Strong Shadow Removal of Text Document Images Based on Background Estimation and Shading Scale

Shadows may bring uncomfortable perception when taking text document images. Previous work mainly focuses on weak shadow removal issue. However, strong shadow removal is still a challenging task. This paper proposes a method based on background estimation and shading scale to remove strong shadows from text document images. Firstly, background color estimation is designed by a number of iterations through neighboring pixels' propagation. Then, umbra and penumbra are separated by morphological operations and processed by a divide-and-conquer strategy. For umbra, shading scale strategy is exploited to obtain unshadowed result. For penumbra, background replace strategy is designed to remove the shadow regions. Finally, a reference background and a text binary image are combined to generate unshadowed image. Experiments conducted on some typical strong shadow images demonstrate the effectiveness of our method.