I haven't been very satisfied by the Yee (--pdiff) method. More specifically,
although it sounds good theoretically, in practice I just haven't found any
situations in which it's a superior approach to just using regular diffs with
carefully thought-out thresholds and percentages.
I'd be very receptive to a proposal for a more robust basis for a perceptual
diff.
-- lg
On Aug 4, 2014, at 7:40 PM, Thiago Ize <[email protected]> wrote:
> --pdiff is a bit too aggressive since it can hide noise, which isn't always
> good when you want to make sure one image doesn't have more objectionable
> noise than another. For now I want something a bit simpler. So we could use
> the RGB->XYZ->LAB conversions already in imagebufalgo_yee.cpp and then use
> one of the differencing metrics I linked to on those LAB colors.
>
>
> On Mon, Aug 4, 2014 at 7:38 PM, Mark Boorer <[email protected]> wrote:
> Hi Thiago,
>
> Have you tried using the perceptual difference option ("-pdiff") to oiiotool?
> Or was your comment more of a request to include this particular differencing
> algorithm?
>
> I think the "-pdiff" option was added to oiiotool a little later, but older
> versions of the idiff utility (which also comes with OpenImageIO) should be
> able to do a perceptual difference with the "-p" option.
>
> Cheers,
> Mark
>
> Cheers,
> Mark
>
>
> On Tue, Aug 5, 2014 at 1:53 AM, Thiago Ize <[email protected]> wrote:
> Has anyone ever looked into having oiiotool compute a more accurate color
> difference (https://en.wikipedia.org/wiki/Color_difference)? For instance,
> the CIEDE2000. This could help when comparing two images where you only want
> the diff to fail when the two images are perceptually different. In other
> words, if my eyes can't see the difference, I don't want to be informed of a
> failure.
>
> Right now, I can get two pixels with the same error difference and yet one
> pixel will clearly look different between the two images while with the other
> pixel I can't see a single difference on my monitor. This means I either
> have to set my fail thresholds really low and accept getting a bunch of
> images for which I can't tell the difference, or I set the thresholds higher,
> but risk being uninformed of visible image differences.
>
> Thiago
>
> _______________________________________________
> Oiio-dev mailing list
> [email protected]
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
>
>
>
> _______________________________________________
> Oiio-dev mailing list
> [email protected]
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
>
>
> _______________________________________________
> Oiio-dev mailing list
> [email protected]
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
--
Larry Gritz
[email protected]
_______________________________________________
Oiio-dev mailing list
[email protected]
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
