On 2016-01-04 06:07 (+0100), Ben Thurston wrote:
On 2016-01-04 02:53 (+0100), su_v wrote:
On 2016-01-02 16:50 (+0100), Ben Thurston wrote:
I developed this math and wrote a python program to do this sort of mapping, I thought maybe it would be interesting to have in inkscape: http://benpaulthurstonblog.blogspot.com/2015/12/parameterized-quadratic-conf...
Out of curiosity, I implemeted the python program as Inkscape extension to apply such a transformation to a linked or embedded bitmap image (using the path data of a helper path for the parameters of the transformation) [1].
Comparing the results (python 2.7.11, Pillow 3.0) with your example in the blog entry of the tranformed Mona Lisa image, I wonder how you filled missing pixels in the output image?
I oversampled the original image, by reading the original image as if each pixel had another pixel (of exactly the same color) at the halfway point between the two integer pixel positions... sometimes you might have to oversample even more but you would just put more pixels of the same color in between...
Oversampling now works AFAICT as expected after this change: http://bazaar.launchpad.net/~suv-lp/+junk/inkscape-extensions/revision/747/m...
Updated demo (using oversample = 2): http://imgh.us/qpconf-demo-091-2.svg
Regards, V