Secrets of Home Theater and High Fidelity posted some interesting results for two players and how they handle video data over HDMI here. There are several things about this report that could use more clarification and different approaches to calculating the effects of these measurements could be taken.
The article states that “The overall dynamic range we would expect from the Sony has gone from 220 values (16-235) down to 209. That’s a loss of 5% of your dynamic range.” From a bit depth standpoint that is true, but these devices use gamma and degamma curves. The actual loss of dynamic range from a linear luminance perspective is actually 10% from 16 to 235. This will cause the image to be coarsened in this process of re-sampling the data into a narrower data range. This is confirmed with their observations of the image being less smooth with the Sony.
The only delta E 1994 values I could reconstruct from the information provided were the ones for the reference Y (luminance levels). The calculation method used in the original article is incorrect for those. The calculated dE1994 values were referenced to a maximum value of 254, a minimum of 0 and a gamma of 2.2. Delta E is a representation of relative luminance error. Data values 16 and below have zero luminance in the Rec. 709 standard for HDTV which is what a Blu-Ray player is intended for. The data should therefore be shifted appropriately by 16 with all values below 16 having zero luminance and therefore 0 delta E error to represent how this would work on an actual image. The corrected values using this correction are shown in Figure 1.
Figure 1 – dE1994 For Black Level Of 16
A more complex situation exists as to what is the best reference luminance for a given data point for a perceptual luminance error. It could be taken as 254 which is the maximum signal level like Secrets did. However, the standard references 235 as reference white. The standard also references 103 cd/m2 which is what the 235 level is translated to on a display. Since most displays can be calibrated to hit 103 cd/m2 whether the input signal is 235 or some other value at this level it seems inappropriate to assign a delta E error other than zero at many values transmitted for the reference white signal. I find that calculating delta E based on the error relative to the gamma curve for each sample is more consistent with what is seen on an image. Using this as the reference will penalize errors in the darker regions much more than what was used in the article. This makes sense since relative contrast errors on images are much more important than relative luminance compared to the reference peak white level. This is probably because within a neighboring region of an image it is more likely to be of similar luminance than not. The resulting values for this are shown in Figure 2. Reductions in these errors are possible by shifts in black level (brightness control), but without more data it is unclear whether that is reasonable or not.
Figure 2 – dE1994 Using Gamma Luminance Reference
The article is unclear in how the data relates to player output versus the source content. The Oppo player data is labeled as percent saturation while the Sony is in percent intensity. How are individual Cb and Cr values used to calculate a dE1994 values? Is a cell in each table a part of a sample or is it the row?
I spent hours examining color bars from my AccuPel HDG-4000 video generator compared to an Avia Pro DVD and a Sony Blu-Ray from a Sony PS3. I do not own the Quantum Data unit to examine the data from the player, but I do own a Photo Research PR-670 spectrophotometer which can easily measure the color and luminance shifts from the display of the type shown in the Secrets article if they were to exist. I could not find any shift in color or luminance no matter the colorspace setting on my Sony PS3 relative to the AccuPel signal generator.
I have seen luminance and color problems like those mentioned in this report, but I find color errors like those claimed in hue and saturation from a player to be very rare. To be fair these types of problems are not just limited to sources, but are also found on displays.
It should be noted that this analysis assumes the data in the original report is correct. It is possible different settings or firmware in the Sony Player could alter the results for that product.
Leave a Reply