessing tasks, thresholding and
of model noise removal, are performed here.

1.3 CONTRIBUTION 2.3 Text Binarization

The algorithm used previously cannot handle complex There are numbers of methods for binarization in document
background and multiple patterns, and extract text analysis but few in text analysis. In this paper, we reviewed
information from hand-held objects. In assistive reading text analysis binarization methods related. Thresholding
systems for blind persons, it is very challenging for users to techniques are quite popular in document analysis. Several
position the region of interest within the center of the improvements over thresholding techniques are also
camera’s view. As of now, there are still no acceptable proposed recently in document analysis and people try the
solutions. same methods to extend for scene text binarization also.

In this project the previous drawback of algorithm can be

minimized and divided the problem in stages. To make sure 2.4 Filter image
the hand-held object appears in the camera view, a camera
with sufficiently wide angle to accommodate users with only It is nothing but the image processing, since the methods take
approximate aim. This may often result in other text objects an input image and create another image as output. Other
appearing in the camera’s view. To extract the hand-held appropriate terms often used are ltering, enhancement, or
object from the camera image, a motion-based method to conditioning. The major notion is that the image contains
obtain a region of interest of the object is used. some signal or structure, which we want to extract, along
with uninteresting or unwanted variation, which we want to
It is a challenging problem to automatically localize objects suppress. If decisions are made about the image, they are
and text ROIs from captured images with complex made at the level of a single pixel or its local neighborhood.
backgrounds, because text in captured images is most likely
surrounded by various background outlier “noise,” and text 2.5 Automatic Text Extraction
characters usually appear in multiple scales, fonts, and colors.
For the text orientations, algorithm used in the previous
paper assumes that text strings in scene images keep There are two popular methods for extracting a text region in
approximately horizontal alignment but that drawback of images are, merge and split method and comparison of two
algorithm will overcome by algorithm which is best suitable. frames. Thus, they take long computing time due to the use of
Many algorithms have been developed for localization of text a whole image. So automatic text extraction algorithm is
regions in scene images. So we will be working on vertical implemented to detect the region containing the label text. In
character recognition problem and try to solve the problem order to handle complex backgrounds, two novel feature
by using Image Processing. maps to extracts text features based on stroke orientations
and edge distributions, respectively are used. Maximally
2. THEORETICAL BACKGROUND stable external region is used in automatic text extraction.

2.1 Image Input 2.6 Optical Character Recognition

The Image or we can say printed text/label is captured by Text recognition is performed by off-the-shelf OCR prior to
camera which is used in project. Initially this image output of informative words from the localized text regions. A
containing noise in background. This complicated text region labels the minimum rectangular area for the
background can be removed by stroke width transform accommodation of characters inside it, so the border of the
algorithm that helps to recognize the character by their shape text region contacts the edge boundary of the text character.
and width by calculating each pixel by their start to end point. However, OCR generates better performance if text regions
are first assigned proper margin areas and binarized to
segment text characters from background. We propose to use

© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1049
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 12 | Dec 2018 www.irjet.net p-ISSN: 2395-0072

Template matching algorithm for OCR. The output of the OCR 4. Result and Discussion
is nothing but a text file containing the product label (its
name) in textual form. Audio output component is to inform
the blind user of recognize text code in the form of speech or
Audio.
3. Algorithm used for Image to Audio Conversion

3.1 Text Detection Algorithm

We describe the text detection algorithm that is MSER

(Maximally Stable Extremal Region) algorithm. MSER is a
method for text detection, blob detection in images. The Fig. OCR based product name Recognition System
MSER algorithm extracts number of co-variant regions from
image. We rst de ne the concept of stroke and then explain
the stroke width transformation. MSER is based on the idea of Here, two images have been taken one is of Dove and one of
taking regions which stay nearly the same through a wide Battery. Toggle button 1 this button is used for Character
range of thresholds. All the pixels above or equal to a given recognition and Character Identification.
threshold are black and all the pixels below a given threshold
are white. MSER uses two important properties to remove
non text regions from image rst is Geometric Properties and
another is Stroke Width Variation Properties.

