基于掩码引导融合与双卷积前馈的去阴影网络OA

Shadows are a common phenomenon in images that often degrade image quality and impair the accuracy of subsequent visual tasks,especially under complex backgrounds and varying lighting conditions.Existing shadow removal methods frequently encounter challenges such as edge residues,color inconsistencies,and texture loss,which significantly hinder their effectiveness and generalizability in practical applications.To address these issues,we propose a Transformer-based shadow removal network—Mask-Guided Fusion and Feedforward Network(MGFF-Net).The network consists of two key modules:the Mask-Guided Fusion Module(MGFM),which in-corporates explicit shadow mask information and employs a gating mechanism to enhance the model's ability to capture shadow regions,significantly reducing color discrepancies and edge blurring in the shadow areas;and the Dual Convolution Feedforward Module(Dual-ConvFFN),which integrates local texture and global semantic features to improve detail restoration and structural preservation.Experiments conducted on three public datasets—SRD,ISTD,and ISTD+—demonstrate that MGFF-Net achieves superior performance across several evaluation metrics,including PSNR,SSIM,and RMSE.Specifically,on the SRD dataset,compared to mainstream methods,the overall image PSNR increases by 1.35 dB,RMSE decreases by 17.3%,and SSIM remains roughly the same;on the ISTD+dataset,the overall image PSNR improves by0.95 dB,and RMSE decreases by11.03%,and SSIM remains roughly the same.The ex-perimental results validate the advantages of the proposed method in detail recovery,structural consistency,and shadow region restoration.