Comparison of methods for document restoration and character retention and study of Marathi character recognition
Aishwarya Nalawade, Anil R. Karwankar · 2019 3rd International Conference on Trends in Electronics and Informatics (ICOEI) · 2019
Numerous techniques are being explored to digitize physical documents which exist in number of scripts and have accumulated since advent of writing. But what makes it even more difficult is when these documents have been partially destroyed by pigmentation, faded ink, charring etc. Even the slightest change in lighting while capturing the document via photograph can make it difficult for simple text recognition algorithms to extract the data. This paper studies methods to reconstruct the document in such cases where the text is not clearly visible due to lack of sharp contrast or pigmentation. Various existing methods and operations in image processing were investigated and compared which include different types of contrast enhancement, binarization, edge detectors, filters and morphological operators. It has been observed here that combining results of thresholding and adaptive binarization proves as a simple but immensely useful operation to retain maximum text while eliminating the unwanted pigmentation at the same time. The outcomes of methods used have been inspected and weighed up through visual inspection method. Further the paper discusses the steps to digitize documents in Devanagari script; largely used in the Indian subcontinent. Papers suggesting algorithms for Devanagari character recognition have been referred and the possibilities discussed.