3.2 Geometric properties

MSER detects almost all text regions from image but

alongside it also detect some non-text regions. To remove
those non text regions rst we apply Geometric Properties on
image. Geometric Properties detects the non-text regions
from image and remove those regions. Fig. MSER Region

Fig. Flow chart of the Proposed Method Fig. Region and stroke width image

3.3 Stroke Width Transform As shown in the Result figure the toggle button 1 Proceed
Character Recognition by using OCR. This image of result
shows the region After Removing Non-Text Regions Based on
It receives the RGB image with the help of algorithm
Geometric Properties OCR is the stand of optical character
the image is converted into grey but of the same size after
recognition which is field of computer science that
that text can be marked from the region of interest. It has
recognizing image-based text from photos and transforms it
three important stages: first the most important stage is
to real digital character. OCR works like human ability in the
stroke width transform, then collection of pixel of images on
brain to recognize the letters, numbers and symbols. OCR can
their stroke width, then pigeonholing letter candidates into
read both handwritten and printed text. The performance of
regions of text. In Stroke Width Transform, the stroke in
OCR is directly related to quality of input documents and
image is converted into constant width with the help of
pictures.
continuous band. Figure shows example of stroke in image.
The Stroke Width Transform is operation for calculating
width of pixel stroke from image.
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1050
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 12 | Dec 2018 www.irjet.net p-ISSN: 2395-0072

 Such system can be implement to read the text from

the object and translate output in different
languages.
 Future work can also extend for localization
algorithm to process text strings with characters
fewer than three and to design more robust block
patterns for text feature extraction.
 We will also extend our algorithm to handle incline
text strings. Furthermore, we will address the
significant human interface issues associated with
Fig. Expanded Bounding Boxes Text reading text by blind users.
In this it will find out all the Region of Interest. And it will ACKNOWLEDGEMENT
identify and find out its Stroke Width of Image. Stroke Width
is a measure of the curves and lines that make up character.
I offer sincere and heartily thank, with deep sense of
Text regions have little stroke width variation, whereas non
gratitude to our guide Prof. Monoj Kathane and co guide Prof.
text regions have larger variations. To remove the non-text
Yogesh Sushir for their valuable guidance, direction and
regions using stroke width we require thresholds. All the
inspiration to my project work without taking care of their
pixels above or equal to a given threshold are black and all
voluminous work. I am also thankful to all teacher for taking
the pixels below a given threshold are white.
personal Interest, giving encouragement and timely
suggestion and notable guidance.
5. Summary and Conclusion
REFERENCES
In this project, we proposed algorithm for solving the
problem of character recognition for blind persons. We had [1] Chucai Yi, Student Member, Ieee, Yingli Tian, Senior
given the input in the form of images of a product. The Member, Ieee, And Aries Arditi “Portable Camera-Based
algorithm was trained on the training data that was initially Assistive Text And Product Label Reading From Hand-Held
present in the database. We have done pre-processing and Objects For Blind Persons”, Ieee/Asme Transactions On
OCR classification and detect the text. The project presents a Mechatronics, Vol. 19, No. 3, June 2014.
brief survey of the applications in various fields along with
experimentation into few selected fields. The proposed [2] Nagarathna, 2Sowjanya V M, “Product Label Reading
method is extremely efficient to extract all kinds of products System For Visually Challenged People”, International Journal
which have text including blur and illumination. The project Of Computer Science And Information Technology Research
includes two different methods of product identification Issn 2348-120x (Online) Vol. 3, Issue 2, Pp: (897-904),
which are successfully implemented. Month: April - June 2015.
The text-image recognition initiatives discussed in the [3] Ms Komal Mohan Kalbhor1, Mr Kale S.D.2, “ A Survey On
preceding sections illustrate research themes at OCR which Portable Camera-Based Assistive Text And Product Label
expand and redefine the role of recognition technology in Reading From Hand-Held Objects For Blind Persons”,
document-oriented applications. These include the International Research Journal Of Engineering And
development of editors which operate directly on scanned Technology (Irjet), Volume: 04 Issue: 03 Mar -2017
image data, and the use of text-image recognition to retrieve
text information from documents with content. Key concepts
[4] Priyanka Patil1, Sonali Solat2, Shital Hake Prof.S.T.Khot4,
embodied in these research efforts include partial document
“Camera Based Product Information Reading For Blind
models, task-oriented document recognition, user
People”, International Journal Of Engineering And Computer
specification and interpretation of recognition models, and
Science Issn:2319-7242 Volume 4 Issue 3 March 2015, Page
automatic generation of recognizers from declarative models.
No. 11072-11075
These concepts enable the realization of a broad range of
signal-based document processing operations, including font,
vocabulary and language-independent classification and text
detection.

6. Future scope

 future work of this project is to recognize the text

from the more complex background and different
type of challenging background surfaces

© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1051