Color tolerances of car paints

Color tolerances of car paints


Some industrial processes produce variable colors, that are adjusted to match a given target. The target is not only a single color value, it must provide tolerances for the color coordinates, in which the product can be considered as good.


Choosing optimal color tolerances is a difficult task : if they are too large, the product will have a poor visual quality. If they are too narrow, sustaining production with a high yield is not possible. Color standards such as CIE 1976, CIE 1994 and CIE 2000, which define color differences, suggest a value of ΔE=1.0 for the Just noticeable difference (JND). However, this value may vary a lot depending on materials and lighting conditions. The best method for choosing the optimal ΔE color difference is visual observation in real conditions, but this may be very expensive. In the example of car paints, this would imply building simplified (and eventually reduced-size) car models with many different paint variations, for several ΔE values, and in several directions of the color space.


We propose in this post to use the Ocean™ software for showing how physical image synthesis may be used for finding optimal color tolerances. This will never replace eye observation, but this can be used before building the real models, in order to reduce their required number. Image synthesis can be very useful for identifying potential issues, such as lighting situations where the sensitivity to the variations is much higher.


In this example, we started from one reference paint named “Ref”, and four color variations V1 to V4 based on measured data. By using linear combinations of reflectance data with the reference material, we generated normalized color variations at given ΔE* values, from 1.0 to 6.0 by steps of 1, for a total numer of twenty-four color variations.

The picture shows an image render of the group with ΔE*=6.0 compared with the reference color on spheres, lit by a uniform white D65 environment. This provides ideal values of the color measured by integrating sphere industrial colorimeters.

Figure 1 – Render of car paint variations – ΔE*=6.0

The table shows the color coordinates measured on the previous picture:

The six groups of four normalized color variations were then applied to a car model using the following scheme.

The results are presented in the gallery below, you may open any picture and navigate between the ΔE* values using the arrow buttons. They should be watched or printed on a professional calibrated device for best results, as standard displays are not reliable for discerning subtle color differences.

From these pictures, the first conclusions are:

  • Color differences at ΔE*=1.0 are not noticeable
  • At ΔE*=2.0, some differences are perceivable. The doors appear slightly greener but the effect is very subtle. Sample #3 appears slightly brighter, but this is not easily discernible from a simple reflection of the sky.
  • At ΔE*=3.0 and higher, the paints are clearly non-uniform
  • When switching between images, the color “jumps” are clearly visible, because the eye has a very high sensitivity to temporal color variations. However, this is not relevant in our case. We are only looking for spatial color differences between areas of the car.


The previous images were rendered under an overcast sky environment. The paint is lit from many directions, averaging the reflections angle-variable reflections : the resulting color is similar to the one rendered with the probe spheres above, and similar to the color that would be measured by a colorimeter with an integrating sphere.

With a directional light source such as sun light, the results may be significantly different. The reflection is not averaged, a single angle is observed instead, and will vary with sun and observer positions. For this reason, it is possible that two paint that cannot be discerned under gray sky will result in very different colors under a clear sky.

To check this, we replaced the overcast environment by a clear sky one, and rendered again the car with ΔE*=2.0.

Under clear sky, the variations #3 and #4 with ΔE*=2 are now clearly unacceptable . These paints have a higher glossiness, maybe related to the metallic particles. If the color difference to the reference paint was only ΔE*=2.0 under omnidirectional lighting, it is significantly higher under directional lighting.

This test shows that for complex materials, with mixed specular, diffuse and glossy reflection components, measuring one color value is not enough to guarantee a good color matching in any lighting condition.

Figure 2 – Car paint variations with ΔE*=2, direct sunlight


This very simple example showed that Ocean™ is a powerful tool for assessing color tolerance of industrial products, in applications where color matching is critical. The full-spectral nature of Ocean™ allow accurate colorimetry predictions from measured spectral data.

For a final conclusion in the diffuse light case, more samples would be required around ΔE*=2.0, with a panel of human observers watching at professionally calibrated display devices. The results should then be validated on a real model.

The direct sunlight test showed that two materials that match under a given lighting, might not match in another environment. Measuring a single diffuse reflection color is not enough for controlling the quality of a colored material such as car paint. Simulation tools like Ocean™ may provide a critical help for defining quality controls and tolerance, allowing to quickly check many cases, and reducing the number of physical samples necessary for validating the quality procedure.


Remark on color difference standard

In this example, we used the CIE 1976 definition for the ΔE* color difference for simplicity. It has been shown that this definition has many flaws due to perceptual uniformity issues. Improved definitions were given in CIE 1994 and CIE 2000, and should be used for defining industrial color tolerances.