Photography & simulation comparison
By Chloé Terreau
Ocean™ is a realistic renderer that, based on 3D model and material data, provides predictive and realistic render. One way to prove that Ocean™ is able to simulate reality is to propose a comparison between simulated images and photographs of a real sample.
In this article, we will illustrate photograph-render comparison with a green car paint sample, already presented and discussed in a previous article. The characterization of this green car paint and the integration of measured data are presented in the same article here.
When doing photograph vs render comparison, several important conditions need to be fulfilled:
. Ensure similar light conditions,
. Ensure similar environment,
. Ensure similar observer.
A sample seen under two distinct lighting conditions may appear differently. This effect is illustrated in figure 1 where the car paint sample was photographed under two light conditions (the D65 standard illuminant, corresponding roughly to the average midday light and the A standard illuminant, corresponding to a typical, domestic lighting). The perception of the green color of the sample is different between those two photographs.
Figure 1 – Photographs of the green car sample under two light conditions (D65 illuminant on the left side and A illuminant on the right side)
The environment in which a sample is observed impact the perception of this sample. This effect is illustrated in figure 2 where simulated images with same light condition (D65 illuminant) are shown into two different environments (with grey walls and blue walls). The green car paint looks different between those two simulations.
Figure 2 – Car paint sample simulated with same light condition but with different environment (grey wall on the left image and blue wall on the right image)
Color perception is different depending on the person looking at the object. The most common example is colorblind people. In case of photograph-render comparison, having the same observer means to have the same camera setting between photographs and simulations, to apply the same post-treatment and to observe both images on the same screen. As an example, in figure 3, the same image is shown with two different camera setting. In this case, the ISO sensitivity (i.e. the camera sensor sensitivity to light) is different.
Figure 3 – Car paint sample simulated with two different ISO sensitivities (800 ISO on the left side and 200 ISO on the right side)
The lightbooth, shown in figure 4, is completely reconstructed in 3D model, as shown in figure 5.
Then, every materials as well as the light emission of the lightbooth are included into Ocean™ after being measured, as explained in the next section. The simulated lightbooth is shown in figure 6.
Figure 4 – Lightbooth photograph
Figure 5 – Lightbooth 3D model
Figure 6 – Simulated lightbooth
MATERIAL AND LIGHT SIMULATIONS
In order to correctly simulate the lighting and environment conditions corresponding to the physical lightbooth, the reflectance properties of each part of the lightbooth (wall, ground and celling) are measured using spectrophotometer at several angles. Those measurements are then imported into Ocean™.
The different between real measurements and virtual ones is determined in the CIELAB (or L*a*b*) color space. This color space is composed of : L* for perceptual lightness, and a* and b* for the four unique colors of human vision (red, green, blue, and yellow). Those parameters are determined for one given illuminant (D65 in this case) and depend on the human eye response and the material reflectance or transmittance spectrum.
Figure 7 – ΔE determined between the real and virtual measurements of the lightbooth ground is given for different angles
The relative perceptual differences between any two colors in L*a*b* can be approximated by the Euclidean distance ΔE of each color expressed as a point in a three-dimensional space L*, a*, b*. It is commonly set that no difference can be seen by eyes if ΔE < 3.
In figure 7, ΔE determined between the real and virtual measurements of the lightbooth ground are given for different angles and are always lower than 3, indicating that the lightbooth material simulation is correct.
The light spectrum and intensity emission are also measured using a luxmeter. A calibration is then made in order to estimate the emission shape and match the one obtains in the Ocean™ model.
Once the sample is placed in the lightbooth, with a stable lighting and environment, the next step is to take the picture of the sample and to make the comparison with simulations.
In order to have a good comparison between those two images, the real camera and the virtual one need to have the exact same setting. This means that the position of the camera (compared to the sample), the focal lens, the shutter speed and the ISO sensitivity have to be identical. In Ocean™, all those parameters can be implemented in an “Ideal perspective camera”.
Finally, the picture and simulation post-treatment need to be the same. That means that the white balance, the colorimetry space of observation and the global tone mapping need to be identical. The DNG camera profile of the camera is determined using X-rite color checker passport (shown in figure 8). A DNG camera profile is used to define the camera sensor behavior under a given illuminant. Thanks to this profile, the exposure, the color normalization, the white balance, and the chromatic variations are set and corrected.
Figure 8 – X-rite color checker used to define the DNG camera profile
EXAMPLE – CAR PAINT SAMPLE
The procedure presented in this article is applied on car paint sample. The result can be seen in figure 9. The question is : which is the real sample and the simulated one ?
Figure 9 – photograph vs simulation comparison of the green car paint sample
The photograph of sample is on the left side of figure 9, and the simulation on the right side. The green color is really similar between the two images. In order to be convince of this, the L*a*b* Euclidian difference (as presented in section “Material and light simulations”) is determined for different point of this images and given in figure 10. The Euclidean difference ΔE is lower than 3, indicated that no large difference can be seen between those two images.
Figure 10 – ΔE determined between the photograph and the simulated image, for different points indicated on the right image
The procedure of the comparison between photographs and simulations with Ocean™ is presented in this article. The example of a green car paint is provided and show good agreement in term of appearance and colorimetry between photograph and simulated image.
This comparison shows that the complete workflow, from measured properties to a complete and accurate simulation of a sample, is achievable with Ocean™ and appropriate material. This is a client service that can be proposed by Eclat Digital.