NES Colour: FAQ

What?

The Nintendo NES generated a NTSC and PAL video directly from its PPU (Picture Processing Unit). It did not do so in common colour spaces like YIQ, YUV or RGB, and then encode them later, like many later popular consoles and computers.
- https://wiki.nesdev.com/w/index.php/NTSC_video
- https://wiki.nesdev.com/w/index.php/PAL_video
When emulators, FPGA consoles, NES RGB and HDMI mods create NES video, they do so from a given palette instead of generating NTSC/PAL video directly
- https://emulation.gametechwiki.com/index.php/Famicom_Color_Palette
There is some debate about accuracy of palettes, and I wanted to find out for myself with the help of some hardware and tools I'm familiar with through my work on film and VFX.

The more I learn about colour science in my work and in my hobbies, the more "by eye" colour matching and calibration always dissatisfies me, because your brain lies to you about colour.
I like measurements and numbers. I like colour science. I like old video games. This covers all three.
I am a terrible software developer (go on, laugh at the code I publish). But I love open source and video game preservation, so this is my small way of contributing back.

I measure 52 colours from 00 (top left) to 0C (top right), then continue 10-1C (second line), 20-2C (third line), 30-3C (fourth line).
Values roughly sweep across hues left to right, and increase in brightness top to bottom.

Specifically, a tristimulus colorimeter.
A device that objectively measures light wavelengths and amplitudes and reports them back to a computer.
It is very accurate for measuring colour specifically - several hundred times more than your eyes for a variety of reasons.
It's a bit rubbish at measuring very low light values, however even the NES's darkest colours don't fall into these problem areas.
Spectrometers (sometimes called spectrophotometers) are more accurate devices than colorimeters, but also cost 50-100 times as much for the given accuracy. As I am not a squintillionaire, I only have a colorimeter.

We plot colours on a 3D graph as a way to represent them easily.
The Delta E is the measured error or difference between two colours, plotted on a 3D graph
It comes in three flavours:
- CIE dE76 - standard euclidean distance, absolute
- CIE dE94 - updated value with respect to perceptual uniformity
- CIE dE2000 - updated once again with respect to perceptual uniformity
"Perceptual uniformity" gives a better indication of how sensitive the human brain is to colour differences at different frequencies
- For example, you are far more sensitive to differences in blues than reds, and far less sensitive to differences in greens than anything else
- See the MacAdam Elipses for a visual representation
If I refer to shorthand "dE", I mean dE2000.
A dE of 2.0 or lower is considered to be "not noticeable without close inspection", and the target dE of most professional display calibrators and production/TV/film/VFX studios.
A dE of 1.0 or lower is considered to be not noticeable at all.
Don't fall into the trap of obsessing that "dE 0.5 is better than dE 0.9". You honestly can't tell. Don't go all audiophile on me.

See the Colour Science video above. It explains why your brain lies to you about colour. Naughty brain.
Still don't believe me? Go on, look: http://www.psy.ritsumei.ac.jp/~akitaoka/color19e.html

Your brain lies to you about colour. Like, a lot.
You might even think that certain repeated colours look different sometimes. See? Your brain lied to you again. Check results with a colour picker tool.
The different backgrounds help your critical analysis
- Black is better for darker colours
- White is better for brighter colours
- Greys help mid tones. Specifically, on an sRGB calibrated screen with a Gamma curve of 2.2, middle grey represents 50% "lightness".

No. This is merely a comparative measurement, with far too much variability to be considered that.
Luminance levels vary wildly between real NES hardware models, let alone FPGA systems and emulator/computer output, which likely affects these results. I do my best to minimise that impact (see the page titled Method).
Each display contains a chip that decodes composite video into useful information for the display (typically RGB). Different chips do slightly different things. I'll do my best to note the display make/model for each comparison I do.

No. These are relative measurements on a given display, and only valuable as a comparison between two colours.
There are much smarter people than I doing work on measuring NES video co-ordinates on vectorscopes measuring electrical signal outputs directly, as well as calculating internal NES voltages with respect to CMOS voltage curve characteristics. That work will yield much more objective results.
At best, these results I'm presenting can assist palette creators to see where they need to tweak values, but again this is relative to a given display.
At worst, this is all useless, and just a bit of fun for me.

I'm not going to make that call, nor could I even if I wanted to.
I'll give numbers, stats and rendered comparisons, with the caveat that it's all relative to certain displays/chips, and you can make up your own mind.

That's not a question.
Play whatever like, however you like. There's no wrong way to play a video game IMHO.
- Correction: Playing 4:3 games stretched to 16:9 is absolutely the wrong way. Don't do what Donny Don't does.