Re: [css3-images] honoring EXIF orientation from images

On Wed, Nov 30, 2011 at 4:09 PM, L. David Baron <dbaron@dbaron.org> wrote:
> One thing we've discussed at least once in the past, and I'm
> surprised not to see in css3-images (other than in a note), is a way
> for Web content to opt in to browsers honoring the EXIF orientation
> on images.
>
> There's a compatibility problem with honoring EXIF orientation by
> default.  However, providing authors a way to opt in to honoring it
> would allow authors to avoid the problem of incompatibility between
> the other image processing software they use and the Web platform.
>
> It seems that an obvious place to put this functionality would be in
> the image-orientation, perhaps as 'image-orientation: from-image'.
>
> However, this would also require extending image-orientation to
> support flipping, since there are 8 EXIF orientations.  It's not
> immediately clear to me how to do this given the current syntax
> (expressing image-orientation as an <angle>).  However, I tend to
> think that <angle> may be the wrong syntax for image-orientation:
> transformations other than the 8 EXIF orientations seem like the
> role of 2D transforms and not a sensible future extensibilility
> direction for the image-orientation property.
>
> I would suggest reworking image-orientation to support the 8 EXIF
> orientations with a non-<angle> syntax and then adding a
> 'from-image' keyword.  (I think the <angle> syntax is also bad
> because it suggests that angles that aren't multiples of 90deg are
> meaningful.)

I discussed this with Chrome engineers recently too, and they also
expressed a desire for this.  My intention was to wait for level 4 to
do so, but if you think it's appropriate to drop into this level, I'm
fine with that.

My suggested grammar is:

image-orientation: from-image | [ <angle>? flip? ]

Rotations from <angle> are done before flipping.  Flipping is done in
the inline direction.

Any non-angle syntax would just be naming quarter-turns with another
syntax.  The current spec just rounds to the nearest quarter-turn,
which already precludes any future extension to non-quarter-turns.
That seems fine to me.

~TJ

Received on Thursday, 1 December 2011 00:38:59 UTC