=Paper=
{{Paper
|id=Vol-3283/Paper22
|storemode=property
|title=An Approach for Handwritten Recognition Using Bayesian Network
|pdfUrl=https://ceur-ws.org/Vol-3283/Paper99.pdf
|volume=Vol-3283
|authors=Rohini A,Richa Choudhary,Tanupriya Choudhury,Sachi Nandan Mohanty,Hitesh Kumar Sharma
|dblpUrl=https://dblp.org/rec/conf/isic2/ACMS22
}}
==An Approach for Handwritten Recognition Using Bayesian Network==
https://ceur-ws.org/Vol-3283/Paper99.pdf
An Approach for Hand Written Recognition using Bayesian
Network
Rohini A1, Richa Choudhary2, Tanupriya Choudhury3, Sachi N. Mohanty4,
Hitesh K. Sharma5
1
Anil Neerukonda Institute of Technology and Sciences, Vishakapatnam, AndhraPradesh,531162, India.
2
University of Petroleum and Energy Studies (UPES),Dehradun, 248007,Uttarakhand, India.
3,5
University of Petroleum and Energy Studies (UPES),Dehradun, 248007,Uttarakhand, India.
4
Singidunum University, Serbia and VIT-AP University, Amaravati, Andhra Pradesh, India.
Abstract
With machine learning, the characters in the digits data set can be recognized. It is a growing
concern in pattern recognition. This study analyzes the behaviors of handwritten characters
by using the quality of features in handwritten such as font, size, styles of writing, and
symbols to make patterns. A process for identifying characters using algorithms. It is difficult
to learn algorithms for traditional human writing recognition. The proposed approach extracts
the spatial information and applies the bilinear fusion to the flow of patterns. In binary image
processing, the Bayesian network is used for classifying the contents and mathematical
methods are used to determine patterns in terms of digits. This approach yielded good results.
Keywords 1
Handwritten Recognition, Bayesian Network Theory, Digits recognition, Bayes Probability.
1. Introduction
It is a traditional approach in machine learning techniques to recognize images and detect them. The
research draws the groups from the digits dataset. Images of scanned documents were normalized in size
after they were scanned and taken from scanned documents. Using deep learning and machine learning,
handwritten character recognition has been used to analyze reading bank check forms, reading postal
addresses, and so forth. The handwritten digit recognition has been recognized by the human hand-written
digits in the topic of the boundless area of research in the emerging field of deep learning techniques. It’s
taken from papers, images, and touch screens, etc. The feasibility of digits has been treated as supervised
learning types of machine learning and significant sources have been retrieved by user digits to understand
and analyze the image recognition. The Quality of lines, space between the words, consistency of sizes,
Connectivity of strokes, Pen pressure, are to improve and identify the recognition of handwritten digits.
The collected data in terms of user attributes and typed characters. The diversities of writing types which
are space between the characters, diminish of letters are challenges to identify the digits in the dataset. The
co-occurrence of the digits has the evidential datasets of the image structure. Strokes are inherently strong
connections for dealing or inferring data to join the digits in the data set. The complexity of data to
recognize the images has more difficult to identify the target node. A large enormous of data has been
taken from the MNIST database to predict the digits given an image. The recognition of image is carried
out three steps. (a) Input of the segmentation into hypothetical symbols. It has been recognized by the
symbol classifier and determined the structural expressions. It takes into different forms of neural .Behind
ACI’22: Workshop on Advances in Computation Intelligence, its Concepts & Applications at ISIC 2022, May 17-19, Savannah, United
States
EMAIL: rohinaruna@gmail.com (A. 1); richachoudhary.86@gmail.com(A. 2); tanupriya1986@gmail.com (A. 3);
sachinandan09@gmail.com(A. 4); hkshitesh@gmail.com(A. 5)
ORCID: 0000-0001-5809-7317 (A. 1); 0000-0003-3277-9920 (A. 2); 0000-0002-9826-2759 (A. 3); 0000-0002-4939-0797 (A. 4);
0000-0001-6816-0324 (A. 5)
©️ 2020 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
203
the recognization of patterns image processing has carry over to analyses the strokes and strings. The study
has been elaborated the approximation of mathematical equations for recognizing the strokes in the
handwritten digits. The digit has increased in the image; it's become more complicated to identify the
input stroke to symbols. That symbols are said to be hypothetical symbols. It has been recognized by the
classifier. The symbol structures are recognized by the expression of structures and it has been analyzed by
the parsing algorithm. It provides the tool for analyzing the structure and coherent approaches are used for
recognition and identifying the images by using a mathematical equation. The pattern recognition tool is
used to analyze the information from raw handwritten images or digits. There are different styles of
handwriting in different communities. Images are not in sharp the redundancy of images were removed.
This approach has to be implementing with the SVM recognition system to perform by the NIST SD19
data set (https://www.nist.gov/srd/nist-special-database-19). Patterns are represented mathematically to
understand the dynamics tools for communication and information. For instance, a handwritten character
set drives the different styles of writing. Hence, if topological properties of the handwritten dataset are
connected, which signifies the redundancy of data, and sharp recognition. The structural dimension
provides to benefit the terms of accessing the large structured images.
