NIET Journal of Engineering & Technology (NIETJET)

Volume 5, Issue Winter 2016 ISSN: 2229-5828 (Print)

Handwritten Numbers and Characters

Recognition Using Machine Learning
Algorithms
Kumar K.1, Singh H.1 ,Khan S., Manu Sharma2
1
(Computer Science and Engineering, NIET Greater Noida, India)
2
(Computer Science and Engineering, NIET Greater Noida, India)

Abstract: One of the most well-known challenges in machine learning and computer vision applications is
handwritten digit recognition. To overcome the challenge of handwritten digit and/or letter recognition, a variety
of machine learning algorithms have been used. The current research focuses on using Neural Networks to
overcome the problem. Deep neural networks, deep belief networks, and convolutional neural networks are the
three most well-known methodologies. The accuracy and performance of these three neural network algorithms
are compared and assessed in terms of a variety of variables such as accuracy and performance. However, the rate
of recognition accuracy and performance is not the only factor in the assessment process; there are more relevant
metrics like execution time. Random and standard dataset of handwritten digit will be used for conducting the
experiments. The trials will be conducted using a random and standard dataset of handwritten digits. The results
demonstrate that Deep belief network is the most accurate algorithm among the three neural network techniques,
with a 98.08 percent accuracy rate. Deep belief network, on the other hand, has a comparable execution time to
the other two methods. Each method, on the other hand, has a 1-2 percent error rate due to digit form similarities,
particularly with the digits (1,7), (3,5), (3,8), (8,5), and (9,5). (6, 9).
Keywords - Handwritten recognition, CNN, deep learning ,KNN, machine learning,

1. INTRODUCTION
Throughout the decades, several scholars have worked on various methodologies and proposed various ways for pre-
processing, segmentation, recognition, and post-processing [1]. During the 1960s, many approaches such as labeling
schema for syntactic description of pictures, syntax directed interpretation of various classes of pictures, description
and generation based, and synthesis method were applied [2]. Even with all of the approaches like KNN, SVM, and
CNN, it may not be feasible to attain 100% accuracy in a pre-processing system [3]. The resemblance between digits
such as 1 and 7, 5 and 6, 3 and 8, 9 and 8, and so on was the basic challenge. It would meet in this digit categorization
task. People also write the same numeral in a variety of ways. Finally, the individuality and variation of various
people's handwriting has an impact on the digit's construction and look.

2. METHODOLOGY
The scope of this study is to study known machine learning algorithms for handwritten character recognition and
implement one of them in a user-friendly Android application. The software will offer a solution for handwriting
recognition using live camera frames or an image file, as well as learning new characters and learning interactively
depending on user feedback. Because of its great performance on nonlinearly separable issues, it is decided to use a
multilayer perceptron, a feed forward artificial neural network, as recognition model. It's also shown to be effective
in optical character recognition. The Python programming language has been used to analyze the perceptron's
performance and adjust its parameters, and then create the Android application using the same perceptron architecture,
learning parameters, and optimization techniques. The programme will be put through its paces on a training set of
numbers and alphabets, with the potential to learn alphabetical or non-alphabetic characters. Image capture, pre-

19 | Page
Publisher: Noida Institute of Engineering & Technology,
19, Knowledge Park-II, Institutional Area, Greater Noida (UP), India.
 NIET Journal of Engineering & Technology (NIETJET)
Volume 5, Issue Winter 2016 ISSN: 2229-5828 (Print)

processing, segmentation, feature extraction, classification, and post-processing are the six processes of handwritten
recognition. Figure 1 shows a block diagram of basic digit recognition.

Fig 1. Overall Methodology

2.1 Preprocessing

Pre-processing is the first step in digit recognition and is essential for a high recognition rate. The major goal of pre-
processing processes is to normalize strokes and eliminate variations that would otherwise make recognition more
difficult and diminish recognition rate. The irregular size of the text, missing points during pen movement collecting,
jitter present in text, left or right bend in handwriting, and unequal lengths of points from nearby places are all
examples of variances or distortions. Size normalization and centering, interpolating missing points, smoothing, slant
correction, and point resampling are the five typical procedures in pre-processing [2].
2.2 Segmentation

The individual digits of an image are separated during segmentation. In most cases, documents are processed in a
hierarchical order. The row histogram is used to segment lines at the first level. Words are retrieved from each row
using a column histogram, and afterwards digits are derived from words [2].
2.3 Feature Extraction

The major goal of the feature extraction phase is to obtain the most relevant pattern for classification. To extract the
characteristics of individual digits, feature extraction techniques such as Principle Component Analysis (PCA), Linear
Discriminant Analysis (LDA), Chain Code (CC), Scale Invariant Feature Extraction (SIFT), zoning, Gradient based
features, and Histogram can be used. These characteristics are employed in the system's training [3].

