topic ReduceBits - dithering alignment? in Intel® Integrated Performance Primitives

ReduceBits - dithering alignment?

gol — Sun, 28 Feb 2010 09:37:06 GMT

I'm looking for dithering & reading about ippiReduceBits, but it looks like it doesn't have a way to align/offset its dithering?

That seems to limit its use, because you don't necessarily need to convert an entire image in 1 pass, thus if you're working in blocks (say in a GUI, or when processing large images in smaller blocks), you will end up with dithering patterns with inconsistent alignments.

Edit: I see that most of the dithering algo's are error-diffusion ones, so I'm no expert in dithering but I assume that an offset would be pointless for those, they really need to work on full data?
However it would make sense for the Bayer one which is pattern-based, right?

ReduceBits - dithering alignment?

Chao_Y_Intel — Thu, 04 Mar 2010 01:53:21 GMT

Hello,

Here are some comments from the function owner:

Unfortunately, you Only can process an image as one unit for with error-diffusion desiring, no blocking.

It is right, you can divide images in blocks and process them with the Bayer.

Thanks,

Chao

ReduceBits - dithering alignment?

gol — Thu, 04 Mar 2010 08:10:26 GMT

It is right, you can divide images in blocks and process them with the Bayer.

Yes, but this would then require the function to get an offset for the Bayer pattern.

It would work if, instead of an IppiSize, it was taking an IppiRect as region of interest. Otherwise the pattern will always start at the top-left of the block to process, thus if you're processing only a random part of the image, as you would if only a rectangle was refreshed in a GUI system, the result would be random.