论文复现——Retinex-Based Signal Enhancement for Image Dark Regions

Retinex模型：参考链接：https://www.cnblogs.com/sleepwalker/p/3676600.html
Linear SSR：线性多尺度增强
Bilateral filter：参考链接：https://blog.csdn.net/l_eop/article/details/81812277

论文中的方法：利用双边滤波替换线性滤波，实现改进。
总体来说，分为两大步。
[Step 1]

1. For each pixel at coordinates (x, y), 0 ≤ x < W,
   0 ≤ y < H,
   a) Initialize the binary map Mlow(x, y) indicating
   the dark regions as Mlow(x, y) ← 1.
   b) Initialize the binary map Mhigh(x, y) indicating
   the bright regions as Mhigh(x, y) ← 1.
2. When the group of pixels affected by a pixel at
   coordinates (x, y) in the filter processing is expressed
   by
   R(x,y) = {(m, n)| x − w ≤ m ≤ x + w,
   y − w ≤ n ≤ y + w }, (19)
   for each pixel at coordinates (x, y), 0 ≤ x < W,
   0 ≤ y < H,
   a) If Y (x, y) > Klow，then set Mlow(m, n) ← 0,
   (m, n) ∈ R(x,y)
   b) If Y (x, y) < Khigh，then set Mhigh(m, n) ← 0,
   (m, n) ∈ R(x,y)
   .
   [Step 2]
   For each pixel at coordinates (x, y), 0 ≤ x < W,
   0 ≤ y < H,
   a) If Mlow(x, y) = 1, set I′i(x, y) = α · Ii(x, y).
   b) Otherwise, if Mhigh(x, y) = 1, set I′i(x, y) =Ii(x, y).
   c) Otherwise, calculate L(x, y) from (7) and (8),
   and obtain I′i(x, y) from (14).