International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064 Index Copernicus Value (2016): 79.57 | Impact Factor (2017): 7.296 Volume 7 Issue 4, April 2018 www.ijsr.net Licensed Under Creative Commons Attribution CC BY Deep Learning based English Handwritten Character Recognition Somil Gupta 1 , Tanvi Bansal 2 , Manish Kumar Sharma 3 1, 2 Final Year Student, Department of CSE, Galgotia College of Engineering and Technology, Greater Noida, India 3 Assistant Professor, Department of CSE, Galgotias College of Engineering and Technology, Greater Noida, India Abstract: In this paper we propose a convolutional neural network based handwritten character recognitionusing SVM as an end clas- sifier. The learning model is based on CNN (Convolutional Neural Network) which is used as a feature extraction tool. The proposed method is more efficient than other existing methods and will be more efficient than modifying the CNN with complex architecture. The recognition rate achieved by the proposed method is 93.3% which is greater than other existing methods. The computation time of train- ing phase is 13.14sec and that of testing phase is 13.27 sec. The proposed system was validated by 6 validation points. The overall accu- racy of system is 93% Keywords: Deep Learning, Convolutional Neural Network, Handwriting recognition, SVM 1. Introduction Handwriting recognition is a very important field of re- search. Both the Machine leraning and computer vision fields have research areas regarding handwriting recogni- tion. [1] A number of different algorithms and techniques have been provided, but still it is an unresolved area. With various different writing styles handwriting recognition is still very complex to implement. Some problems in handwriting recognition are due to uncer- tainty of inut data, as characters of different persons are dif- ferent, some [2] are disconnected and distorted, gthe thick- ness of characters varies. Handwritten character recognition is an area of pattern rec- ognition which defines an ability of a machine to analyze patterns and identify the character. Pattern recognition is the science of making inferences from perceptual data based on either a priori knowledge or on statistical information [1] [2]. The subject of pattern recogni- tion spans a number of scientific disciplines uniting them in the search for the solution to the common problem of recog- nizing the members of a class in a set containing elements from many patterns in classes. A pattern class is a category determined by some given common attributes. We have used Covolutional Neural Network algorithms for feature extraction. For the classifier we have used the SVM. 2. Related Work Some researches have been conducted to develop a variety of methods and algorithms that can be used to recognize a handwritten character Azmi [7] have used Freeman Chain Code with the division of the region into nine regions and normalization of chain code as feature extractor. He also explains four features con- sisting of top, right, wide, and high-ratio of charac- ters.Artificial Neural Network (ANN) is used as a classifier. In paper [8] the author uses Zernike Moments to extract the features of characetrs and uses SVM for classification. In [9] Nasienproposes Freeman Chain Code to remove a feature of uppercase characters. Hallale [10] proposed a 12- directional method for feature extraction ofEnglish charac- ters. The data are classified based on the similarity between the feature of data training and the feature of data testing. 3. Proposed Method A. CNN (Convolutional Neural Network) The learning model is based on Convolutional Neural Net- work (CNN) as a powerful feature extraction. The learning model was used to construct a handwriting recognition sys- tem to recognize a more challenging data on form document automatically [3]. The pre-processing, segmentation and character recognition are integrated into one system. The output of the system is converted into an editable text. [4] Convolutional Neural Network (CNN) is one of the deep learning architecture. [4] It can extract multiple features from low-features to high-features automatically. Currently, CNN is a state of the art of handwriting characters recogni- tion Convolutional Neural Networks are very similar to or- dinary Neural Networks, they are made up of neurons that have learnable weights and biases. Each neuron receives some inputs a dot product and optionally follows it with a non-linearity. Figure 1: Architecture of CNN for feature extraction B. SVM (Support Vector Machine) SVM first introduced by Vapnik. It widely used for classifi- cation and regression. The basic idea of SVM is finding the Paper ID: ART20181693 DOI: 10.21275/ART20181693 1402