Two Pass Processing

suggest a way to improve Neat Video
Post Reply
gomactrading
Posts: 10
Joined: Sat Jan 13, 2007 10:55 pm

Two Pass Processing

Post by gomactrading »

I have been using NV for years in conjunction with 8mm film processing. Most of the times the footage on one reel is such a mix that it has to be broken in many pieces in order to apply satisfactory noise removal specific for each piece. I am suggesting, if possible to have a two pass process where in pass one the software determines the best option for each frame, save the information to a log/data file and in pass 2 execute the log/data file to achieve the best possible effect. For people working with VirtualDub, like I do, Deshaker is probably the best example of how it could be done.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

This is a good idea and it could perhaps work even without two passes if there was an easy and simple way for the software to automatically determine the best option for each piece. Automating that is not trivial, which is the key problem.
gomactrading
Posts: 10
Joined: Sat Jan 13, 2007 10:55 pm

Post by gomactrading »

Well, right now enduser selects a frame and does an auto profile. That can be automated. Then the clip is advanced one frame. Than can be automated. The profile information is written to log/data file fore each such frame. That file can be opened with a low level command and info can be appended programatically. The only problem that I can see is when the software is unable to find a uniform/large enough area in a frame to get a good DNP. That frame may be skipped/zeroed out in the log/data file and info from previous frame may possibly be used. Eventually, the log file will contain a large number of individual profiles that would be far more beneficial than the current system where there is only one profile that is applied across one entire clip. As I work on daily basis with film and NV I encounter these situations time and time again where the profile is just not right for the entire clip and I keep thinking "there must be a better way"
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam »

gomactrading wrote:Well, right now enduser selects a frame and does an auto profile. That can be automated.
Not completely.

When you build a noise profile manually, but hand, you do two things in the simpest case: (1) you click Auto Profile and (2) you visually inspect the area chosen by Auto Profile and if there are any details there or some technical issues (like clipping), then you manually select another area and click Auto Profile again to build another profile based on that sample area. The part that is difficult to fully automate (to make it work the same well as if done by a human operator) here is the visual inspection. It is easy for a human operator because he can easily determine if the selection contains only noise or something else, like useful details. For example, the selection may contain sand on the beach. It generally looks just like noise to the software. But a human operator recognizes such a case easily, even though that is in fact a pretty complicated task, which requires experience and intelligence. Solving such kind of task is not easily automated in software. Humans are just much better at solving it than any software program.

If we accept the risk of potentially using a bad area for profiling (as to avoid the need for visual inspection at the time of profiling) then a potential outcome would be that, because a noise profile might come out wrong in some part of the video (say, in one specific scene or shot), that part of the output video could be processed inaccurately. To deal with such a possibility, you would have to manually inspect the whole output clip to make sure all its parts look fine. If some part looks bad then you would have to redo it manually.

So basically, this would replace one problem (the need for visual inspection during profiling) with another (need to inspect the whole output clip). The first problem still seems easier and faster to solve for a human operator.
Post Reply