IRJET- Handwritten Character Recognition using Artificial Neural Network
0 likes29 views
This document discusses a study on recognizing handwritten English characters using artificial neural networks. The researchers used a convolutional neural network (CNN) model that was trained on segmented handwritten character images. The CNN model classified the characters through multiple layers, including convolution, ReLU, pooling, and fully connected layers. This allowed the model to accurately recognize individual handwritten English characters. The researchers segmented the input images, extracted features, then classified and recognized the characters with the CNN model to identify handwritten words. The study aims to reduce manual work by automating the recognition of handwritten texts using deep learning techniques.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2841
Fig 1 Methodology of working
4.1 Pre-processing
Pre processing involves various stages that had to be
performed such as binarization coverting image pixels into
binary digits and next is noise filtering were were various
white spaces unwanted strokes are eliminated and
smoothing process takes place were contrastand brightness
of the image is enhanced.
4.2 Segmentation
In this stage the image is segmented into various sub
images so that the representation can simplify and change
the image in better enhanced way so that it can be analyzed
easily.
4.3 Feature extraction
When the input data that we provide for an
algorithm is large or redundant we need to reduce the
representation of thatimages,thisconversionoftransferring
input data to reduced form is known as feature extraction.
Feature extraction is type were in we extract different
geometric shapes, lines and textual features to enhance the
representation of the character. In this system we provide
certain data set which as particular limitation of featurethat
as to be considered such input image must be cropped and
no white spaces must be seen and handwritten should be
need and in cursive writing so that it iseasilyrecognizable by
users.
4.4 Classification and recognition
We are mainly using convolution neural networks for
classification and recognition process were we pass the
input data into different layers of convolution such as
convolution layer, relu layer, pooling layer, and fully
connected by passing these through each layers we can
easily classifify by representing the data in pixel form in
matrix representation and we can recognize the words.
5. CONCLUSION
Artificial neural network whichisclusterofnetworks
which helps in recognition of various texts. Here we have
seen the recognition of handwritten English characters by
using various methodological steps and by passing the data
through various laers of convolution layers we can classify
and recognize the desired character.
REFERENCES
[1] Handwritten text recognition using Deep learning by
Batuhan Balci, Dan Saadan, Dan Shiferaw.
[2] Offline Handwritten Character recognition using Neural
network by Anshul Gupta, Manisha Srivastava, Chitralekha
Mahanta IIt Guwahati, india
[3] Online Handwritten recognition problem: Issues and
Techniques by Vikas kumar computer science and
information technology department MIT, Moradabad, U.P ,
INDIA
Pre processing
Classification and
recognition
Feature
extraction
segmentation](https://image.slidesharecdn.com/irjet-v6i4605-190706095548/85/IRJET-Handwritten-Character-Recognition-using-Artificial-Neural-Network-2-320.jpg)
IRJET- Handwritten Character Recognition using Artificial Neural Network
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2840 HANDWRITTEN CHARACTER RECOGNITION USING ARTIFICIAL NEURAL NETWORK Archana1, Deepana. L2, Arun. G3, Lohith Raj. B4, Akshatha. M5 1,2,3,4Students Dept of Computer science and Engineering, Vidya Vikas Institute of Engineering and Technology College, Mysore, Karnataka, India 5Professor , Dept of Computer science and Engineering , Vidya Vikas Institute of Engineering and Technology College, Mysore, Karnataka, India ---------------------------------------------------------------------***---------------------------------------------------------------------- Abstract – Character recognition, due to wide application many research activities are been taking place in past and in present generation. In this paper we are dealing with handwritten English words by recognizing single characters. This recognition can be done by classifying into two classes holistic and segmentation. In holistic only limited size words are considered and images of different features can be recognized. In segmentation we employ bottom up approach where neural networks are used to identify segmented individual character and recognize the words. We have used convolution neural network that trains a model accurately and recognize words. In later work we have also used long term memory networks that uses convolution networktobuilt a bounding boxes for each character, the segemented character are then used by CNN for classification and then built a word on a basis of output that is obtained from classification and segmentation process. 