32bits is per channel while 24bits is in total...
It's 32bpc images... OR... 96 bit images
24bit is 3*8 bit.
If it's a 32bit Image, it's actually an RGBA with still 8bits per channel.
Why is it impossible to change the image without loosing most of the information? Couldn't you simply not translate it to 8bpc after doing all the operations?

