Document Image Shadow Removal Guided by Color-Aware Background

Existing works on document image shadow removal mostly depend on learning and leveraging a constant background (the color of the paper) from the image. However, the constant background is less representative and frequently ignores other background colors, such as the printed colors, resulting in distorted results. In this paper, we present a color-aware background extraction network (CBENet) for extracting a spatially varying background image that accurately depicts the background colors of the document. Furthermore, we propose a background-guided document images shadow removal network (BGShadowNet) using the predicted spatially varying background as auxiliary information, which consists of two stages. At Stage I, a background-constrained decoder is designed to promote a coarse result. Then, the coarse result is refined with a background-based attention module (BAModule) to maintain a consistent appearance and a detail improvement module (DEModule) to enhance the texture details at Stage II. Experiments on two benchmark datasets qualitatively and quantitatively validate the superiority of the proposed approach over state-of-the-arts.

1. Introduction

Documents, such as textbooks, newspapers, leaflets, and receipts, are available daily, often saved as electronic doc-uments for digital document archives or online message transfer. Since the wide use and convenience of mobile phones, people currently prefer to use mobile phones for digital document copying. However, the captured document images become highly susceptible to shadows when the light sources are blocked. The low brightness in shadow regions reduces the quality and readability of the documen-t image, resulting in illegible content and unpleasant user experience [2], [10], [12], [19], [28], [29]. Thus, shadow removal for document images is a required image processing task in var-ious vision applications [3], [4], [33], [37], [46], [48]. Figure 1.

Document image shadow removal. With the assumption of constant background, results of [2] and [24] may cause color distortion or shadow remnant. By using a spatially varying background, our method can produce more desirable result.