Catherine Meininger, Color Scientist, Portrait Displays

While the various ∆E color difference metrics, specifically ∆E00, have served well for assessing perceptual color differences in standard dynamic range (SDR), the introduction of high dynamic range (HDR) and wide color gamut (WCG) technology raised the question of whether these metrics were still applicable with these new capabilities. To provide an answer, researchers at Dolby Laboratories evaluated color difference metrics against experimental just noticeable difference (JND) data, where the accuracy of the metric at estimating perceptual color differences is indicated by how closely the results align with the JND data once normalized. This JND data set consists of seven colors (neutral, red, green, blue, cyan, magenta, and yellow) at three different luminance levels (0.1 cd/m2, 25 cd/m2, and 1000 cd/m2). Color difference metrics included in this evaluation were ∆E94, ∆E00, ∆CAM02-UCS", ∆L*u*v*, and a new metric designed specifically for use with HDR/WCG technology, ∆ICtCp. Results showed that no single metric outperformed the others for all stimuli, however, an argument is made that ∆ICtCp performs the best overall given this set of data. Most important to note is the failure of  ∆E00 for the 0.1 cd/m2 data set, where the metric over predicts color differences, indicating a larger perceptual difference than what actually exists. ∆E00 performs fairly well for brighter colors, however, ∆ICtCp is more consistent in aligning with the JND data set. Based on these results, it is suggested that ∆ICtCp be the metric used for the assessment of color accuracy on HDR and WCG displays [1, 3].

The initial equation of ∆ICtCp is given above [1]. Note the scalar of 0.25 for the CT axis difference. The CT scalar is required due to the intentional manipulation of the ICtCp color space from that of a pure threshold space to one that makes the best use of the available digital code values within a BT.2100 color gamut. To revert back to a threshold space, the CT axis requires a division by 2, thus resulting in a scalar of 0.25 when squared in the color difference formula [1].

Due to the completely different derivation principles behind ∆ICtCp and its ∆E counterparts, the values that result from these two metrics are incomparable to one another, thus an additional scalar modification is required to equate the two metrics. It was concluded that a scalar of 240 on ∆ICtCp creates an equivalence with the average ∆E00 result from the JND data set, and then a scalar of 3 on both metrics equates them to a JND [2]. It should be noted that in [2], there is an additional scalar on the I axis. This has been redacted in the most recent published paper on ∆ICtCp [3]. The following are the known ∆ICtCp equations based on this information and are available in CalMAN 2018 R3:

References:
  • [1] J. Pytlarz, E. Pieri, and R. Atkins, “Objectively Evaluating High-Dynamic-Range and Wide-Color-Gamut Color Accuracy,” SMPTE 2016 Annual Technical Conference and Exhibition, 2016.
  • [2] J. A. Pylartz and E.G. Pieri, “How Close is Close Enough? Specifying Colour Tolerances for HDR and WCG Displays,” IBC2017, IET Journals, Sep. 2017
  • [3] E. Pieri and J. Pytlarz, “Hitting the Mark - A New Color Difference Metric for HDR and WCG Imagery,” SMPTE 2017 Annual Technical Conference and Exhibition, Apr. 2017.