https://hgpu.org/?p=5294
Acceleration of an improved Retinex algorithm