3D-LUT Optimization for High Dynamic Range and Wide Color Gamut Color Processing

In A Review of 3D-LUT Performance in 10-Bit and 12-Bit HDR BT.2100 PQ, published in the SMPTE Motion Imaging Journal in March 2020, JD Vandenberg and Stefano Andriani concluded that HDR requires larger LUTs to achieve error-free interpolation comparable to those used for SDR, writing:

“Tetrahedral interpolation outperformed all other interpolations for both SDR and HDR applications. It can achieve the same quality as trilinear interpolation using a 3D-LUT 20%–25% smaller on average. The widely used trilinear interpolation generated the largest errors among the tested images. In 10-bit SDR, a 3D-LUT larger than 41 × 41 × 41, if using the trilinear interpolation, and larger than 31 × 31 × 31, if using the tetrahedral interpolation, should result in unnoticeable interpolation errors. In 12-bit HDR, a 3D-LUT larger than 72 × 72 × 72, if using the trilinear interpolation, and larger than 55 × 55 × 55, if using the tetrahedral interpolation, should result in unnoticeable interpolation errors. The saturated and dark images resulted in a much larger minimum size 3D-LUT to achieve a JND < 1, thereby confirming the findings by Pieri and Pytlarz that ICtCp might not be suited for those types of images, and other color difference metrics should be considered to evaluate the 3D-LUT size for those types of images. Finally, reducing the bit depth from 12 to 10 bits requires a considerably greater 3D-LUT size to obtain the same image quality.”

Some 10-bit images in the study even required a 3D-LUT size of 149 × 149 × 149 and higher! What we were not aware of is that the following year, Andriani, A. Zabot, G. Calvagno and JD Vandenberg published 3D-LUT Optimization for High Dynamic Range and Wide Color Gamut Color Processing in the SMPTE Motion Imaging Journal, where they acknowledge that larger LUTs aren’t feasible for on-set monitoring:

“Vandenberg et al proved that even with the most accurate 3D-LUT interpolation (i.e., tetrahedral) on average a 49 × 49 × 49 size should be used to avoid visual artefacts, i.e., in HDR the memory requirement for an error-free 3D-LUT should be more than three times larger than an equivalent 3D-LUT in SDR. Unfortunately, at the moment, no camera or on-set device can afford this complexity.”

They then go on to propose a novel method using non-uniform 3D LUTs that achieves the same color distortion of uniform ones while consuming 4X less memory space. To the best of our knowledge, the only cinema camera used in high-end production that accepts 65cube LUTs is the Alexa 35. 

JD Vandenberg is an imaging and color science manager for Walt Disney and Marvel Studios. Back in 2010, Vandenberg joined ARRI, Burbank, CA, as an ARRI ALEXA technical support engineer where he spent much of his time training cinematographers and Digital Imaging Technicians (DITs) on the features of ALEXA. In 2014, Vandenberg joined Technicolor, Los Angeles, CA, as the product manager of their on-set color correction system Director of Photography (DP) Lights. He joined Walt Disney Studios in 2018.

Stefano Andriani joined the Research and Development Department, Arnold & Richter Cine Technik (ARRI) in 2007, where, as a member of the Image Science Group, he develops cutting edge color reconstruction and sensor correction algorithms implemented in the award-winning ALEXA camera family.