https://community.intel.com/t5/Intel-Integrated-Performance/Inverse-alpha-premultiplication/m-p/1634951#M28852 <P>Working with images that have the alpha channel premultiplied over the color channels has compositioning benefits, but persistence of such images in PNG format requires undoing the premultiplication. As such I'm looking for something to reverse the effect of the AlphaPremul function.</P><P>Is there any such thing readily available? In particular, I'm looking to use this on 8u_C4 images where one of the 4 channels contains the alpha channel by which to devide the remaining 3 channels. I believe something along those lines could be achieved using a Copy (to extract alpha channel), followed by a Dup (to duplicate alpha channel into 3 channels), and finally a Div in '<SPAN class="">8u_AC4IRSfs</SPAN>' variant where each of the 3 color channels is divided by the duplicated alpha channel. Given the number of passes through the data this would involve, I suppose it is not optimally efficient. Is there an established approach to this seemingly common scenario?</P> Thu, 03 Oct 2024 14:47:46 GMT zzattack 2024-10-03T14:47:46Z https://community.intel.com/t5/Intel-Integrated-Performance/Inverse-alpha-premultiplication/m-p/1634951#M28852 <P>Working with images that have the alpha channel premultiplied over the color channels has compositioning benefits, but persistence of such images in PNG format requires undoing the premultiplication. As such I'm looking for something to reverse the effect of the AlphaPremul function.</P><P>Is there any such thing readily available? In particular, I'm looking to use this on 8u_C4 images where one of the 4 channels contains the alpha channel by which to devide the remaining 3 channels. I believe something along those lines could be achieved using a Copy (to extract alpha channel), followed by a Dup (to duplicate alpha channel into 3 channels), and finally a Div in '<SPAN class="">8u_AC4IRSfs</SPAN>' variant where each of the 3 color channels is divided by the duplicated alpha channel. Given the number of passes through the data this would involve, I suppose it is not optimally efficient. Is there an established approach to this seemingly common scenario?</P> Thu, 03 Oct 2024 14:47:46 GMT https://community.intel.com/t5/Intel-Integrated-Performance/Inverse-alpha-premultiplication/m-p/1634951#M28852 zzattack 2024-10-03T14:47:46Z