Setting alpha to 0 means that the color is completely transparent. Since cairo uses pre-multiplied alpha, you have to set the pixel to 0, since otherwise the color components could have higher values than the alpha channels. I think cairo chokes on those super-luminscent pixels.
So instead of this code:
if ((z & 0xffffff) == 0xffffff) {
z = (z & ~0xff000000) | (alpha & 0xff000000);
*((unsigned int*) &ptr [y * stride + x * 4]) = z;
}
You should try the following:
if ((z & 0xffffff) == 0xffffff) {
*((unsigned int*) &ptr [y * stride + x * 4]) = 0;
}
And while we are at it:
- Doesn't
(z & 0xffffff) == 0xffffff
check if the green, blue and alpha channels are all at 100% and ignores the red channel? Are you sure that's really what you want?z == 0xffffffff
would be opaque white. - Instead of using
unsigned int
, it would be better if you useduint32_t
for accessing the pixel data. Portability! - Your code assumes that
cairo_image_surface_create_from_png()
always gives you an image surface with format ARGB32. I don't think that's necessarily always correct and e.g. RGB24 is possible as well.
I think I would do something like this:
for (y = 0; y < height; y++) {
uint32_t row = (uint32_t *) &ptr[y * stride];
for (x = 0; x < width; x++) {
uint32_t px = row[x];
if (is_expected_color(px))
row[x] = 0;
}
}