image-denoising-benchmark
image-denoising-benchmark copied to clipboard
lbasek
Is the figure "Peak signal-to-noise ratio" correct?
The second figure "Peak signal-to-noise ratio" in README shows that PSNR of restored images increases as noise power increases, which is against the common sense. So is the figure "Peak signal-to-noise ratio" correct?
The second figure "Peak signal-to-noise ratio" in README shows that PSNR of restored images increases as noise power increases, which is against the common sense. So is the figure "Peak signal-to-noise ratio" correct?
Dear @TomHeaven ,
first of all thank you for showing interest in our project. Yes, later we noticed that too. We think that the error could be because of two reasons:
- Bad image choice.
- Wrong parameters in algorithms.
That diagram was made by running algorithms on only one image segment (because of limited computer power) so it is possible that chosen image segment wasn't good enough for that kind of algorithm comparison. Also the structural similarity (SSIM) on graph below is also rising as the noise is becoming greater which also doesn't make sense.
Other possible explanation would be that some parameters in algorithms were wrong. Our main goal in this project was to compare algorithms taken from other articles listed in references so it is possible that we didn't get some of them right.
If you are interested you could try to pull the repository and fix the problem following possible solutions mentioned here. We would be happy to add you as contributor.
Best regards!
