A brief history of color science — from Newton's prism to Helmlab's 64K human observations

Close your eyes for a moment, then open them. Everything you see — the white of your screen, the brown of your desk, the blue beyond the window — is the result of billions of neurons in your brain interpreting light. But it took humanity centuries to learn how to express that interpretation with mathematics.

In this post, we will trace the 360-year journey of color science. From the moment Newton held a prism to sunlight, to the emergence of a color space built on 64,000 human observations. At each step, what was gained and what was lost?

1666: Newton and the Prism — It All Starts with Light

Isaac Newton was 24 years old, conducting an experiment in his darkened room at Cambridge. He closed the curtains, allowing only a thin beam of light to enter, and passed it through a glass prism. The prism split the white light into a rainbow: red, orange, yellow, green, blue, indigo, and violet.

Before this experiment, people believed that color was an intrinsic property of objects. An apple was red because "redness" was in its nature. Newton turned this idea completely on its head: color was a property of light, not of objects. White light was a combination of all colors.

Newton also added a second prism and recombined the rainbow back into white light. This demonstrated that color was something that could be decomposed and recomposed — in other words, something that could be expressed mathematically.

What was gained: The physical foundation of color — the seeds of the wavelength concept. Color was no longer mystical; it was a measurable phenomenon.

What was lost: Nothing yet, because the journey had just begun. But Newton's model treated color as pure physics; human perception was left out of the equation entirely.

1802-1850: Young and Helmholtz — Three Colors Are Enough

Newton had observed an apparently infinite number of hues in his rainbow. But did the human eye perceive each of them separately?

In 1802, Thomas Young proposed a bold hypothesis: the human eye has only three types of color receptors. All colors are perceived by stimulating these three receptors in different proportions. We do not need a separate receptor for every wavelength — three are enough.

Roughly half a century later, Hermann von Helmholtz expanded and solidified this theory through systematic experiments. Helmholtz showed controlled light sources to subjects and measured their perceptual responses. The conclusion: Young was right. The retina contains three types of cone cells — sensitive to red, green, and blue light.

This "trichromatic theory" still forms the foundation of color science today. Every pixel on your screen emits red, green, and blue light — because your eyes operate through these three channels.

What was gained: A biological model of color perception. The idea of reducing infinite wavelengths to three numbers laid the groundwork for modern technology, from digital displays to cameras.

What was lost: The "three channels" had been discovered, but how differently sensitive each channel was had not yet been measured. The fact that we are far more sensitive to green light than to red, and least sensitive to blue, would only become clear later.

1931: CIE XYZ — The First International Standard

By the early 20th century, everyone who worked with color — painters, printers, lighting engineers — faced the same problem: there was no way to be certain that two people meant the same thing when they said "that color." When you said "this red," the other person might understand a completely different red.

In 1931, the International Commission on Illumination (Commission Internationale de l'Eclairage, or CIE) took a major step to solve this problem. In dozens of experiments, they showed people mixtures of light and asked them to "match these two colors." From the resulting data, they derived color matching functions.

The CIE XYZ color space, built on these functions, expresses every visible color as three numbers (X, Y, Z). Regardless of physical spectral distribution, two lights with the same X, Y, Z values look identical to a human — a phenomenon called metamerism.

What was gained: A universal language of color. Anywhere in the world, when a color was specified as X=0.42, Y=0.35, Z=0.15, everyone understood the same thing. This became the shared language of international trade, printing, and lighting engineering.

What was lost: CIE XYZ measured color matching, not color difference. A 0.05 difference in X could sometimes look dramatic and sometimes be nearly invisible. The space was not perceptually "uniform" — it could tell you whether two colors were the same, but not how different they looked.

1976: CIE Lab — The Attempt at Perceptual Uniformity

The biggest shortcoming of CIE XYZ was its failure in distance calculations. How different are two colors? XYZ could not give a reliable answer.

In 1976, the CIE defined the CIE Lab color space to address this problem. The core idea rested on an elegant mathematical transformation: applying a cube root to XYZ values. Why cube root? Because the human eye's response to luminance roughly follows such a curve — highly sensitive in dark tones, less sensitive in light tones.

CIE Lab introduced three axes:

L* — Lightness (0 = black, 100 = white)
a* — Red-green axis (positive = red, negative = green)
b* — Yellow-blue axis (positive = yellow, negative = blue)

Now the distance between two colors could be calculated with a simple formula: the Euclidean distance in Lab space. This was called Delta E. Practical rules emerged like "1 Delta E is a barely perceptible difference."

CIE Lab became the standard for the paint, textile, printing, and food industries for decades — and is still used in many fields today.

What was gained: For the first time, color difference became calculable. A printer could express numerically how much a printed color deviated from the original. The foundation of quality control was established.

What was lost: Lab's promise of "perceptual uniformity" was only half-fulfilled in practice. Particularly with vivid (highly saturated) colors, in the blue region, and in dark tones, it made significant errors. The same Delta E value could look like a huge difference in some color regions while being nearly invisible in others. The cube root was a good approximation — but real human perception was far more complex than that.

2000: CIEDE2000 — Patching Lab

CIE Lab's errors became increasingly apparent over the years. The textile and paint industries in particular suffered from Lab's failures in dark tones and vivid colors.

The CIE, in a formula published in 2001 (but based on data from 2000), attempted to correct Lab's errors with CIEDE2000. This formula did not change the Lab space itself — instead, it added correction terms to the distance calculation formula:

SL — Lightness-dependent weighting. Compensates for distinctions Lab could not make in dark tones.
SC — Chroma-dependent weighting. Adjusts distance calculations for vivid colors.
SH — Hue-dependent weighting. Provides different sensitivity in different color regions.
RT — Rotation term. Corrects for the interaction between hue and chroma in the blue-purple region.

