IRJET- Advanced Character based Recognition and Phone Handling for Blind Peoples

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1751
Advanced character based recognition and Phone handling for
Blind peoples
R.Aarthi1,N.R. Kruthika Devi2, S. Prithi3
1,2 B.E/ ECE, Prince Shri Venkateshwara Padmavathy Engineering College
3Asst.Prof /ECE, Prince Shri Venkateshwara Padmavathy Engineering College
------------------------------------------------------------------------***-------------------------------------------------------------------------
Abstract—Blindness makes life rather difficult for people
who suffer from this health problem, but the use of
technology can help them in day-to- day tasks. This paper
describes an embedded device dedicated for blind or visually
impaired people. Technology poses a challenge for blind
people as well. For example, a blind person cannot read the
information on a web page. Searching the internet requires
screen reading software will read the information on a
website, but this may require a significant amount of time
to learn the process. In this context, the present work
focuses the development of a text-to- speech application
for the blind in both Tamil and English font. In this project
raspberry pi as the main unit which has an USB camera that
is used to scan any written document and uses Optical
character recognition (OCR) to convert the image into a
digital text. We then use a text to audio system that will
enable us to convert the digital text into a synthesized
voice. In this project, we develop a system in which the text
written are been analyzed and Optical Character
Recognition (OCR) is been performed. For feature
extraction, Gabor filter algorithm is been proposed. To
classify the input with the trained dataset, Support
Vector Machine (SVM) is used. Additionally phone calls are
established to the frequently contacted number just by
their actions using gyroscope sensor. So it is an attempt
to make the life of blind people quite independent.
Key Words - text detection, optical character recognition
and speech synthesis, gyroscope sensor, GSM.
I. INTRODUCTION
With increase in growing population, the number of
blind people across the world is set to triple from about 36
million to 115 million by 2050. The uniqueness in
characters were identified on physiological characteristics
and behavioral characteristics. Uniqueness in signature,
handwriting and voice comes under behavioral
characteristics while face, iris, retina impression falls
within physiological characteristics. We know that hand
writing differs from person to person and even with
respect to time which the person write. Hence a well-
defined trained algorithm must be implemented to
recognize the hand written documents. The efficiency of
the system is 80-90% in order to recognize the slightly
varied hand written documents. The document which has
to be converted are captured as a image using USB camera
connected with raspberry pi controller. Further the
captured image are converted into binary image followed
by line segmentation and word segmentation using
horizontal and vertical profile components. For English
font tesseract algorithm were inserted to identify the
English characters and words up to 200 different types of
fonts. And Tamil documents are recognized by separating
number of horizontal, vertical and curved region present
with in the character. A phone call is established through
GSM just by detection their actions with gyroscope sensor
so that it helps the blind people at emergency time. Were
the detected actions were fed into ADC converter and the
digital outputs were given to the GSM module. The process
architecture describes the overall process involved in the
device such as scanning, preprocessing, segmentation,
feature extraction, classification and text detection as well
as call establishment steps followed by description about
the obtained results and conclusion.
II. PROCESS ARCHITECTURE
The recognition of both printed documents and hand
written documents recognition involves in process such as
Scanning, Preprocessing, Segmentation, Feature
Extraction, self- organizing, Map classification and
Recognition which is shown in the figure 1.And calls are
established through GSM module by sensing the actions
with gyroscope sensors.
Figure 1: System Architecture

2.1 SCANNING
The documents are scanned using USB camera which is
connected with one of the four ports of raspberry pi
controller. The scanned documents are also viewed in
personal computers. There would be a change in value of
the pixels values at every steps of scanning which are
termed as noise. The scanned images are stored in
document is sent to a program that saves in TIF, JPG or GIF
format.
2.2 PREPROCESSING
Preprocessing is the step involved after scanning
process which consist of three steps namely, binarization,
Noise removal and Skew correction. There are two peak
values where high peak corresponds to the white
background and a smaller peak corresponds to the
foreground. The binarized image is preprocessed for noise
removal which occurs due to the poor quality of the
document or accumulated while scanning that has to be
removed before further processing. The resultant image
is checked for skewing. The image can either skewed with
left or right orientation. The images are brightened, then
angle of orientation between ±15 degrees are checked and
if detected then image rotation is carried out until the lines
match with the true horizontal axis, which produces a
skew corrected image as given in figure 2.
Fig. : 2 Histograms for skewed and skew corrected
images
2.3 SEGMENTATION
The pre-processed image which is free from noise and
skew corrected image is passed to the segmentation
phase, where the image is decomposed into individual
characters. Text lines can be identified with the help of the
horizontal projection profile. Projection profile of a
document in a particular direction is the running sum of
the pixels along that direction. The profile exhibits valley
points at line boundaries and the location of these minima
points mark the line boundaries as shown in the figure 3.
For binary images, these are the points where the profile
goes to zero.
Fig. : 3 Line segmentation
Word segmentation can be performed with the aid
of vertical projection profile. The vertical projection profile
shows valleys at points corresponding to word gaps as
shown in the figure 4. These word boundaries can be
identified with the help of these minima points.
Fig. : 4 Word segmentation
Fig. : 5 Character segmentation
2.4 FEATURE EXTRACTION
The next phase to segmentation is feature
extraction where each character is represented as a
feature vector, which becomes its identity. Feature
extraction forms the backbone of the recognition process.
The major goal of feature extraction is to extract a set of
features, such as height, width of the character, number of
short and long horizontal lines present, number of short
and long vertical lines present, number of circles present,
number of horizontally and vertically oriented arcs,
centroid of the image and pixels in the various regions of
the character which maximizes the recognition rate.
Length of the character is computed by subtracting the
column number of the last pixel to the column number of
the first pixel. Height of the row is calculated by
subtracting the row number of the last pixel to the row
number of the first pixel. Area is the total number of pixels
in the region. It is calculated by making the product of
height and length. The shape of the character sometimes
may be circular or curved forming a loop. Junction is the
interface between two regions in the characters depending
upon length and height of the pixels.
2.5 KOHONEN’S SELF ORGANIZING FEATURE MAP
The process of classification of documents was carried out
in three phases. The first phase is document preprocessing.
The second phase is the training process. The third phase is
The character segmentation are performed by
detecting the pixels of the character and are segmented as
shown in the figure 5.

