G'MIC: How to make 890+ Color LUTs fit in 3.3Mb ?

Patrick David & David Tschumperlé

Day 29th May

TALK

17:00

In this talk, we will present the compression/decompression algorithm, integrated into the G'MIC opensource image processing framework (https://gmic.eu), that allows us to store more than 890+ non-parameterized color transformations to apply on images (also known as "Color LUTs", or "CLUTs"), in less than 3.5 Megabytes. It roughly achieves a 99% compression gain over the original data.

In this talk, we will present a compression/decompression algorithm, integrated into the G'MIC opensource image processing framework (https://gmic.eu), that allows us to store more than 890+ non-parameterized color transformations to apply on images (also known as "Color LUTs", or "CLUTs"), in less than 3.5 Megabytes. This algorithm determines a sparse encoding of any input color LUT and defines an associated (lossy) reconstruction scheme, based on the application of multi-scale, volumetric, anisotropic, diffusion PDE's. The reconstruction process can be parallelized and can reconstruct a CLUT at any spatial resolution. The whole compression algorithm roughly achieves a 99% compression gain over the original CLUT data.