BRENNEN

Industries / Mining and materials / Chemicals and petrochemicals / Specialty chemicals / Paint and coatings

Color matching

Color matching is the work of describing a color in numbers and working out which colorants, in what amounts, will reproduce it in paint.

5

Milestones achieved, from 1927 to 2003.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1927

    Forty colors get numbers

    RAL, a German committee for delivery terms founded in 1925, began standardizing color cards in 1927. Its first collection, now called RAL Classic, held 40 shades listed in the register RAL 840-HR, and from 1928 the cards were produced with the firm Hans Hansen-Schmidt. Paint makers outside Germany still sell to these numbers: Cloverdale Paint in Canada calls RAL the global standard for powder coating colors, used to specify, match and reproduce finishes across suppliers.

    LessonA number names a color but does not guarantee it; the US maker Diamond Vogel warns that its paints may differ from the RAL standard because of pigments or manufacturing.

    ral-farben.decloverdalepaint.comdiamondvogel.com

    Credit

    RAL (Reichsausschuss für Lieferbedingungen)

  2. 1931

    A theory of paint layers

    Paul Kubelka and Franz Munk published equations for how a matte paint film absorbs and scatters light, so that its color over a surface, and the thickness needed to hide that surface, could be worked out from two constants of the paint. They wrote that only experiments had addressed these questions before. Their model became the basis of computer color matching: the COMIC computer of the late 1950s solved its equations, and colorists used its two-constant form for pigments in paint and plastics.

    graphics.cornell.edusdc.org.ukcs.union.edu

    Credit

    Paul Kubelka and Franz Munk

  3. 1959

    A computer finds the recipe

    Before computers, a colorist matched a sample by mixing trial batches and judging them by eye. Hugh Davidson and Henry Hemmendinger announced COMIC, the Colorant Mixture Computer, in 1958 and put it on the market in 1959. It solved the Kubelka-Munk equations with a network of resistors and showed on an oscilloscope how close a mixture of colorants came to the sample. It sold for about $10,000 in 1960, and Chemical and Engineering News counted 172 sold by 1966, to dye and paint mills among others. Ford used it to check for metamerism, when two colors match in one light but not in another.

    LessonIts low price let it spread before general-purpose computers were affordable, and digital programs replaced it once they were; it went out of production in 1967.

    cs.union.edusdc.org.uk

    Credit

    Hugh Davidson and Henry Hemmendinger, Davidson & Hemmendinger Inc

  4. 2001

    A better color-difference formula

    A color difference is a number for how far apart two colors look, and paint makers and their customers set match tolerances in it. M. R. Luo, G. Cui and B. Rigg published CIEDE2000, which corrects the CIELAB color difference for lightness, chroma and hue and for blues and grays. Tested on four sets of visual judgments, it outperformed the earlier CMC and CIE94 formulas, and the CIE, the international standards body for light and color, adopted it. ASTM's practice for calculating color tolerances between painted panels and other opaque samples lists it among its formulas.

    api.crossref.orgastm.org

    Credit

    M. R. Luo, G. Cui and B. Rigg; the CIE

  5. 2003

    Three angles for metallic colors

    Metallic paints hold aluminum flakes that make their color change with the viewing angle, so one reading cannot describe them. ASTM's test method E2194, first issued in 2003 and developed by its subcommittee E12.12, requires measurements at three angles for paints with metal flakes, for quality control, color matching and formulating.

    LessonThe standard warns that two samples can measure the same at all three angles and still not look alike if their gloss, texture or flake orientation differ.

    astm.org

    Credit

    ASTM subcommittee E12.12

Watch for

  1. On-screen help for dark colors

    Surventis says LightAssist, a new viewing mode in the ImagePLUS tool of its Refinity color platform for vehicle repair, makes small differences in color, texture and effect easier to see on screen, especially in black and dark effect colors, and can cut unnecessary spray-outs (test panels sprayed to check a match), material use and process time. It gives no figures.

    Achieved whenrepair shops or an outside party publish counts of spray-outs per repair with and without it.

    indianchemicalnews.com

    Claimed by

    Surventis, September 2026

  2. AI and physics match metallics

    Voloboy and colleagues report a matching method for metallic car paints that joins a neural network, trained on a proprietary database of real refinishing cases, with a physical model of how the flakes reflect light. They report an 85% success rate for standardized pigments in a controlled setting, and close agreement between calculated and measured samples.

