Enblend layers (was Re: [ptx] PTmender Version 0.2 "The enlightment")

[JD Smith]

On Sun, 2006-01-15 at 21:11 -0500, Daniel M. German wrote:
> 
>  Andrew> On Sun, 15 Jan 2006, Daniel M. German wrote:
>  Andrew> Do you want to find an intuitively-editable layer mask such that image A,
>  Andrew> image B, the layer mask, and photoshop normal blending mode equals the 
>  Andrew> same as the Enblend result?
>  >> 
>  >> Yes, that is the idea. 
> 
>  Andrew> With photoshop blending modes, each output pixel is a function of the
>  Andrew> corresponding layer A pixel, layer B pixel, and mask pixel. In Enblend,
>  Andrew> each output pixel is a function of a neighborhood of layer A pixels, layer
>  Andrew> B pixels, and mask pixels. These neighborhoods can be quite large as they
>  Andrew> grow exponentially in the number of blending levels you use. And you have
>  Andrew> to consider various boundary conditions (wraparound +/- 180 or not). I 
>  Andrew> think it is impossible to reduce this behavior to a function of only three 
>  Andrew> pixels that works for every pixel in the image.
> 
> I understand. But what I'd like is each layer colour corrected, and
> to keep them all in the output. If the image is flattened, then the
> result is what enblend creates currently. 
> 
> Is this feasible at all? I don't know.

Andrew basically said it is not.  There exists no mask which can be
applied to two unaltered layers to produce the Enblend output.  Unlike
the old PTStitcher, enblend does not simply "feather" a mask to
transition from one image to another: it actually intelligently selects
regions within all overlapping images to blend.  In order for this to
work, Photoshop would need some type of curve or mask layer that allows
you to specify, for each pixel, a surrounding neighborhood of pixels to
operate on, and the specific way in which to operate on them.  If this
were true, Enblend could probably be re-written as a Photoshop plugin.

Enblend can be told to ignore certain regions on input by masking them
out in the input image(s) alpha channel(s).