the test phase in which a document is classified and the
weights of neighboring units are updated.
Kohonen’s SOFMs are a type of unsupervised learning. The
goal is to discover some underlying structure of the data.
With this approach an input vector is presented to the
network and the output is compared with the target vector.
If they differ, the weights of the network are altered
slightly to reduce the error in the output. This is repeated
many times and with many sets of vector pairs until the
network gives the desired output. The network is created
from a 2D lattice of 'nodes', each of which is fully
connected to the input layer. Fig.2.5.1 shows a very small
Kohonen’s network of 4 X 4 nodes connected to the input
layer representing a two dimensional vector. All neurons
in the output layer are well connected to adjacent neurons
by a neighborhood relation depicting the structure of the
map. Generally the output layer can be arranged in
rectangular or hexagonal lattice.
Fig. : 6 Kohonen’s network of 4 X 4 nodes
Lots of activities in pre-processing stages helps to process
this stage very easy. Self-organizing feature maps (SOFM)
are unsupervised machine learning that learns by self-
organizing and competition. The main idea for this is to
make it simple and acceptable for Kohonen SOM. It reduces
a remarkable amount of time. SOM is clustering the input
vector by calculating neuron weight vector according to
some measure (e.g. Euclidean distance), thus weight vector
that closet to input vector comes out as winning neuron.
However, instead of updating only the winning neuron, all
neurons within a certain neighborhood of the winning
neuron are updated using the Kohonen rule [20]. The
algorithm is described as follows, suppose the training set
has sample vectors X, trains the SOM network has
following steps:
Step 1: Node’s weights are initializing weights for each
nodes.
Step 2: From the set of training data a vector is chosen at
random and presented to the lattice.
Step 3: Every node is examined with the input node and
whichever node most likely matches with the input vector
is termed as the Best Matching Unit (BMU).
Step 4: The radius of the neighborhood of the BMU is now
calculated whose value starts large, typically set to the
‘radius’ of the lattice but diminishes each time-step. Any
nodes found within this radius are deemed to be inside the
BMU’s neighborhood.
Step 5: The neighboring node’s weights are adjusted in
such a way it’s equal to the input vector. The node which is
closer to the BMU has to alter its weight more.
Step 6: For N iterations repeat step 2.
i) Firstly, all neuron nodes weights, defined as W j (1), j =
1…L, are initialized randomly. L is the number of neurons
in the output layer.
ii) K =Maximum (X,(k)), for iteration step k=1...K, get an
input vector X(k) randomly or in order.
iii) Calculate Distance = X (k), k = 1…n, 1…n refers to
neuron nodes.
iv) Select the winner output neuron j * with minimum
distance.
v) Update weights Wj(k+1) to neurons j * and its
neighborhood
W j (k+1) = Wj(k) + [ (k+1) ∩ (j, j*(k+1),(k+1)][X(k+1) W j
(k)], j=1…..L
vi)If k= K go to step (ii).
2.6 TEXT TO SPEECH
The scope of this module is initiated with the conclusion
of the receding module of Character Recognition. The
module performs the task of conversion of the transformed
Tamil text to audible form. The Raspberry Pi has an on-
board audio jack, the on-board audio is generated by a
PWM output and is minimally filtered. A USB audio card
can greatly improve the sound quality and volume. Two
options of attaching a microphone into Raspberry Pi. One is
to have USB mic, another to have an external USB sound
card.
2.7 CALL HANDLING PROCESS
The actions from the blind people were detected with
the help of gyroscope sensor with respect to changes in the
XYZ axis direction. Hence the output of the gyroscope
sensor is of analog form. In order to convert them into
digital signal which is suitable for further computations
we use ADC converter. The output digital signal from ADC
block is then fed into Raspberry pi controller which sends
them to the GSM module that has an inserted sim in its sim