    Achieved whenanother group or a repair shop reports a similar success rate on its own cases or on public data.

    api.crossref.org

    Claimed by

    Voloboy, Ershov, Lyoushkin, Kuznetsov and Galaktionov, Scientific Visualization, August 2025

Open

  1. 01

    Car colors by paint code

    A public set of multi-angle color measurements of factory car paints, taken from salvaged body panels and labeled with each car's paint code, and a test that scores how often a method names the right paint code from a measurement.

    The brief
    Credit · unclaimed
  2. 02

    Recipe software checked with paint

    An open-source program that takes a target color and a set of tinting colorants and proposes a recipe for an opaque paint, using the two-constant Kubelka-Munk model that accounts for white pigment, and a public test in which its recipes, and those of other open tools, are mixed, painted out and measured.

    The brief
    Credit · unclaimed

Open 01 · the brief

Car colors by paint code

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of multi-angle color measurements of factory car paints, taken from salvaged body panels and labeled with each car's paint code, and a test that scores how often a method names the right paint code from a measurement.

Why it is still open

Color lookup in repair shops runs on closed databases: Axalta says its Irus system draws on more than two million formulas, and the 2025 metallic-matching study above trained on a proprietary database. Public reflectance data for car paints are a handful of samples made for computer graphics: Cornell's set names two automotive paints supplied by Ford, and EPFL's material database lists car paints among many other materials. Forensic databases of factory car paint exist but are closed: the Royal Canadian Mounted Police's Paint Data Query holds the chemistry of paint from most vehicles sold in North America after 1973 and is free only to agencies that supply at least 60 paint samples a year, and the FBI's National Automotive Paint File holds more than 40,000 samples, by a 2007 account in the NIJ Journal. Searches of Zenodo and Hugging Face for this file found no set labeled by paint code.

bodyshopbusiness.comapi.crossref.orgbowers.cornell.edurgl.epfl.chojp.gov

Built with what exists

A multi-angle spectrophotometer that meets ASTM E2194, borrowed from a repair shop, a paint supplier or a university; panels from scrapyards, with the paint code read from each vehicle's label; colour-science, an open-source Python library, for CIEDE2000 color differences; a nearest-neighbor baseline and a small model fine-tuned on one GPU (see the Note "One GPU to fine-tune").

  1. First step

    Measure 30 panels in the ten most common colors at one scrapyard, three readings each, and compare the spread within and between codes.

  2. Done looks like

    1,000 panels covering at least 300 paint codes, each measured three times at every angle the instrument offers; the spread within each code reported, which shows how much factory paint varies; the baseline's share of held-out panels assigned the right code published with the data under an open license.

Open 02 · the brief

Recipe software checked with paint

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open-source program that takes a target color and a set of tinting colorants and proposes a recipe for an opaque paint, using the two-constant Kubelka-Munk model that accounts for white pigment, and a public test in which its recipes, and those of other open tools, are mixed, painted out and measured.

Why it is still open

Open tools on GitHub calculate paint recipes but publish no test against paint actually mixed and measured. K-M Paint Unmixer solves recipes for Golden Heavy Body acrylics from measured drawdown data with a one-constant model; the paint-mixing library says its colors "are estimates, not measurements".

github.comgithub.com

Built with what exists

Artist acrylics or water-based tint bases and colorants; drawdown cards with black and white areas, so each paint can be measured over both; a handheld spectrophotometer such as the X-Rite Ci64 or the Nix Spectro 2; colour-science for CIEDE2000; the two GitHub tools as baselines; a coding agent that writes and runs its own tests, such as one built on Inkling 2 Small (see the Note "Checked without being asked").

  1. First step

    Draw down tint ladders of six colorants mixed into white at four levels each, over black and white, and fit their absorption and scattering constants.

  2. Done looks like

    100 target colors taken from a public standard such as RAL Classic; each tool's first recipe mixed, drawn down, dried seven days and measured three times; the share of first recipes within ΔE00 2.0, and of recipes within ΔE00 1.0 after one correction, published for each tool with the data and code.