Help: How to get rip of the artifact

[jinkazuya]

Hello...I would greatly appreciate it if anybody could help me with this. I am just wondering how to get rid of the blue band artifact, below is the picture.

http://i59.tinypic.com/w7biq.jpg

[johnmeyer]

Is it exactly the same in every frame, or does it move around, change shape, or change color? It is tough to give a specific answer without more information.

If it is only happening on a few frames, then you could use motion estimation to replace the entire frame. You could use one of the RT_Stats functions to detect the presence of the "glitch frame," and then us a conditional function to replace that one frame with one synthesized with MVTools2.

If it happens on many frames, but is in the exact same position each time, then you can use a mask and change the chroma. It it moves around a lot, and changes shape, the fix is a lot more difficult, and there may not be a good solution. If you can post a few seconds of the video, it would help.

[jinkazuya]

Here it is

http://www.firedrive.com/file/51343955470E5829

[colours]

There's been some ugly temporal processing done to the source but I don't see any blue bands. There is, however, some corruption which may cause decoding errors. Try reripping the DVD. (And if you're using DirectShowSource, use a proper source filter like MPEG2Source.)

(TinyPic is a horrible image host, by the way. Use imgur or something.)

[johnmeyer]

I didn't see any evidence of bad temporal processing. I looked at the video, frame-by-frame, and then also field-by-field. Everything looked good except for frames 55-64 all of which have some sort of local horizontal offset (to the right), but only on her forehead, just above the eyebrows. I don't have any easy solution to suggest.

I definitely did not see anything that looked like the still image you posted from one frame of the video. The fact that the artifact shows up differently under different circumstances does strongly suggest some sort of decoding error, as colours already suggested. I think his suggestion of re-ripping the DVD is a good one, but I would further suggest that you change something: use a different ripper, or come up with a different decoder for the MPEG stream.

[colours]

Well, with MPEG2Source at least, requesting the frames in 55-64 nonlinearly produces nondeterministic results so it's pretty safe to assume that either MPEG2Source is buggy (unlikely) or the source is corrupted (likely).

As for the temporal processing, there's at least some blending and probably because of vertical sharpening applied after interlacing (why would anyone do this), some of the fields around scene changes exhibit "negative" ghosting. Do a SeparateFields and compare fields 211-216, for example.