Ocr Of Scribbled Devanagari Lipi Using Deep Neural Learning

Abstract

Optical character recognition otherwise known as text recognition or text extraction is a tool that enables us to extract scribbled or printed text and convert them into editable files which can be further used for translations and many more purposes. Scribbled character recognition is a very crucial need of today's society and is much more challenging also. Scribbled character recognition has been the interest of so many researchers around the world. Though there is so much research on scribbled character recognition in different languages that have been done to date and many researchers are working on it as well. But comparatively the works on the "Devanagari Lipi" are very less. Although there are several works carried out on character recognition in the Devanagari Lipi character recognition is comparatively harder than in any other lipi (like Roman, Chinese, Japanese, etc.). Working with the Devanagari Lipi is sometimes difficult due to the existence of the header line, also known as "Shirorekha," which connects each syllable of the lipi to make a cohesive word with a specific meaning. The existence of this specific Shirorekha across each word complicates word segmentation. Another significant difficulty was that each human being has their own handwriting. The writing patterns of two distinct people cannot be the same; there must be some variances, and even the same person writes in changing styles over time, adding to the tool's complexity. As a result, when compared to printed characters, most scribbled characters are misread. We provide ConvNet (ConvNet) based Optical Character Recognition (OCR) for scribbled Devanagari Lipi in this study, which is reported to distinguish consonants and vowels more correctly.