y4mdenoise version 4 beta 1 is checked in!

Measured by lines of source-code modified, it was a really minor change, but measured by quality increase, it was a huge change!  So I'm calling it "version 4 beta 1".

About a year ago, I got stumped by "earthquaking" artifacts in the dark, cloudy areas of the frame.  Last weekend, I decided I was tired of not getting good results (and not having done any video processing for that whole year), and got obsessed with finding the answer.  Finally, I figured it out!

The problem was that the match-size-throttle was picking a bad match before it ever had a chance to find a good match.  Now, the match-size-throttle isn't applied unless the match-count-throttle is also exceeded, or the end of a line is reached.  This has led to a SERIOUS quality increase, without making y4mdenoise any slower than it was before!

Finally, the face of my craggy old sailor looks like it should!  (BTW, I was talking about Sean Connery at the beginning of "The Hunt For Red October".  I'm still converting my LaserDisc collection to DVD, and hope to finish that before my LaserDisc player crumbles into dust.)

If we could only get the mpeg2enc guy to come in from the cold, and fix the nasty bugs that I can reproduce easily, we could finally have a bona fide good reason to release a new version of mjpegtools!

Next, I plan to tackle the multiprocessing issue again.  I've been working it out on paper for a long time, and all I find are complicated problems.  If I split frames into pieces and denoise each section separately, I run the risk of artifacts along the bounardies between slices, but if I make the denoising algorithm itself parallel internally, I run the risk of killing my performance through abuse of operating-system mutexes.  I still haven't solved these, but I decided to start coding the necessary infrastructure anyway, so that I can at least experiment with the issues.

For once, I have a day job that doesn't completely exhaust me, so I have some spare brainpower left over on weekend to do this.  I'll let you know if I get anywhere with it!

Steven Boswell

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

_______________________________________________
Mjpeg-developer mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mjpeg-developer