Maximus_G
16th January 2004, 07:55
Hello, everyone,
i've posted this at virtualdib.everwicked.com, but unfortunately that thread received little or no attention at all.
We are now working on some WWII footage ("Memory" project, http://www.translate.ru/url/tran_url.asp?l...ransliterate=on )
The big problem that we have faced is the quality of the material - there are so many scratches... and the only way to fight them that i've found is to use these plugins:
1. http://i4004.0catch.com/ctmedian.htm
2. http://bag.hotmail.ru/despot/despot.dhtml
3. http://observatory.playland.ru/article21_a...icle_24_1.phtml
The main problem with these filters is:
1. They produce artifacts, more or less.
2. They have a big constraint - a fast-moving picture is treated as noise/scratches to be removed. If we raise the threshold parameters, than we just cannot detect scrathes effectively. So these plugins work well only when the picture is not shaky and doesn't contain too much motion. Needless to say that aviation footage is both shaky and fast.
The question is - are there any other scratch removal filters/plugins, equally or more effective as those mentioned, including plugins for programs other than VDub and AviSynth?
On the Doom9.org forum i've found information on project "Brava" http://brava.ina.fr/index.en.html
It's even more than we need, and they have produced the plugins for Premiere, but as we presume, those plugins cannot be obtained.
Maybe there's something like that?
If no, than we'd like to discuss the aspects of a special filter development, involving object recognition and restoration techniques.
---
SO, in 2 words: :)
If we use an image recognition technique, the filter would be able to distinguish dust and scratches from the main material and temporal noise just the way WE (humans) do - and that would be the best way to localize scrathes properly, and treat'em as needed. Furthermore, using motion map, we could restore lost parts by transforming them from previous/next frames.
Do such filters exist? If no, are they needed by the community?
Best regards,
Maximus_G
i've posted this at virtualdib.everwicked.com, but unfortunately that thread received little or no attention at all.
We are now working on some WWII footage ("Memory" project, http://www.translate.ru/url/tran_url.asp?l...ransliterate=on )
The big problem that we have faced is the quality of the material - there are so many scratches... and the only way to fight them that i've found is to use these plugins:
1. http://i4004.0catch.com/ctmedian.htm
2. http://bag.hotmail.ru/despot/despot.dhtml
3. http://observatory.playland.ru/article21_a...icle_24_1.phtml
The main problem with these filters is:
1. They produce artifacts, more or less.
2. They have a big constraint - a fast-moving picture is treated as noise/scratches to be removed. If we raise the threshold parameters, than we just cannot detect scrathes effectively. So these plugins work well only when the picture is not shaky and doesn't contain too much motion. Needless to say that aviation footage is both shaky and fast.
The question is - are there any other scratch removal filters/plugins, equally or more effective as those mentioned, including plugins for programs other than VDub and AviSynth?
On the Doom9.org forum i've found information on project "Brava" http://brava.ina.fr/index.en.html
It's even more than we need, and they have produced the plugins for Premiere, but as we presume, those plugins cannot be obtained.
Maybe there's something like that?
If no, than we'd like to discuss the aspects of a special filter development, involving object recognition and restoration techniques.
---
SO, in 2 words: :)
If we use an image recognition technique, the filter would be able to distinguish dust and scratches from the main material and temporal noise just the way WE (humans) do - and that would be the best way to localize scrathes properly, and treat'em as needed. Furthermore, using motion map, we could restore lost parts by transforming them from previous/next frames.
Do such filters exist? If no, are they needed by the community?
Best regards,
Maximus_G