ShaDocFormer: A Transformer-Based Approach for Document Shadow Removal
Główne pojęcia
ShaDocFormer is a Transformer-based architecture that effectively removes shadows from document images, outperforming current methods.
Streszczenie
Document shadow impacts readability on mobile devices.
ShaDocFormer integrates traditional and deep learning techniques.
Components include Shadow-attentive Threshold Detector (STD) and Cascaded Fusion Refiner (CFR).
STD uses attention mechanism for precise shadow detection.
CFR facilitates restoration of shadow areas in a coarse-to-fine process.
ShaDocFormer excels in detecting variations in shadow and illumination.
Outperforms state-of-the-art methods in experiments.
ShaDocFormer
Statystyki
現在の最先端の方法を上回るという実験結果が示されています。
Cytaty
"ShaDocFormer excels in accurately detecting and capturing variations in both shadow and illumination."
"Extensive experiments demonstrate that ShaDocFormer outperforms current state-of-the-art methods."