[AGG] Re: RE: image-filtering bug (Maxim Shemanarev)

[Ivan N. <in...@pd...>]

Please note that the original test files are wrong because they don't test
the right thing (i.e. the interpolation between edge pixels and border
color). The source test files have a white border so the white gradient
surrounding the image is as expected (PhotoShop, Acrobat & AGG agree in this
case). To illustrate the problem I cropped the original image
(http://216.70.66.169/tb1a/ps-bilinear-1-v2.png 5x5 pixels). This is
PhotoShop output: http://216.70.66.169/tb1a/ps-bilinear-1-v2-CS2.png.
Besides resolving serious problems I mentioned previously, PhotoShop output
looks more truthful to the original image.

Please let me know if this helps.

Ivan


---

> > http://cyphre.mysteria.cz/tests/agg-bilinear-1-result.png
> > http://cyphre.mysteria.cz/tests/agg-bilinear-2-result.png
> > They should look the same, without the gray border in the first image.
> 
> Hmm, this is really strange as I'm actually using the pixfmt_nnnn_pre:

I tried to reproduce your picture and it works as expected.
http://antigrain.com/stuff/image_filters2.zip (compiles with pilot agg-2.4)
http://antigrain.com/stuff/image_filters2.png

Maybe you blend layers (blend_from() function)? If so, the destination pixel
format must also be premultiplied.

Also, it might be something wrong with the inverse alpha channel you use.

McSeem