Most important project for version 8

[tomczak]

Here is a link to one relevant patent application - pretty dry but informative reading, and nothing really that revolutionary I think, but the way I understand it, it separates the image into Y'CbCr components, then uses blurring kernel separately on Cb and Cr, weighing the blurring by the variances of pixels within the kernel in the the values of chroma channel being blurred and also the Luminance (Y') channel.

http://www.faqs.org/patents/app/20130064448

It suggest that the low frequency colour noise component especially in blue channel may be the result of initial high frequency sensor noise spread over the neighboring pixels by inventing missing values for a B (and R) channels during the RAW demosaicking of Bayer matrix.

Would anyone know of perhaps an academic publication or an application of similar system with examples on how it actually performs?