CIEDE2000 significantly improved Lab's unreliable distance measurements and has been the industry standard for 25 years. Paint, printing, textiles, food — it is used wherever professional color matching is needed.

What was gained: Repair of Lab's worst errors. Reliable color difference measurement in industry was finally possible.

What was lost: CIEDE2000 is not a color space — it is a distance formula. It operates in Lab space, using Lab's coordinates. This made it useless for color generation: building palettes, creating gradients, or finding "the closest in-gamut color" all required Lab's distorted coordinates. Furthermore, the formula was enormously complex: instead of a single equation, it was a long chain of nested conditions, angle corrections, and special cases.

2020: Oklab — The Power of Simplicity

In 2020, Swedish software engineer Björn Ottosson brought a fresh approach to a familiar problem in color science. Ottosson's Oklab — "OK Perceptual Lab" — was focused on simplicity: just 18 parameters. A 3x3 matrix and cube roots. That was it.

Oklab attempted to re-deliver on CIE Lab's original promise — a simple, uniform color space — using modern data. The results were impressive: far more uniform than Lab, no need for CIEDE2000's patches, and critically — because it was a proper space, it could be used directly for color generation, interpolation, and palette building.

Its greatest achievement was its inclusion in the CSS Color Level 4 specification. The oklch() function now works in all modern browsers — perceptual color space access in pure CSS, no JavaScript required.

What was gained: Simplicity and accessibility. Web developers could use a perceptual color space directly in CSS for the first time. Palette generation, color transitions, lightness adjustments — all produced more consistent results.

What was lost: Eighteen parameters cannot capture the full complexity of human perception. Oklab cannot model the Helmholtz-Kohlrausch effect (vivid colors appearing brighter than they are). It cannot apply different calibrations for different color regions. It cannot fully capture how perception in dark tones differs from light tones. It is an elegant but approximate solution, aware of its own limitations.

2025: Helmlab — 64,000 Human Observations

All previous approaches had tried to model human perception with a handful of parameters or patch-like correction terms. Helmlab asked a different question: what happens if you ask enough humans and provide enough parameters?

Behind Helmlab are more than 64,000 human color-difference observations. These observations were gathered from thousands of people evaluating 3,813 color pairs. The answer to "How different do these two colors look?" — in numbers, measurably.

A 72-parameter, 13-stage transformation was optimized against this massive dataset. Helmlab's processing pipeline works as follows:

Matrix transformation — Projects XYZ coordinates into perceptual-basis channels
Power function compression — Instead of CIE Lab's cube root, gamma values learned from data
Second matrix — Constructs perceptually meaningful L, a, b axes
Hue correction — Different calibrations for different color regions (red, blue, green...)
Helmholtz-Kohlrausch effect — Models the perception that vivid colors appear brighter
Dark tone correction — Handles the perceptual differences in dark tones separately
Chroma scaling — Hue-and-lightness-dependent chroma calibration
Neutral correction — Guarantees that grays map to exactly zero chroma

This complex pipeline has a single goal: minimize the gap between the "distance between two colors" that Helmlab calculates and the distance humans actually perceive.

The result? A STRESS score of 23.2. That is 23% more accurate than CIE Lab (~30.3), 20% more accurate than the industry standard CIEDE2000 (29.2), and 16% more accurate than the modern Oklab (~27.5). Here is the comparison at a glance:

Helmlab is also a full color space — not merely a distance formula. Unlike CIEDE2000, it can generate palettes, create gradients, perform gamut mapping, and export color tokens. It is available as a 10KB npm package with zero dependencies:

npm install helmlab

What was gained: The lowest STRESS score to date. A data-driven color space that models human perception more closely than any alternative. Usable for both distance calculation and color generation.

What was lost: Simplicity. Seventy-two parameters and a 13-stage pipeline are a far cry from Oklab's 18-parameter elegance. There is no CSS integration — JavaScript is required. But this is a deliberate trade-off between accuracy and simplicity: if human perception is not simple, the system that models it cannot be simple either.

Looking Back at 360 Years

At each step, a different question was asked and answered at a different cost:

Newton asked "What is color?" He answered with the physics of light — but ignored perception.
Young and Helmholtz asked "How does the eye see color?" They found the three-channel model — but could not measure the sensitivity of those channels.
CIE XYZ gave a universal answer to "Are these two colors the same?" — but could not answer "How different are they?"
CIE Lab answered "How different?" for the first time — but the answer was approximate, especially wrong for vivid and dark colors.
CIEDE2000 patched Lab's errors — but remained a formula, not a space; useless for color generation.
Oklab brought perceptual color to the web with simplicity and accessibility — but 18 parameters could not capture the full complexity.
Helmlab optimized 72 parameters guided by 64,000 human observations — trading simplicity for accuracy.

There is a recurring theme in the journey of color science: each new step sees what the previous one could not, but brings its own limitations along for the ride. Newton's model was physically correct but perceptually incomplete. CIE Lab was perceptually ambitious but data-starved. CIEDE2000 was data-driven but structurally constrained.

Helmlab, as the latest link in this chain, was trained on the largest human perception dataset ever assembled. But the history of color science teaches us something: this will not be the last link either. New data, new models, new understanding will come.

What matters is the direction of this 360-year journey: from physics to perception, from intuition to data, from approximation to precision. Newton began by looking at a prism. Today we continue with data gathered from the eyes of 64,000 people.

npm install helmlab

Add the most current link in 360 years of color science to your project — 10KB, zero dependencies.

A Brief History of Color Science: From Newton to Helmlab