Post Processing?

[chemmajik]

I'm having some problems with white shades, it seems its more blocky I've tried just about everything with filters and settings. What technique is the DIVX5 using on there post processing that gives it a glossy smoothing, we don't need anything that extreme. Maybe its a temporal filtering thats needed. What I am talking about are shades from light to dark on say a arm, or a nose & lips. The dark area's are shading fine with no blocks. I've tested with higher then normal bit rates to see if it would help but thats not the problem. No filter will solve this on the encode process. You can say its left over noise from deinterlacing, but its not that, its the size of the blocks need to be smaller on white edges. Also I think I've seen where the shade isnt going white to dark from left to right, its like its upside down for that block, lighter on the top part of the block & darker on the bottom of the block. Has anyone noticed this, I'm going to find me another clip, it could be a bad source. I'm not complaining, I just want to solve it. I've not seen alot of discussion on flaws, but if people don't speak up, how will the developers now what may need fixed or changed. This is some complicated stuff, don't believe me do a search on temporal filter on google. FPGA IMPLEMENTATION OF ADAPTIVE TEMPORAL KALMAN FILTER FOR REAL TIME VIDEO FILTERING.
Maybe there's a perimeter I'm missing to make tiling smaller in the white area's.

[-h]

I think I'd need to see a screenshot or two, as I'm not sure how Kalman filtering relates to the issue you're describing.

-h

[chemmajik]

I knew I shouldn't have destroyed my test image's... No biggy its easy to recreate. I just wanted to get a discussion going on about the white area's, and I'm not sure that KALMAN is the answer. The combinations are endless. But what exactly decides on the white area's of the nose & arms the block size? I said lips earlier but thats not correct. Make my clips & get back.

[chemmajik]

Well I found a way to solve it, but it would require a 2d cleaner filter. But the only problem is it takes a 9x9 radius & threshold of 10-12. This conversion goes from 1min up to 10mins for just a 30sec test clip at 640x480.

[chemmajik]

Field bob seems to help on the whites & runs alot faster, but it does mess with the colors to much. There's a new Donald graph smart de-interlacer 2.7 beta that I am testing out. You can kill this thread. I still believe Vertical Temporal Smoother Filter of some sort can improve things. But with the post processing code nonGPL'd I guess we'll never now.... poof

[-h]

Field bob seems to help on the whites & runs alot faster, but it does mess with the colors to much. There's a new Donald graph smart de-interlacer 2.7 beta that I am testing out. You can kill this thread. I still believe Vertical Temporal Smoother Filter of some sort can improve things. But with the post processing code nonGPL'd I guess we'll never now.... poof

That may be true, but I'm still not sure what the artifact looks like, so I can't say whether it's something we could fix on the encoding side :)

-h

[chemmajik]

It can be ultimately fixed on the encode side, but I could try to develop a post processing to temporarily solve it with a switch on or off if the post processing DSF source was available. I don't want to interfer with the encode bec that is for the experts to mess with, wants I figure out the technical term for the problem. I did post a picture of the problem, I took the link down after a day. What program is best for using to take screenshots & saving in the best format? I thought tif was the best, but hypersnap doesnt prove that. But my problem I believe is being discussed by two others in this forum that are more closer to its problem. Even thou they mention color on one, and I mention white shades on mine, and both mentioning foreground vs background.