I still cannot imagine what you are going to do. Could you be more specific ?
I've understood, you have several images. Let's say image1, image2, image3 and image4. Now you want to convert or combine the images applying any new image to the former result. So you load image1, apply image2, save result1, apply image3, save result2, apply image4 and save result3.
Currently I don't see why you need a pipe or where you want to speed up. It is a sequential processing, one image is processed after the other. As we are on a single-processor computer, asyncronous loading and saving won't speed up things. Loading from harddisk (which is the only thing which would benefit from asyncronous processing because of DMA) is very fast. The main issue is decoding the data and this is done by the processor.
Certainly you should keep the results in memory. If you reload e.g. result1 before you apply image3, you have a major performance decrease.
Bye,
Thomas