20 | Page
Publisher: Noida Institute of Engineering & Technology,
19, Knowledge Park-II, Institutional Area, Greater Noida (UP), India.
 NIET Journal of Engineering & Technology (NIETJET)
Volume 5, Issue Winter 2016 ISSN: 2229-5828 (Print)

2.4 Classification

When an image is provided to the HCR method, its characteristics are retrieved and fed to a trained classifier such as
an artificial neural network or a support vector machine (SVM). Classifiers compare an input feature to a previously
stored pattern to determine the best input class matches [3].
2.5 Post-processing

Post-processing is the process of using linguistic knowledge to fix misclassified data. The result of shape recognition
is processed in post processing. The accuracy of pure shape identification can be improved by using language
information. Some form recognizers provide a single string of digits from handwritten input, while others produce a
number of choices for each digit, frequently with a level of confidence for each possibility.

3. RESULTS AND DISCUSSIONS

One of the most practical difficulties in pattern recognition applications is handwritten character recognition. Digit
recognition is used in postal mail sorting, bank check processing, form data entry, and other applications. The capacity
to design an efficient algorithm that can detect handwritten digits and which is provided by users via a scanner, tablet,
and other digital devices is at the heart of the challenge. The following are some examples of applications:
• The total amount of bank checks written
• On Postal Cards and Envelopes, there is a postal code.

21 | Page
Publisher: Noida Institute of Engineering & Technology,
19, Knowledge Park-II, Institutional Area, Greater Noida (UP), India.
 NIET Journal of Engineering & Technology (NIETJET)
Volume 5, Issue Winter 2016 ISSN: 2229-5828 (Print)

5. CONCLUSION
Handwritten digit recognition using a neural network has shown to be pretty effective. The approach of using extracted
characteristics from sample photos of each digit and character to train a neural network offers a higher detection
accuracy. The system's total efficiency is 87.4%. The first constraint is due to the robustness of the classifier. Even if
the segmentation is done well, the classifier may not be able to recognise the segmented fingers due to their form. To
overcome this constraint, it is suggested that this digit be added to the training set in order to improve the performance
of SVM, ANN, and CNN classifiers.

REFERENCES
[1.] M Ozaki, Y. Adachi, Y. Iwahori, and N. Ishii, Application of fuzzy theory to writer recognition of Chinese
characters, International Journal of Modelling and Simulation, 18(2), 1998, 112-116

[2.] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, S. Huang, A. Karpathy, A. Khosla,M. Bernstein,
A.C. Berg, and F.F. Li. “ Imagenet large scale visual recognition challenge”. International Journal of Computer
Vision” December 2015, Volume 115, Issue 3, pp 211- 252.

[3.] Liu, C.L. et al.,.” Handwritten digit recognition: Benchmarking of state-of-the-art techniques”. Pattern
Recognition 36, 2271–2285. 2003 17.

[4.] Xiao-Xiao Niu n , Ching Y. Suen “A novel hybrid CNN–SVM classifier for recognizing handwritten digits”
Pattern Recognition 45 (2012) 1318–1325, 2011 19.

[5.] Samidha Garg, S.N.Singh; 'Performance Comparison Of High Frequency Single Phase PWM Inverters Using
PSpice',Volume No.1,Issue No.1,2013,PP.001-008,ISSN :2229-5828

[6.] Satyendra Sharma, Dr.Brahmjit Singh; 'Planning and Optimization of UMTS Network Radio Link with
Limited Interference',Volume No.2,Issue No.1,2013,PP.009-012,ISSN :2229-5828
[7.] L.NavinkumarRao, S Gairola; 'Steady -State and Dynamic Performance of the Static Var Compensator
(SVC) Phasor Model',Volume No.2,Issue No.1,2013,PP.013-016,ISSN :2229-5828
[8.] Rishi Singhal, S.P. Singh and Dr V. K. Pandey; 'Design of Phase Compen-sation Circuits For Multistage OP-
AMP',Volume No.2,Issue No.1,2013,PP.017-019,ISSN :2229-5828
[9.] Mohd Yunus Khan, Saroj Kumar Padhi and R.K. Singh; 'Extraction ,Characterization And Engine
Performance of Jatropha Curcas Oil',Volume No.2,Issue No.1,2013,PP.021-028,ISSN :2229-5828
[10.] Srivastav, Vivek; 'Analytical Study of Traffic Behaviour in a Network on Chip Architecture',Volume
No.2,Issue No.1,2013,PP.029-032,ISSN :2229-5828

22 | Page
Publisher: Noida Institute of Engineering & Technology,
19, Knowledge Park-II, Institutional Area, Greater Noida (UP), India.