Volume 18, No. 6, 2021
A Printed And Handwritten English Scanned Document Text Segmentation Using Deep Learning Method
Mr Roshan D Suvaris , Dr. S Sathyanarayana
Abstract
The detection of text in a document is critical step in all text recognition systems. The state of the art techniques to locate the words are grounded on the handcrafted heuristics which is fine tuned by the image processing communities’ experience. They only work if specific conditions are met, such as a relatively consistent background. We propose deep learning technique to recognize the text in printed and handwritten text documents. Our method can be used to process damaged documents or documents with variety of backgrounds. Experimental results show that our method perform better than older methods with slight engineering efforts and fewer parameter tuning. The method is tested on different types of scanned image documents and it we got an accuracy of 96.5%.
Pages: 7072-7077
Keywords: word segmentation, text segmentation, deep learning.