Document Rectification via Hierarchical Attention-Guided Control Point Regression

In daily life, a vast amount of information is recorded on paper-based documents, making the efficient digitization of such content a critical challenge. During the capture process, paper documents often exhibit surface deformations, such as folds and curls, while perspective distortions and geometric aberrations may further degrade image quality due to suboptimal shooting angles. These factors collectively impair the effectiveness of document digitization. To address these issues, this study proposes a control point regression-based method for rectifying distorted document images. By incorporating a hierarchical attention mechanism, the proposed approach enhances the model’s ability to adaptively focus on both local and global structural features, thereby improving correction accuracy. Experimental validation on publicly available benchmark datasets demonstrates that our method achieves competitive performance compared to state-of-the-art techniques, offering a robust solution for high-quality document digitization.