Paper Contents
Abstract
This paper introduces an OCR system using LSTM networks designed for motion and out-of-focus blur in digital camera-captured documents. Achieving a 12.3% error rate on the SmartDoc-QA dataset, surpasses ABBYY Fine Reader's 38.9%. The paper proposes a method for predicting OCR accuracy through local blur estimation, categorizing characters into readable, intermediate, and non-readable classes. The approach is validated on synthetic blurred images, making it promising for OCR quality assessment. Emphasizing smartphones in document capture, the paper presents a dataset for benchmarking OCR accuracy and quality enhancement methods, addressing the lack of databases in this field.
Copyright
Copyright © 2024 Anushree K C. This is an open access article distributed under the Creative Commons Attribution License.
