Abstract: When scanning paper documents, the resulting document images are often suffer from skew issues due to factors such as document placement and thickness. skewed document images adversely affect subsequent digital processing procedures including layout analysis and text recognition, resulting in low recognition accuracy. To address this, a document image skew correction method based on rank decomposition is proposed. A coarse-to-fine correction strategy is adopted. Firstly, rough skew angle estimation is performed on the document image. Subsequently, the region of interest (ROI) is automatically selected through the global projection of the document image. Rank decomposition is then conducted on the ROI image. Finally, by analyzing the rank solution information under different fine-grained angles, the optimal rotation angle is identified to complete the skew correction of the image. On the public dataset of the DISEC'13 competition, two performance metrics, the Average Error Distance (AED) and Top-80% Average Error (TOP80), were selected to compare the proposed method with other skew correction approaches. The results demonstrate that the proposed method achieves superior effectiveness in correcting skewed scanned document images. It performs excellently in both performance metrics, and can reduce the computational complexity while ensuring high document recognition accuracy.