Revitalizing Digitized Historical Collections: AI-driven Quality Control with Locally Trained Computational Models

Abstract

<p>This paper explores <strong>AI-driven approaches</strong> by applying methodological paradigms to <strong>Quality Control (QC)</strong> for digitized collections. We propose a practical framework – <strong>the first automated QC process</strong> <strong>covering pivotal operations and workflows</strong> in <strong>digitizing ancient and fragile collections</strong>. Using a dataset of digitized images (primarily from a <strong>Chinese Rare Book Collection</strong>, we explored supervised <strong>Deep and Machine Learning</strong> algorithms for defect detection and computational methods to measure the color variations using <strong>Jensen-Shannon (JS) divergence</strong>. Our <strong>Convolutional Neural Network (CNN) model</strong> demonstrated high accuracy for our target, making them favorable for detecting digital defects in this context. The experimental results indicate that automation can significantly reduce human labor in quality assurance, thereby facilitating further character recognition accuracy and seamless access to online academic collections.<br></p>