 | | | | | |
| |
Hmm, I don’t think so. Traditional antialiasing spreads the partial
pixel errors from a line (or edge) into the adjacent pixels. In order
to undo it you’d need to know where the line is, but that’s the very
information you’re trying to recover from the image.
|
Strictly speaking though, (almost) any filter is invertible. Of course you do
need to know what the original filter was which is easier said than done.
| |
In this case you’re not working from an antialiased image, but rather
one that has been sampled by a scanner or a digital camera. There was
no antialiasing applied. They just save what they see, blended colors
and all.
|
Although it even worse as the image has been stored a a jpeg which bleeds the
colours even more in its compression algorithm.
| |
I’m pretty certain they already have edge sharpening and unblur
convolution filters which should be useful for undoing some of the
blending, before converting to the closest ldraw colors.
|
Yes but edge sharpening actually seems to make the antialiasing worse. If it
finds an ‘edge’ between the ‘real’ colour and its fake neighbour it tries to
exaggerate this. I suspect it has something to do with the frequency offsets in
Fourier space.
Will do,
Tim
| | | | | | | | | | | | | | | |
In lugnet.cad, Timothy Gould wrote:
| |
| |
Hmm, I don’t think so. Traditional antialiasing spreads the partial
pixel errors from a line (or edge) into the adjacent pixels. In order
to undo it you’d need to know where the line is, but that’s the very
information you’re trying to recover from the image.
|
Strictly speaking though, (almost) any filter is invertible. Of course
you do need to know what the original filter was which is easier said
than done.
|
Yeah, but if you think of line antialiasing as a sort of blur filter,
you’ll notice the convolution kernel is directional. The blurring occurs
perpendicular to the direction of the line. This implies a different
filter for each line segment on an edge. Knowing the direction of ALL
the filters is the same as knowing where the lines are.
| | | | | | | | | | | | | | | | |
In lugnet.cad, Timothy Gould wrote:
| |
| |
Hmm, I don’t think so. Traditional antialiasing spreads the partial
pixel errors from a line (or edge) into the adjacent pixels. In order
to undo it you’d need to know where the line is, but that’s the very
information you’re trying to recover from the image.
|
Strictly speaking though, (almost) any filter is invertible. Of course you do
need to know what the original filter was which is easier said than done.
|
That may be true, but antialiasing isn’t a filter in that sense. When applied
to computer-generated graphics, it’s calculated during rendering using
information that isn’t available in the final image. When aplied during image
scaling (shrinking), it makes use of the image information in the original
(higher-resolution) image. The antialiasing you see in digital camera images is
a combination of a physical antialiasing filter placed in front of the camera
sensor (often integrated into the camera’s IR filter, apparently) and smoothing
algorithms. So even if you knew which form of antialiasing was used on a
particular image, I don’t think you could create a filter to undo it.
--Travis
| | | | | | |