slot. With respect to the python code the calls are
established to the corresponding person. The analog signal
are viewed in the pc.
III. SUPPORT VECTOR MACHINE
Supervised learning technique support vector machine is
used for pattern classification. The binary classifier is built
by standard SVM by constructing a hyper plane which
separates to classes of data. The subset of informative points
called support vectors are identified by SVM automatically
which uses them to represent the separating hyper plane.
With a set of training samples, (xi,yi) the machine are
accessible where the xi are the real world data instances
and the yi are the labels signifying which class the instance be
appropriate to. When two class pattern recognition
delinquent, yi = +1 or yi = -1. A training example (xi,yi) are
called positive if yi = +1 and negative otherwise. Hyper plane
that separates two classes are constructed by SVM which
tries to achieve all-out parting between the classes.
Extrication the classes with a huge margin minimizes a
bound on the expected generalization error.
IV. FINDINGS OF THE PROPOSED SYSTEM
To test the effect handwriting style has on character
recognition with this system, samples from the five
persons were scanned using USB camera. To test in an
environment where 100% accuracy was obtainable, only
the first 8 letters of each sample were used. This also
reduced the amount of time and processing power needed
to run the experiment. Each time in a slightly different
position as shown in figure 7 and its accuracy where
calculated. Letters from the sentence in the handwriting
sample were used to create the test set to determine
accuracy. Similarly English fonts were also implemented
for handwritten English character recognition (ie for
word) using Support vector machine.
Figure: 7 Scanned handwriting
V. CONCLUSION
We investigated a new representation of Tamil Character
Recognition, and used Kohonen SOM techniques efficiently
classifies handwritten and also Printed Tamil characters.
More effective and efficient feature detection techniques
will make the system more powerful. There are still some
more problems in recognition. They are, during letter
segmentations and abnormally written characters.
Misrecognition could be avoided by using a word
dictionary to look-up for possible character composition.
The presence of contextual knowledge will help to
eliminate the ambiguity. We show that, in practice, the
proposed approach produces near optimal results besides
outperforming the other methodologies in existence. Our
future work in this regard will be analyzing the features of
joined letters and incorporating better segmentation
accuracy. Results indicate that the approach can be used
for character recognition in other Indic scripts as well. The
call handling system helps the blind people to make out the
calls easily without anyone’s help. On the whole this
system is used to establish the phone calls and recognize
both English and Tamil font with different size and hand
written documents too.
REFERENCES
1. Majid Sepahvand, Fardin Abdali-Mohammadi,
Farhad Mardukhi,” Evolutionary Metric-Learning-
Based Recognition Algorithm for Online Isolated
Persian/Arabic Characters, Reconstructed Using
Inertial Pen Signals ” , 2016.
2. Siddhesh R.Baravkar, Mohit R.Borde, Mahendra K.
Nivangune,” Android text messaging application
for visually impaired people”,2013.
3. Ms. Pratiksha More, Ms. Ketaki Tharkude, Ms.
Sonali Kothey, Ms.Aishwarya
Wayase,Prof.Kanchan Varpe,” Smart Android
Application for blind people based on object
Detection”, 2015.
4. K.H.Aparna, Vidhya Subramanian, M. Kasirajan, G.
Vijay Prakash, V.S. Chakravarthy,” Online
Handwriting Recognition for Tamil”, 2015.
5. Thendral T, Vijaya MS, Karpagavalli S,” Analysis of
Tamil Character Writings and Identification of
Writer Using Support Vector Machine”, 2014.
6. Akshay Apte and Harshad Gado, “Tamil character
recognition using structural features” ,2010
7. Kanimozhi.V.M, Muthumani.I, “Optical Character
Recognition for English and Tamil Script”,
International Journal of Computer Science and
Information Technologies, 2014.

8. G. Siromony, R. Chandrasekaran, M.
Chandrasekaran, “Computer Recognition Of
Printed Tamil Characters”, Pattern Recognition
(1978) 243-247.
9. S.T.Deepa, Dr.S.P.Victor (2012), “Tamil Text
Extraction “, International Journal of Engineering
Science and Technology, Vol. 4 No.05 May 2012,
pp 2176 – 2179.
10. T.Rubesh Kumar, C.Purnima, “Assistive System for
Product Label Detection with Voice Output For
Blind Users “, IJREAT International Journal of
Research in Engineering & Advanced Technology,
Volume 1, Issue 6, Dec-Jan, 2014, pp 1- 4.
11. Aparna K G and A G Ramakrishnan , “A Complete
Tamil Optical Character Recognition System”,
12. Naveen Sankaran and C.V Jawahar, “Recognition of
Printed Devanagari Text Using BLSTM Neural
Network”, IEEE, 2012

IRJET- Advanced Character based Recognition and Phone Handling for Blind Peoples

More Related Content

What's hot (19)

Similar to IRJET- Advanced Character based Recognition and Phone Handling for Blind Peoples (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Advanced Character based Recognition and Phone Handling for Blind Peoples