I was wondering if Intel would like to implement a multi scale filtering framework in IPP.
Decompose an image into 12 frequency band images (using wavelets), then allo manipulating (filtering) those images, and lastly, recompose into final resulting image.
The method can give more precise control over how the image is filtered.
It is a rather cpu demanding process, and I think it will multi-thread nicely.
which kinds of filters you want to use? IPP includes the Wavelet functions for image transforms. You can find at Chapter 13, "Wavelet Transforms", in ipp manusl (ippiman.pdf), Can these functions be used there?
I know about the IPP wavelet functions, but I was looking for a high-level framework, just as the IPP sample with UIC for Jpeg2000, in which the IPP implemementation of Jpeg2000 consists of a framework (the UIC sample) and core functions (IPP library).
As for what filters I need, I was thinking of two filter classes that can be processed with multi scale decomposition:
- Filter for local brightness correction "Shadow/Highlight". Isolate low frequency areas, and then adjust dark- or bright pixels there.
- Filter for local sharpening and contrast enhancement. Decompose image in 12 frequency bands, enhance each to a varying degree, then recompose.
I was reading these pages:
There is also a (patented) algorithm "MUSICA" with a similar idea.