Filter: Ink bleed with bad results

[uwesch]

System: Windows 7 Eval 71000 + Inkscape devel pre4-1-dev 22516

If i use the filter "ink bleed" with an object, the result contains some weird lines or edges from the bounding box or bounding box edges.

For better understanding i attached a image and put black recktangles around the "bad" area".

[su_v (suv-lp)]

reproduced with Inkscape 0.46+devel r22516on OS X 10.5.8

depending on the size of the filtered object the edges of the filter seem cut off

[su_v (suv-lp)]

… the bug is visible in the bitmap export from Inkscape as well.

[Jaspervdg (jaspervdg)]

feDisplacementMap used a different region from feTurbulence, causing the last row/column to be read as transparent black (it wasn't set by turbulence). This caused the lines at the edges. This is fixed now.

In addition the region feDisplacementMap got was a bit too big (this was true for other filters as well). This was due to strange(?) offsets to the bounding box computation in nr-arena-shape, nr-arena-image, etc. I've changed these computations to use floor and ceil instead of round-to-zero and offsets (which roughly did the same as floor and ceil, but not quite). I haven't found any rendering regressions caused by this yet (I've executed quite a few of the rendering tests and loaded a few complex files in Inkscape).

[su_v (suv-lp)]

fix committed in revision 9350

[su_v (suv-lp)]

Rendering of bitmaps with transparent background: regression in Inkscape 0.47+devel r9350 compared to r9349?

While there seem to be less artifacts at the edges of the 'Ink Bleed' filter in r9350, I noticed that importing a bitmap image like '455302-ink_bleed_error-test.png' from comment #3 (or a bitmap export from r9350 of the same SVG file) has black pixelized borders around the filtered area which are not visibly rendered in other image viewers, browsers or previous Inkscape revisions.

[su_v (suv-lp)]

@Jasper - could you take a look at this rendering regression (comment #6)?

[ScislaC (scislac)]

r9354- it really appear that this is not fixed

[Jaspervdg (jaspervdg)]

Sorry for not responding earlier, but I am unable to reproduce the problems you're seeing. I have a few uncommitted changes, but nothing that's directly related, so I'll do a clean build and see what happens then... In the mean time, what platform are you using, and what exactly did you do to get that result? (And could you try a clean build as well.)

[Jaspervdg (jaspervdg)]

I'm afraid I can't reproduce the problem you're seeing, even after a full rebuild. I doubt the problem is related to my fix for this bug, as that only affects the bounding box/filter region (and this looks more like an issue with blending?), but I'll ask on the list whether anyone else has the same problem.

[Jaspervdg (jaspervdg)]

The problem was actually with my fix for bug #165529 (I made a very stupid mistake which caused any pixel with alpha value 1 to be handled the wrong way).

[su_v (suv-lp)]

fix (committed in rev. 9358) confirmed with Inkscape 0.47+devel r9359 on OS X 10.5.8

easiest steps to reproduce (prior to r9358) had been:

1) open new document, zoom 1:1
2) draw curve (or spiral)
3) blur it
4) make bitmap copy