2. Materials and Methods
The efforts are carried out by pattern expressions and recognition of characters. The images were
extracted from the user interaction data along with the properties of communication. The Extracted
data has been analyzed and tested by R-Programming for statistical analysis. It is used to provide
functions and data types for identifying the patterns. A supervised Learning algorithm has easy to
identify the strokes and structure, size, and intensity of images in the data set. Handwritten
mathematical equations have been tested in the proposed study and compared with the existing
algorithms. The dataset consists of 5000 training sets and 1000 for testing data set which helps find
the patterns of recognition. Images were accommodate and tested the Machine Learning algorithms
of Support Vector Machine, K-Nearest Neighbor and Convolution Neural Network.
Image Acquisition: The image has been captured from the scanner. It has numbers, symbols,
characters and a special set of characters has been collected and put into the sequential form of a
database.
Preprocessing: Given image has converted into a greyscale image with a threshold value of 0.5. It
removes blur spots from images and it inverts and reshapes the images.
Step 1
Step 3
Step 2
204
In the handwritten recognition to improve the features has been computed by the input images.
The process of detection and correlation is the challenging problem in the preprocessing images.
Generally categorized the process into removal of noise, normalization, and smoothing the images in
the input images.
The input image has been formatted up to the saturation level of gray scale image. Then it has been
turned into binary format. By the two categories binarization method has taken place. (1) Global
threshold (2) Local Threshold.
In the algorithm of Global Threshold: The study has taken a single threshold value for the overall
images based on the uniformity and measuring the shapes in the image.
Local Threshold: the angle has been determined by the threshold values for pixel using their
spatial information. After binarization pixel densities and angle of the text is chosen in the text image
to realize the actual angle.
Segmentation: The recognition of the process has been analyzed by the segmentation. The size of
each digit and gap between the digits are unknown. To fill the gap by the digit segmentation algorithm
to remove the noisy images and connect the components in the hand written digit. The aim of the
paper is to crop the correct segmentation for handwritten recognition by using single touch and
multiple touches of strings.
Functional Analysis is used to separate the features of digits.Input images are resized into 5*7
pixels towards the training data.
Feature Extraction: By the structural extraction the morphological features of edges, regions, and
curves has been analyzed for indexing and labeling the data to help the classification of handwritten
digits. It extracts the information from the input image. The parametric features were area, centroid,
density, line segment also extracted to find the weight and center of location in the box area.
Classification: By using soft computing techniques of multilayer perceptron is arranged in layers
which are input layer, hidden layer, and outer layer of the node. It gives the classification performance
of extracting the features. Step 1,2,3 describes how from the input images , the binarization and
features and testing’s are processed.
3. Bayesian Network Approach
BN is a system to minimize the classification error and plays a role in the prior probability of
information. It is a statical approach to quantify the various decision.
The conditional probability and prior probability has been computed as:
𝑃(𝑥⁄𝑚 ) ∗ 𝑃 (𝑚
𝑚 𝑛 𝑛)⁄
𝑃 ( 𝑥𝑛 ) = 𝑃(𝑥) (1)
Where 𝑃(𝑥) = ∑𝑘𝑗=1 𝑃(𝑥⁄𝑚 ) ∗ 𝑃 (𝑚 (2)
𝑛 𝑛)
𝑃(𝑥⁄𝑚 ) ∗ 𝑃 (𝑚 -- Likelihood event
𝑛 𝑛)
𝑃(𝑥) -- Probability of Evidence
Outliers of the decision probability
𝑃(𝐸𝑟𝑟𝑜𝑟⁄𝑋) = P(m1 /x) if we choose m1,
205
P(m2 /x) if we choose m2,
Observe x and action αi, the true nature of state is αi. Incur the λ (αi |mn).
R (αi |x) = Conditional Risk.
4. Simulation Results
The implementation results were shown below, the mapping of the model has been implemented in
Graphical User Interface components of Java programming by the eclipse tool.
Fig. 1, Shows the Handwritten character has converted into binary, removal, and segmentation.
Fig. 2, shows the performance of feature extraction using Bayesian Decision theory.
Figure 1: Segmentation of Noise removal and Binary classification
Figure 2: Recognition of Digits using Bayesian Approach
5. Result and Discussion
206
Figure 3: Compare Accuracy of Learning algorithms
Word recognition is the number of words in transcription. we combined the online and offline
mode of the system have recognized and measured the rate has compared with four existing
algorithms shown in Fig.3. which are KNN, SVM, CNN, and BN, BN has incrementally recognized
the rate of recognition as 75.20% and the accuracy as 66.10%. Each output position of KNN
recognition rate as 65.90% and the accuracy as 61.4%. SVM as 73.4% and accuracy as 65%when
compared to KNN, SVM has a high accuracy rate. Eventually, the CNN recognition rate has 73.8%
and accuracy was 65.3%.
6. Conclusion
By using the appropriate parameter parameters of feature and quality of recognition, the proposed
work of handwritten character recognition fits into the input image of the meta-database. Based on the
Bayesian network approach, we have classified effectively, decreased errors, and increased accuracy
in character recognition.
