Noise / Image Artifact Detection

Question

Is there a reliable way to detect noise or artifacts in an image, consisting of text and images (page from a PDF file), but at the same time does not harm the text or "real" images. Only the removal of noise, stains, stains, etc.

Answer 1

In general, there is no reliable, non-destructive method - this would mean asking the computer program to "magically" find out what noise and what is not. However, there are methods that are approaching practice.

One generally accepted method, which is quite simple and often not very destructive, is a small radius (3-5) median filter . The median filter is good at detecting scratches or "wrong pixels."

Another method of noise reduction is a double-sided filter , which in the conditions of a layman is mostly blurry, which respects the function.

Another way to detect and filter noise will be akin to the Pixar method used in their wavelet noise algorithm:

• reduce the image (for example, one level mip, aka 1/2 in all directions)
• subtract the reduced sampled image from the original (implicitly upsampling)
• what remains is the difference, something that cannot be represented in a lower resolution image, so the noise

Answer 2

There is no statistical way that accurately removes noise; you can use super-resolution GANs, create synthetic data, and train your model in noise-free image cleaning.