I am trying to take advantage of the convolution capabilities of the GIL (with the numeric extension of course). I have some text that I render first using Cairo and then I pass the raw image data along to the code I'm using with the GIL with some simple glue code. This much of my code works great - I can write out a PNG using png_write_view after handing the data off to the library and it contains the rendered text as I would expect. There is a problem I've encountered, however, where the anti-aliasing pixels of the white text I have rendered against a transparent background appear dark and the blurred part of the text after convolving the image is also dark. I'm pretty sure this is due to the fact that the raw RGBA data I pass along already contains premultiplied alpha from Cairo. I'm guessing that the data is treated as if the alpha has not been premultiplied. I've been trying to figure out the best way to handle this situation and have had very little success in getting anything to work. I have recently been attempting to try one of the 2 things listed in this mailing list post by Lubomir Bourdev where he gives an example of approaches one could take to handle premultiplied RGBA data: http://lists.boost.org/Archives/boost/2006/10/112532.php. No matter what I try, however, I just keep getting endless compile errors and am probably going about it all wrong. Right now I've been trying to do something where I create types to represent premultiplied RGBA data: namespace boost { namespace gil { struct premultiplied_rgba_t : public rgba_t {}; typedef layout premultiplied_layout_t; typedef pixel premultiplied_pixel_t; typedef image premultiplied_image_t; } } ...so far this has just resulted in compile errors and I can't even begin to work on any kind of custom color conversion sort of thing that uses this new type yet (though actually I tried to once and couldn't make it work either). I'm guessing that what I need to do is either unmultiply the RGBA data first or I need to somehow get the library to work correctly with the premultiplied data. Can anyone help me figure out how to make this work? David Warkentin