7. Future Work
To propose new classification models to improve the performance of segmentation and reduce the
complexity of algorithm to enhance faster computation.
207
8. References
[1] approached the content-based classification used by the techniques of K-Nearest Neighbour,
Decision tree, and Support Vector Machine for analyzing the classification techniques and compared
the results. SVM has not been supported and many outliers are identified in the dataset. It becomes a
core problem identifying the image classification and clustering the data. [2]proposed the
dimensionality reduction techniques used for handwritten recognition with different images in the
document and it has been analyzed by the parameters of extending the radius of complex background
with high intensities of detersive images. The pre-processing has included in the stretches of images,
normalization, binarization, segmentation, and morphological techniques. When compared the
accuracy of an image has very less and up to the mark has been identified. [3]analyzed the
performance of the neural network with geometric and gradient features. It has been efficiently
observed through the Levenberg algorithm. [4] proposed the nearest neighbor method to classify the
images with different wavelets. [5] compared the efficiency with the existing algorithms of
Hierarchical, Anamoly detection, Apriori Algorithm, and Principal Component Analysis(PCA). PCA
has yielded a good classification and efficiency of data. [7][8] implemented the character recognition
using deep learning and machine learning techniques. It has been used by the Random Forest, Neural
network, and Support Vector Machine.
[1] Mahmoud M. Abu Ghosh; Ashraf Y. Maghari, “A Comparative Study on Handwriting Digit
Recognition Using Neural Networks”, IEEE, 2017.
[2] H. Du, P. Li, H. Zhou, W. Gong, G. Luo, and P. Yang, "WordRecorder: Accurate
Acoustic-based Handwriting Recognition Using Deep Learning, "IEEE INFOCOM 2018 -
IEEE Conference on Computer Communications, 2018, pp. 1448-1456, DOI:
10.1109/INFOCOM.2018.8486285.
[3] V. V. Mainkar, J. A. Katkar, A. B. Udate, and P. R. Pednekar, "Handwritten
Character Recognition to Obtain Editable Text," 2020 International Conference on Electronics
and Sustainable Communication Systems (ICESC), 2020, pp. 599-602, DOI:
10.1109/ICESC48915.2020.9155786.
[4] J. Dalal and S. Daiya, "Image processing based optical character recognition using
Matlab", International journal of engineering sciences & research technology, vol. 7, no. 5, pp.
406-411, 2018.
[5] Anuj Dutt, AashiDutt,” Handwritten Digit Recognition Using Deep Learning”, International
Journal of Advanced Research in Computer Engineering & Technology (IJARCET) Volume 6,
Issue 7, July 2017, ISSN: 2278 – 1323.
[6] M. Zanoni, “Data mining techniques for design pattern detection,” Ph.D. dissertation, Universita
degli Studi di Milano-Bicocca, 2012.
[7] Handwritten Digit Recognition Using Machine Learning
https://medium.com/@himanshubeniwal/handwritten-digitrecognition-using-machine-learning-
ad30562a9b64”.
[8] J. Niere, W. Schafer, J. Wadsack, L. Wendehals, and J. Welsh, "Towards pattern-based design
recovery," in Proceedings of The 24th International Conference on Software Engineering, 2002,
pp. 338–348.
[9] Darmatasia and Mohamad Ivan Fanany,” Handwriting Recognition on Form Document Using
Convolutional Neural Network and Support Vector Machines (CNN-SVM), Machine Learning
and Computer Vision Laboratory Faculty of Computer Science, University of Indonesia.
[10] M. Elleuch, R. Maalej, and M. Kherallah, “A New Design Based - SVM of the CNN Classifier
Architecture with Dropout for Offline Arabic Handwritten Recognition,” Procedia - Procedia
Comput. Sci., vol. 80, pp. 1712–1723, 2016.
[11] Tina Vaz , Nagaraj Vernekar,”A Survey on Evaluating Handwritten Iterative Mathematical
Expressions”, International Journal for Research in Applied Science & Engineering Technology
(IJRASET), 2019.
208
[12] Malla V, S., Sharma, H.K., Choudhury, T. (2022). Handwritten Digit Recognition with Neural
Network. In: Sharma, D.K., Peng, SL., Sharma, R., Zaitsev, D.A. (eds) Micro-Electronics and
Telecommunication Engineering . Lecture Notes in Networks and Systems, vol 373. Springer,
Singapore. https://doi.org/10.1007/978-981-16-8721-1_51.
[13] M. Mishra, T. Choudhury and T. Sarkar, "Devanagari Handwritten Character Recognition," 2021
IEEE India Council International Subsections Conference (INDISCON), 2021, pp. 1-6, doi:
10.1109/INDISCON53343.2021.9582192.
[14] Chakraborty, P., Jahanapi, S.S., Choudhury, T. (2022). Bangla Handwritten Digit Recognition.
In: Tavares, J.M.R.S., Dutta, P., Dutta, S., Samanta, D. (eds) Cyber Intelligence and Information
Retrieval. Lecture Notes in Networks and Systems, vol 291. Springer, Singapore.
https://doi.org/10.1007/978-981-16-4284-5_14.
209