Research on Document Image Binarization: A Survey
Jinlong Ho, Min Liu · 2024
In the digital age, the majority of paper documents are converted into electronic formats for preservation purposes. However, the conversion process can result in significant degradation of the document images, which can subsequently lead to difficulties in reading, text recognition, and other digitisation processes. Document image binarisation is a crucial pre-processing step for automatic document recognition and data analysis. The effectiveness of this binarisation directly impacts the accuracy of the digitisation operation. This paper presents a summary of the research results based on traditional thresholding methods, image feature methods and some learning methods, as well as the evaluation indexes of binarisation performance. It also discusses the characteristics and limitations of these methods. Finally, it presents a summary of relevant methods for document image binarisation and points out several promising research directions.