1. INTRODUCTION It is a very interesting and challenging task to develop handwritten character recognition.Theglobal featureswere introduced to get better accuracy. Another module in this system of character recognition is pattern matching. These character recognition can be classified into four approaches of pattern recognition such as template matching, statistical techniques, structured techniques, neural networksbyusing these techniques it will be better accurate for recognition of handwritten character recognition. 1.1 Problem definition There are many Problems that have been seen to recognize handwritten character such as to built a profile of syntactic surface structure that helps to describe ambiguous words using geometric classification method by combining strokes and to use the advance nextgenerationsequencoung. There are another problems that have been seen that is recognizing offline verses online handwritten characterand then is handwritten segmentation, the reliable identifying online stroke cluster that forms an individual character or words can be difficult. 1.2 Objectives The objectives of handwritten character recognition that we have seen in real time is it has reduced manual work so that the old literatures can be converted into digitized for manually. Using neural signs in literature domain and proposed system serves as guide in recognizing words most accurately and efficiently then older system. 2. Exixting system The current working system deals with recognition of English printed or written text words by a system. In this process the image of text character that is individual character wors by word is scanned and analyzes is done for the scanned image. The next step is translating character image into character codes like as ASCII codeforfurtherdata processing. 3. Proposed system In the proposed systemwemainlysubdivideintotraining and testing phases. In training phase we pass a sample training mage as an input and then there are pre processed and segmentation of images into various categories and by categorizing into various features techniques such as geometric shapes and strokes and extracted and we classify the rained image in training phase. In testing phase firstlywe pass a input image and sementationprocesstakesplacewere we categorize inputs into different parts and we apply various future extraction process to modify the words to recognize it in a efficient way and we finally simulate the classifiedwords.webasicallyuseconvolutionneuralnetwork algorithm in classifying and recognizing the words were in wecomeacross variouslayersinconvolutionneuralnetwork they are convolution layer, relu layer, polling layer and fully connected layers. 4. Methodology The main working part of the system is discussed in this contest where in we comeacross variousstagestheyarepre- processing, segmentation, future extraction and finally classification and recognition
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2841 Fig 1 Methodology of working 4.1 Pre-processing Pre processing involves various stages that had to be performed such as binarization coverting image pixels into binary digits and next is noise filtering were were various white spaces unwanted strokes are eliminated and smoothing process takes place were contrastand brightness of the image is enhanced. 4.2 Segmentation In this stage the image is segmented into various sub images so that the representation can simplify and change the image in better enhanced way so that it can be analyzed easily. 4.3 Feature extraction When the input data that we provide for an algorithm is large or redundant we need to reduce the representation of thatimages,thisconversionoftransferring input data to reduced form is known as feature extraction. Feature extraction is type were in we extract different geometric shapes, lines and textual features to enhance the representation of the character. In this system we provide certain data set which as particular limitation of featurethat as to be considered such input image must be cropped and no white spaces must be seen and handwritten should be need and in cursive writing so that it iseasilyrecognizable by users. 4.4 Classification and recognition We are mainly using convolution neural networks for classification and recognition process were we pass the input data into different layers of convolution such as convolution layer, relu layer, pooling layer, and fully connected by passing these through each layers we can easily classifify by representing the data in pixel form in matrix representation and we can recognize the words. 5. CONCLUSION Artificial neural network whichisclusterofnetworks which helps in recognition of various texts. Here we have seen the recognition of handwritten English characters by using various methodological steps and by passing the data through various laers of convolution layers we can classify and recognize the desired character. REFERENCES [1] Handwritten text recognition using Deep learning by Batuhan Balci, Dan Saadan, Dan Shiferaw. [2] Offline Handwritten Character recognition using Neural network by Anshul Gupta, Manisha Srivastava, Chitralekha Mahanta IIt Guwahati, india [3] Online Handwritten recognition problem: Issues and Techniques by Vikas kumar computer science and information technology department MIT, Moradabad, U.P , INDIA Pre processing Classification and recognition Feature extraction segmentation