Of course since the kernel is no longer separable, it becomes awfully
slow with large images. Here is a tileable brushed metal texture:
# pipi pipi:random512x512 --wrap --blur 20x0r25 -o image.png
git-svn-id: file:///srv/caca.zoy.org/var/lib/svn/libpipi/trunk@2715 92316355-f0b4-4df1-b90c-862c8a59935f
routine, which correctly recognises the Gaussian kernel as a separable
filter. As a result, it is now 50% faster because the normalisation
occurs at kernel build time.
git-svn-id: file:///srv/caca.zoy.org/var/lib/svn/libpipi/trunk@2661 92316355-f0b4-4df1-b90c-862c8a59935f
colour spaces for a given image. For instance, if pipi_gaussian_blur is
applied to a 32-bpp image, it is automatically converted to gamma-corrected
32-bit floats beforehands, then converted back to normal.
* TODO: clipping, regions of interest, more formats, getpixel macros...
git-svn-id: file:///srv/caca.zoy.org/var/lib/svn/libpipi/trunk@2605 92316355-f0b4-4df1-b90c-862c8a59935f
use the separation property, but it's just for a test.
git-svn-id: file:///srv/caca.zoy.org/var/lib/svn/libpipi/trunk@2603 92316355-f0b4-4df1-b90c-862c8a59935f
sam