文章的题目是：《Joint Enhancement and Denoising Method via Sequential Decomposition》 考虑到原始图像I I I L L L R R R ，所以使用ADM轮流更新L L L R R R L L L 一种序列评估模型，分别估计L L L R R R ，最后将评估后反射图R ^ \widehat{R} R L ^ \widehat{L} L
序列评估整体框架图：L L L R R R a r g m i n L ∥ L − L ^ ∥ F 2 + α ∥ ▽ L ∥ 1 \underset{L}{argmin}\left \| L-\widehat{L} \right \|_{F}^{2}+\alpha \left \| \bigtriangledown L \right \|_{1} L a r g min ∥ ∥ ∥ L − L ∥ ∥ ∥ F 2 + α ∥ ▽ L ∥ 1 a r g m i n R ∥ R − S / L ∥ F 2 + β ∥ W ∘ ▽ R ∥ F 2 + ω ∥ ▽ R − G ∥ F 2 \underset{R}{argmin}\left \| R-S/L \right \|_{F}^{2}+\beta \left \| W\circ \bigtriangledown R \right \|_{F}^{2}+\omega \left \| \bigtriangledown R-G \right \|_{F}^{2} R a r g min ∥ R − S / L ∥ F 2 + β ∥ W ∘ ▽ R ∥ F 2 + ω ∥ ▽ R − G ∥ F 2 α \alpha α β \beta β ω \omega ω ∥ ⋅ ∥ F \left \| \cdot \right \|_{F} ∥ ⋅ ∥ F ∥ ⋅ ∥ 1 \left \| \cdot \right \|_{1} ∥ ⋅ ∥ 1 ▽ \bigtriangledown ▽ ∘ \circ ∘ / / / W W W S S S G G G S S S
∥ L − L ^ ∥ F 2 \left \| L-\widehat{L} \right \|_{F}^{2} ∥ ∥ ∥ L − L ∥ ∥ ∥ F 2 L L L L ^ \widehat{L} L
∥ ▽ L ∥ 1 \left \| \bigtriangledown L \right \|_{1} ∥ ▽ L ∥ 1
∥ R − S / L ∥ F 2 \left \| R-S/L \right \|_{F}^{2} ∥ R − S / L ∥ F 2 S / L S/L S / L R R R
∥ W ∘ ▽ R ∥ F 2 \left \| W\circ \bigtriangledown R \right \|_{F}^{2} ∥ W ∘ ▽ R ∥ F 2 R R R
∥ ▽ R − G ∥ F 2 \left \| \bigtriangledown R-G \right \|_{F}^{2} ∥ ▽ R − G ∥ F 2 R R R S S S
假设三通道共享照明图或者假设三通道具有相同的反射率。在YUV 颜色空间中，设置Y Y Y L ^ \widehat{L} L W W W W = 1 ∣ ▽ S + ϵ ∣ W=\frac{1}{\left |\bigtriangledown S +\epsilon \right |} W = ∣ ▽ S + ϵ ∣ 1
待续…
