Open CarbonPool opened 3 years ago
To my eyesight, that appears to be a smaller difference than the dirt on the monitor. Also, noise level 3 is the model with the worst preservation of the original image.
I further sharpened this original image under the cnet denoise level3 model. Compared with level2, the results are surprising. Could it be a problem with the program model? But I prefer level3, which can eliminate the jaggedness of some animated illustration lines.
I tried to use the fastNlMeansDenoisingColored function in opencv to reduce noise, using a stronger level, the same serious color band appeared
A typical adaptive smoothing algorithm applies smoothing where the color gradient is low and not where the gradient is high. This preserves the color banding (steps) once it occurs.
I captured some pixels.
The original image:![p](https://user-images.githubusercontent.com/26106045/95009260-2653a400-0653-11eb-8240-cb36f0a4d1b5.png)
UpPhoto DeNoise 3:![p(UpPhoto)(noise)(Level3)](https://user-images.githubusercontent.com/26106045/95009275-4b481700-0653-11eb-9292-04b3893b9f85.png)