OCRNet - Light-weighted and efficient Neural Network For Optical character recognition

Abstract

The developing expertise of neural networks has validated extraordinary outcomes within the text detection. The study seeks to enhance the accuracy of textual content identity to be able to make contributions to the improvement of existing technology. Two numbers, one additives, text detection and textual content reputation are used to identify the optical character. In this paper, provided a way to determine the degree of similarity between every unique character, in order, that each word may also in the end to be diagnosed. Extensive testing of two datasets, which includes TotalText and CTW-1500, indicates that the optical character detection at character level outplays State of the Art. According to findings, this endorsed technique assures that complex textual content pix, which include letters randomly orientated, bent or distorted would be recognized as being very adaptable.

Status - Accepted, to be presented on 18-20 November

Download paper here