In most of the cases, researchers bring a free noise image, next they add noise to this image, then they apply denoising approach and finally they compare between the first free noise image and the result image and if there is a big similarity, so that’s mean they have successfully remove most of the noise from the image , and WORK DONE!
What if I don’t have free noise image, and I have only a real image, I’m asking about how will be the approach of denoising in this case. (I don’t think I’m gonna add noise to a image which is Already noisy!!, and then try to remove it by denoising ) Please what to do?? Thanks
Welcome to the community
I don’t really understand your problem, the whole point of denoising is to remove noise from an image so if you implement a denoising algorithm, you should be able to apply it on your “real image”.
The only thing that you can’t do is to compare the result to a free noise, perfect image that is used by researchers but in real life applications, you don’t need this.
Sorry if my question was not that clear. So when you denoise an image that you had add to it noise, in this case if you want to compare between the two we substrate the denoised image from the noisy one, and that in order to validate your result.
I want to know different way to make a study of filtering an image which has natural noise.
Thanks in advance for your help.
A quick search on internet using SSIM (structural similarity) :
And also :
As a photographer I’m well familiar with noise. Results of denoising are evaluated visually. Does the image look better? I understand the set up researcher use, but doesn’t it just verify that denoising algorithm can handle added noise. Results are useful if added noise actually resembles noise in real life. I would actually use set up where at first an noise free image is taken and the second image is taken with strong neutral filter. Second image would have less light due to the filter and so it would have noise. Noise that you would encounter in photography.
Your question how to evaluate results of denoising without noise free image to start with. Perhaps you could use humans to evaluate? Difference between images tells you what has changed. It could be compared to added noise. Or you could add noise to denoised image and compare results to the original image.
One way to evaluate denoising would be to use reference images. You could create a set of test images. Naturally you would need corresponding set of noise free images to compare the results with. But those reference images don’t have to be photographs, they could be created just for testing.