ZigZag: A Robust Adaptive Approach to Non-Uniformly Illuminated Document Image Binarization

In the era of mobile imaging, the quality of document photos captured by smartphones often suffers due to adverse lighting conditions. Traditional document analysis and optical character recognition systems encounter difficulties with images that have not been effectively binarized, particularly under challenging lighting scenarios. This paper introduces a novel adaptive binarization algorithm optimized for such difficult lighting environments. Unlike many existing methods that rely on complex machine learning models, our approach is streamlined and machine-learning free, designed around integral images to significantly reduce computational and coding complexities. This approach enhances processing speed and improves accuracy without the need for computationally expensive training procedures. Comprehensive testing across various datasets, from smartphone-captured documents to historical manuscripts, validates its effectiveness. Moreover, the introduction of versatile output modes, including color foreground extraction, substantially enhances document quality and readability by effectively eliminating unwanted background artifacts. These enhancements are valuable in mobile document image processing across industries that prioritize efficient and accurate document management, spanning sectors such as banking, insurance, education, and archival management.