Apparently TIFF has not one but two types of Alpha channel Associated
and UnAssociated. The premultiplication of the r,g and b channels by
255/a on load should only be done with UnAssociated alpha channels
(which seems to be counter-intuitive to me but what the hell).
The previous patch that Raster did for me didn't take this into account.
Attached is a patch against CVS loader_tiff.c to fix that.
This possibly means that line 400 (in save) should be changed to
EXTRASAMPLE_UNASSALPHA.
Cheers,
Simon
--- loader_tiff.old 2006-01-05 11:58:04.000000000 +0000
+++ loader_tiff.c 2006-01-05 12:05:56.000000000 +0000
@@ -79,6 +79,7 @@
uint32 *pixel, pixel_value;
int i, j, dy, rast_offset;
DATA32 *buffer_pixel, *buffer = img->image->data;
+ int alpha_premult =
(EXTRASAMPLE_UNASSALPHA==img->rgba.alpha);
image_width = img->image->w;
image_height = img->image->h;
@@ -104,7 +105,7 @@
g = TIFFGetG(pixel_value);
b = TIFFGetB(pixel_value);
pixel_value = (*(pixel++));
- if ((a > 0) && (a < 255))
+ if ((a > 0) && (a < 255) && (alpha_premult))
{
r = (r * 255) / a;
g = (g * 255) / a;
