Is it Chroma Shift?

[MichaelRow]

I can't figure out whether the source need to shift the chroma or not.
Here is the original frame:
http://i59.tinypic.com/w0nmn4.jpg
See the edge of the character. It seems the orange of her hair has shifted to the left. Similarly, right edge of her hair looks a little blue.
Then, I tried to used the ChromaShift filter (http://www.geocities.com/siwalters_uk/chromashift.html) to correct the chroma.
Code:
ChromaShift(C=2)
Shifted frame:
http://i61.tinypic.com/j0w3lu.jpg
But it goes to far. The blue of the sky shifted to the left edge of the her hair. As the doc(included the filter pack)says, each UV pair of values are tied to 2 pixels in YUY2, so the value can't be odd(such is C=1)
So, I'm wonder if it is chroma shift.
:thanks:

[Emulgator]

You can have subpic chroma shift with ChromashiftSP().

[Ghitulescu]

If you can't see then it's not needed :) You moved the chroma to the right and now the hair extends into the blue, which I find it more objectionable.

nevertheless, chroma shift should be performed after deinterlacing.

[SEt]

You can shift chroma (and luma too if you wish) even by fractions of pixel by calling resize functions with the same resolution but non-zero offsets. MergeChroma after that to keep original luma.

Also, are you sure you see actual chroma shift and not just using wrong chroma positions (MPEG2/MPEG1) when converting to RGB?

[MichaelRow]

If you can't see then it's not needed You moved the chroma to the right and now the hair extends into the blue, which I find it more objectionable.

Good Point.

You can have subpic chroma shift with ChromashiftSP().
You can shift chroma (and luma too if you wish) even by fractions of pixel by calling resize functions with the same resolution but non-zero offsets. MergeChroma after that to keep original luma.
Thanks for the help.

Also, are you sure you see actual chroma shift and not just using wrong chroma positions (MPEG2/MPEG1) when converting to RGB?
Not sure, I previewed it in AvsPMod,but it's a MPEG4-AVC source not MPEG1/MPEG2. afaik, the source is upscaled from lower definition. Is it true that upscaling 4:2:0 video may also cause chroma shift?

[SEt]

MPEG1/MPEG2 is just conventional way of describing center-center/left-center chroma alignment. While wrong upscaling can introduce chroma shift, you'd likely get more from incorrect YV12->RGB24 conversion.

[Motenai Yoda]

imho there is too much linedark and halo to check well, maybe try with mergechroma(awarpsharp2()) ?

[SEt]

I think his concerns about misaligned chroma are valid, but without source it's pretty much impossible to say any more.

Script and EBMP of the frame in question would be quite useful even if you can't post the source video.