nibus
30th July 2011, 05:17
In order to achieve maximum compressibility, I'm looking for some ways to denoise without destroying the "focal point" of each frame.
For example, if there is an area of the image with higher detail (like a face) I would like to avoid denoising that area, but at the same time heavily denoise the surrounding areas (the background, walls, etc). It doesn't matter if this area is heavily blurred.
The only denoiser I know of that does this to some extent is TTempSmooth. It works ok but I'm looking for something that works at a higher contrast, meaning heavily blur some parts of the image but leave the focal point untouched (it also tends to leave artifacts on the edges of objects). I know it works on movement between frames, but I'm looking for more of something that denoises according to the detail of each frame as well. The video in question will be mostly live-action cinema.
The alternative I guess is to do heavily denoise and then do some selective sharpening along with it - but just sharpen the focal point of each frame. I've experimented with LSFMod and fft3dGPU but have not had much success.
Encoding time is not an issue because these filters will be applied to video at 640x480 or less. Compressibility is the first priority, with some detail retention being 2nd.
For example, if there is an area of the image with higher detail (like a face) I would like to avoid denoising that area, but at the same time heavily denoise the surrounding areas (the background, walls, etc). It doesn't matter if this area is heavily blurred.
The only denoiser I know of that does this to some extent is TTempSmooth. It works ok but I'm looking for something that works at a higher contrast, meaning heavily blur some parts of the image but leave the focal point untouched (it also tends to leave artifacts on the edges of objects). I know it works on movement between frames, but I'm looking for more of something that denoises according to the detail of each frame as well. The video in question will be mostly live-action cinema.
The alternative I guess is to do heavily denoise and then do some selective sharpening along with it - but just sharpen the focal point of each frame. I've experimented with LSFMod and fft3dGPU but have not had much success.
Encoding time is not an issue because these filters will be applied to video at 640x480 or less. Compressibility is the first priority, with some detail retention being 2nd.