Please use this identifier to cite or link to this item: http://hdl.handle.net/10603/355739
Title: Design Of Devanagari Handwritten Character Recognition System Using Various Machine Learning Algorithms
Researcher: Deore Shalaka Prasad
Guide(s): Pravin,A
Keywords: Computer Science
Computer Science Artificial Intelligence
Engineering and Technology
University: Sathyabama Institute of Science and Technology
Completed Date: 2021
Abstract: Identification of handwritten characters is gradually becoming popular in the field of research due to its many important applications like post card sorting, historical document processing, predicting age and gender of the person, digital signature verification and helping visually impaired peoples. In regard to automation, written characters detection becomes difficult. Its recognition is a deliberating challenging task due to a number of reasons like the shape resemblance of characters, different handwriting styles, different ways of writing the same character, variations in strokes and complexity of compound characters. Writing style is also impacted by the environment around the writer and the mental and physical fitness of writers. newline newline newlineIn this work, we constructed isolated Devanagari handwritten character datasets. These datasets consist of 12 vowels, 36 consonants, newline10 numerals, composite characters and 45 mostly used compound characters. They are created in pen-paper mode first, then scanned and stored as images to their particular class. Different age group writers are considered while taking samples. No restrictions are kept on age group, pen, ink, paper, etc. newline newline newlineWe proposed a novel stage-wised efficient model to recognize Devanagari Handwritten Characters. In the first stage, Convolutional Neural Network (CNN) model is implemented to recognize handwritten Devanagari characters. The model consists of stack of five Convolution2D-BatchNormalization-ReLU-MaxPooling2D layers. The transfer learning methodology is applied in the next stage and the model is trained using the bottleneck characteristics of a pre-trained network of newline newline newline newlineVGG16. Finally, to achieve better performance, the model is fine-tuned on top of the pre-trained network. We achieved promising results with fine-tuned model. Several Convolutional Neural Network models like VGG16, VGG19, ResNet and MobileNet are also explored on our newly created dataset as well as standard datasets. newline newline newlineCapsule Network (CapsNet) is one of the popula
Pagination: A5
URI: http://hdl.handle.net/10603/355739
Appears in Departments:COMPUTER SCIENCE DEPARTMENT

Files in This Item:
File Description SizeFormat 
01. title.pdfAttached File105.72 kBAdobe PDFView/Open
02. certificate.pdf746.29 kBAdobe PDFView/Open
03. acknowledgement.pdf128.05 kBAdobe PDFView/Open
04. abstract.pdf20.19 kBAdobe PDFView/Open
05. table of contents.pdf1.9 MBAdobe PDFView/Open
06. chapter 1.pdf8.28 MBAdobe PDFView/Open
06. chapter 2.pdf6.5 MBAdobe PDFView/Open
06. chapter 3.pdf16.93 MBAdobe PDFView/Open
06. chapter 4.pdf6.35 MBAdobe PDFView/Open
07. conclusion.pdf3.51 MBAdobe PDFView/Open
08. references.pdf6.35 MBAdobe PDFView/Open
09. curriculam vitae.pdf329.33 kBAdobe PDFView/Open
10. evaluation reports.pdf3.06 MBAdobe PDFView/Open
80_recommendation.pdf105.72 kBAdobe PDFView/Open
Show full item record


Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

Altmetric Badge: