Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/373388
Title: | Segmentation of offline handwritten Devanagari words and their recognition into characters |
Researcher: | Kohli, Monika |
Guide(s): | Satish Kumar |
Keywords: | ANFIS CBF DWT FOV MIF |
University: | Panjab University |
Completed Date: | 2022 |
Abstract: | Optical Character Recognition (OCR) is a field of pattern recognition. Online character recognition caters to convert text written on special devices like PDA, digitizer into letter codes whereas Offline converts image of the text into letter codes. In Devanagari script, present state of art is unable to deal with touching characters, compound characters, shadowed characters. The conventional approaches fails to deal with the complexity of the script.in such which share upper and middle zone segment components and thus reducing recognition accuracy in handwritten data. Present work focuses on the different phases caters to OCR process for the recognition of handwritten Devanagari handwritten words using deep learning technique. There are many challenges which are discussed in the work. Headline line detection, Cropping individual components from the word image, presence of shadowed characters, differentiating touching and nontouched components etc. have been proposed in this study. Handwritten word image must be segmented to extract individual components for recognition. It is the pivotal essence of the pre-processing phase. This work focuses on techniques which facilitates segmentation in Devanagari script (Hindi) for offline handwritten words. In the present study, for word database, benchmark dataset (dataset-1) of 80 writers consisting of legal amount words is used. Another dataset (dataset-2) of 15 writers is prepared which consists of legal amount words with touching components. Classification is performed using supervised learning methods. For the purpose of training, database consisting 65 characters having 2000 images each is used. The benchmark database for 46 basic characters is extended to include modifiers and half characters. Feature extraction is one of the vital phase for recognition. In this work, Spatial Transfer Network model is applied for recognition of segmented components with the accuracy rate of 93.60% newline |
Pagination: | 131p. |
URI: | http://hdl.handle.net/10603/373388 |
Appears in Departments: | Department of Computer Science and Application |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_title.pdf | Attached File | 4.36 kB | Adobe PDF | View/Open |
02_certificate.pdf | 1.14 MB | Adobe PDF | View/Open | |
03_acknowledgements.pdf | 5.59 kB | Adobe PDF | View/Open | |
04_abbreviations.pdf | 12.97 kB | Adobe PDF | View/Open | |
05_ table of contents.pdf | 76.28 kB | Adobe PDF | View/Open | |
06_list of table.pdf | 71.31 kB | Adobe PDF | View/Open | |
07_list of figures.pdf | 37.65 kB | Adobe PDF | View/Open | |
08_chapter 1.pdf | 136.57 kB | Adobe PDF | View/Open | |
09_chapter 2.pdf | 98.54 kB | Adobe PDF | View/Open | |
10_chapter 4.pdf | 307.45 kB | Adobe PDF | View/Open | |
11_chapter 5.pdf | 334.71 kB | Adobe PDF | View/Open | |
12_chapter 6.pdf | 243.05 kB | Adobe PDF | View/Open | |
13_chapter 7.pdf | 208.86 kB | Adobe PDF | View/Open | |
14_chapter 8.pdf | 25.52 kB | Adobe PDF | View/Open | |
15_references.pdf | 120.72 kB | Adobe PDF | View/Open | |
80_recommendation.pdf | 23.56 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: