Introduction: Why Measuring Instruments Must “See Like Humans”#
The ultimate goal of optical metrology is not to measure the physical energy of light itself, but to measure human perception of light. For the same 1 watt of radiant power, green light at 555nm excites more than ten times the brightness perception in the human eye compared to red light at 650nm. If a luminance meter treated light of all wavelengths equally, its measured values would be seriously disconnected from the human visual experience.
For this reason, the hardware design of an imaging colorimeter is not aimed at “capturing as many photons as possible,” but at “accurately simulating the spectral response of the human eye” as its core principle. The mathematical foundation of this principle is the CIE 1931 standard colorimetric system, the key to engineering implementation is customized tristimulus filters, and the quantitative metric for evaluating the degree of matching is the f1’ error. This article will dissect this complete chain from physiological vision to optical engineering layer by layer.
Physical Meaning of the CIE 1931 Color Matching Functions#
From Cone Cells to the Standard Observer#
Three types of cone cells on the human retina—L-type (Long-wavelength, sensitive to red light), M-type (Medium-wavelength, sensitive to green light), and S-type (Short-wavelength, sensitive to blue light)—form the physiological basis of human color vision. Each type of cone cell has a different response sensitivity to different wavelengths in the visible spectrum (approximately 380nm to 780nm), a wavelength-dependent response characteristic known as the spectral sensitivity function.
In 1931, the International Commission on Illumination (CIE), based on statistical data from a large number of human color matching experiments, defined the three Color Matching Functions (CMFs) of the CIE 1931 Standard Colorimetric Observer: $\bar{x}(\lambda)$, $\bar{y}(\lambda)$, and $\bar{z}(\lambda)$.
These three functions are not directly identical to the spectral sensitivities of L, M, and S cone cells, but are mathematical equivalents obtained through linear transformation, designed to satisfy the following special constraints:
- $\bar{y}(\lambda)$ is identical to the photopic spectral luminous efficiency function $V(\lambda)$: That is, the Y tristimulus value directly corresponds to the luminance perceived by the human eye. $V(\lambda)$ peaks at 555nm, reflecting the physiological fact that the human eye is most sensitive to yellow-green light.
- All function values are non-negative: This is achieved through linear transformation from the original experimental data, making the tristimulus values XYZ always positive, which facilitates engineering calculations.
- Tristimulus values of equi-energy white light are equal: X = Y = Z, providing a symmetrical reference benchmark for the colorimetric coordinate system.
Calculation of Tristimulus Values#
For any light source with a Spectral Power Distribution (SPD) of $S(\lambda)$, its CIE 1931 tristimulus values are calculated through the following integrations:
$$X = K \int_{380}^{780} S(\lambda) \cdot \bar{x}(\lambda) \, d\lambda$$$$Y = K \int_{380}^{780} S(\lambda) \cdot \bar{y}(\lambda) \, d\lambda$$$$Z = K \int_{380}^{780} S(\lambda) \cdot \bar{z}(\lambda) \, d\lambda$$where $K$ is a normalization constant. The Y value is the luminance (in units of cd/m², after appropriate normalization). The CIE 1931 color coordinates (x, y) are defined by the ratios of the tristimulus values:
$$x = \frac{X}{X+Y+Z}, \quad y = \frac{Y}{X+Y+Z}$$The color coordinates (x, y) describe the Hue and Saturation of a color, independent of luminance. The color of any light source can be represented by a single point on the CIE 1931 chromaticity diagram.
Why Color Matching Functions are the Cornerstone of Colorimetry#
The above integrations express a core idea: to accurately determine the color of a light source, knowing its total luminous flux alone is not enough; one must know how the distribution of light energy across the visible spectrum interacts separately with the three human response channels.
There are two engineering paths to implement this integration:
Spectral Method: Uses a spectroradiometer to directly measure the complete spectral distribution $S(\lambda)$, and then calculates XYZ through numerical integration. This is the most accurate method, but the instruments are large, expensive, slow, and difficult to use for spatially resolved area measurements.
Filter Method: Designs three optical filters such that the combined spectral response of each filter and the sensor approximates $\bar{x}(\lambda)$, $\bar{y}(\lambda)$, and $\bar{z}(\lambda)$, respectively. The signals captured by the sensor through these three filters directly approximate the XYZ tristimulus values. This is the method used by imaging colorimeters, trading the spectral matching accuracy of the filters for the spatial resolution and speed advantages of area measurement.
Customized Filters: From CIE Curves to Physical Optical Components#
Engineering Challenges of Spectral Matching#
Transforming CIE color matching functions from mathematical curves into physical optical filters is the most technically difficult part of imaging colorimeter hardware design.
Taking the $\bar{x}(\lambda)$ function as an example, it presents a bimodal structure in the visible spectrum—with peaks at approximately 445nm (blue-violet region) and approximately 595nm (orange-red region). This means the corresponding filter must have high transmittance in these two wavelength bands and low transmittance in the green wavelength band between them. Such a non-monotonic spectral transmittance curve places extremely high demands on filter design and manufacturing.
Customized filters in imaging colorimeters are typically manufactured using Multilayer Dielectric Interference Coating technology. By alternately depositing dielectric thin film layers of high and low refractive indices (up to dozens of layers) on a glass substrate, the thickness and refractive index of each layer are precisely controlled. The interference effect of light is utilized to achieve selective transmission or reflection of specific wavelengths.
Luther-Ives Condition#
Theoretically, a perfect colorimeter requires that the spectral responsivities $S_1(\lambda)$, $S_2(\lambda)$, and $S_3(\lambda)$ of its three measurement channels (the product of filter transmittance, lens transmittance, and sensor quantum efficiency) be a linear combination of the CIE color matching functions $\bar{x}(\lambda)$, $\bar{y}(\lambda)$, and $\bar{z}(\lambda)$. This condition is known as the Luther-Ives Condition:
$$\begin{bmatrix} \bar{x}(\lambda) \ \bar{y}(\lambda) \ \bar{z}(\lambda) \end{bmatrix} = \mathbf{M} \cdot \begin{bmatrix} S_1(\lambda) \ S_2(\lambda) \ S_3(\lambda) \end{bmatrix}$$where $\mathbf{M}$ is a non-singular 3×3 constant matrix.
If the Luther-Ives condition is strictly satisfied, then for any light source with any spectral distribution, the signals from the instrument’s three channels can be accurately transformed into CIE XYZ tristimulus values through the matrix $\mathbf{M}$. In other words, a 3×3 linear matrix would be sufficient to achieve perfect color correction.
However, in engineering practice, due to filter manufacturing tolerances, residual response of silicon sensors in the near-infrared band (>700nm), lens dispersion, and the complex bimodal structure of the $\bar{x}(\lambda)$ function, strictly satisfying the Luther-Ives condition is physically impossible. There is always a deviation between the actual spectral response and the CIE curves—the magnitude of this deviation directly determines the upper limit of the instrument’s colorimetric measurement accuracy.
Design Strategies for High-End Filters#
To approximate the CIE curves as closely as possible, high-end imaging colorimeter filter designs typically adopt the following strategies:
Monochrome Sensor + Filter Wheel Architecture. Uses a high-sensitivity monochrome sensor with full spectral response, equipped with a rotating filter wheel at the front. Each filter is specifically responsible for one CIE channel. This architecture avoids the spectral crosstalk issues of RGB dye filters in Bayer arrays, and each pixel directly outputs a real measurement value for a single channel.
Near-Infrared (NIR) Blocking. Silicon-based sensors still have significant quantum efficiency in the near-infrared band above 700nm, while CIE color matching functions are near zero in this band. If near-infrared is not effectively suppressed, the infrared radiation received by the sensor will contaminate the colorimetric measurement results as a “false signal.” High-end filters typically integrate NIR blocking functionality to suppress transmittance above 700nm to an extremely low level.
Multi-Filter Combination Optimization. For the bimodal structure of $\bar{x}(\lambda)$, some systems adopt a four-filter solution—splitting $\bar{x}(\lambda)$ into short-wavelength and long-wavelength sub-channels for separate measurement, then synthesizing them in software. Although this increases acquisition time, it significantly reduces the difficulty of filter design and improves matching accuracy.
f1’ Error: The Core Metric for Evaluating Spectral Matching Quality#
Definition and Calculation of f1'#
The spectral mismatch error $f_1'$ is an international standard metric for quantifying the degree of deviation between an instrument’s spectral response and the CIE standard functions, derived from CIE Publication 69 and related IEC standards. The calculation method for $f_1'$ is as follows:
First, perform the best linear fit between the instrument’s actual relative spectral responsivity $s_{rel}(\lambda)$ and the target CIE function (taking $V(\lambda)$ corresponding to the Y channel as an example) to obtain the normalized $s^*_{rel}(\lambda)$. Then calculate:
$$f_1' = \frac{\int_{380}^{780} |s^*_{rel}(\lambda) - V(\lambda)| \, d\lambda}{\int_{380}^{780} V(\lambda) \, d\lambda} imes 100\%$$A smaller $f_1'$ value indicates more accurate spectral matching.
Engineering Significance of f1’ Values#
The $f_1'$ value directly affects the colorimetric measurement accuracy of the instrument, especially when measuring narrowband light sources:
Broadband Light Sources (e.g., Incandescent Lamps, D65 Simulators): Since the spectral energy distribution is smooth and continuous, positive and negative deviations of the filter at certain wavelengths tend to cancel each other out during the integration process. Therefore, even if the $f_1'$ value is not ideal, the colorimetric error under broadband light sources may be small.
Narrowband Light Sources (e.g., Monochromatic LEDs, RGB Sub-pixels of OLEDs, Lasers): Spectral energy is concentrated within a narrow wavelength band, and the matching deviation of the filter at that specific band will be completely “exposed” and cannot be cancelled out through integration. Therefore, the $f_1'$ value is decisive for the measurement accuracy of narrowband light sources. In OLED and LED display inspection scenarios, requirements for $f_1'$ are particularly strict.
f1’ Levels of Different Grade Instruments#
According to CIE and related standards, photometers can be classified into different accuracy grades based on $f_1'$ values:
| Instrument Grade | Typical f1’ Range | Representative Applications |
|---|---|---|
| High-Precision Imaging Colorimeter | < 1.5% | Display panel line inspection, precise lab measurement |
| General Grade Imaging Colorimeter | 1.5% - 3% | General industrial optical inspection |
| Calibrated RGB Industrial Camera | 5% - 15% | Inline Pass/Fail determination, low-precision color screening |
Note that $f_1'$ only describes the spectral matching quality of a single channel (usually the Y channel). Complete colorimetric measurement accuracy also depends on the matching quality of the X and Z channels, as well as the degree of crosstalk between channels.
Tristimulus Filter-based vs. RGB-based Imaging Colorimeters: The Origin of the Accuracy Gap#
Working Principle of RGB-based Imaging Colorimeters#
RGB-based imaging colorimeters use industrial cameras (or consumer cameras) equipped with Bayer color filter arrays, where the spectral responses of the R, G, and B channels are determined by dye filters on the sensor. These dye filters are designed to generate visually “natural” color images, rather than accurately simulating CIE color matching functions.
RGB-based systems convert device-dependent RGB values into CIE XYZ values through mathematical calibration (Color Calibration/Characterization). The core of calibration is establishing a mathematical mapping model from RGB to XYZ—ranging from simple 3×3 linear matrices to high-order polynomial regression or root-polynomial regression models.
Physical Roots of the Accuracy Difference#
The accuracy difference between tristimulus filter-based and RGB-based imaging colorimeters is not just a matter of “good vs. bad calibration algorithms,” but stems from a fundamental physical fact: the spectral response shapes of RGB dye filters deviate too much from the CIE curves, beyond what can be corrected by linear transformation.
Specifically:
Spectral Overlap. There is significant spectral overlap between the R, G, and B filters in a Bayer array—for example, the green filter still has significant transmittance in the red and blue wavelength bands. This crosstalk means that signals from other colors are mixed into a single channel’s signal, reducing the signal-to-noise ratio for color differentiation.
Spectral Shape Mismatch. The shapes of the spectral transmittance curves of RGB filters (usually a single-peak structure approximating a Gaussian distribution) differ systematically from the shapes of the CIE curves. In particular, the bimodal structure of $\bar{x}(\lambda)$ has no corresponding channel in an RGB system to match directly.
Metamerism Issues. This is the most fatal limitation of RGB-based systems. If two light sources appear the same color to the human eye (i.e., they have the same XYZ tristimulus values) but have different spectral power distributions, they are called a metameric pair. For an instrument satisfying the Luther-Ives condition, the measurement results for a metameric pair are also identical (because XYZ are identical). However, for an RGB-based system, since its channel responses do not match the CIE curves, the same pair of metameric light sources may produce different RGB readings, leading to deviations in colorimetric calculation results. Calibration algorithms can only optimize the mapping relationship for known spectral types in the training set; when facing spectral distributions outside the training set, metameric errors are unpredictable.
Quantitative Comparison#
In actual measurement, the accuracy difference between the two systems in typical display measurement scenarios is usually reflected in the following metrics:
| Accuracy Metric | Tristimulus Filter-based Imaging Colorimeter | Calibrated RGB-based System |
|---|---|---|
| Y Channel f1' | < 1.5% | 5% - 15% |
| Broadband White Light Δu’v' | < 0.002 | 0.003 - 0.010 |
| Narrowband LED Δu’v' | < 0.005 | 0.010 - 0.030 (Highly dependent on calibration conditions) |
| Metamerism Resistance | High (Suppressed at hardware level) | Low (Cannot be fundamentally cured by software) |
While RGB-based systems using advanced calibration algorithms (such as root-polynomial regression) can significantly narrow the accuracy gap with tristimulus filter-based systems under specific conditions, this accuracy is conditional—it highly depends on the spectral similarity between the standard light source used for calibration and the actual object being measured. Once the spectral characteristics of the measured light source deviate from the calibration conditions, accuracy will rapidly degrade.
Beyond Three Channels: Four-Color Matrix and Spectral Mismatch Correction#
Four-Color Matrix Method for Specific Light Sources#
Even for high-end tristimulus filter-based imaging colorimeters, residual errors caused by $f_1'$ may still not meet the strict production line accuracy requirements when measuring specific types of narrowband light sources (such as the RGB sub-pixels of a particular model of OLED panel).
The Four-Color Matrix Method provides a targeted solution. The specific process is as follows: the display being tested is made to show four pure color screens—red, green, blue, and white; the imaging colorimeter captures these four screens and extracts channel signal values, while a high-precision spectroradiometer (such as Konica Minolta CS-2000) measures the reference XYZ values of the same area. Then, a 3×3 correction matrix dedicated to that display model is calculated, ensuring the imaging colorimeter’s output is accurately aligned with the spectroradiometer’s reference values.
The essence of this method is: giving up the pursuit of universal accuracy for all light sources, and instead targeting objects of specific spectral types to eliminate the influence of $f_1'$ under those specific spectra through the combination of “known spectrum + reference value.” On display panel production lines, the spectral characteristics of the same panel model are highly consistent, so the four-color matrix method can compress inter-instrument consistency errors to extremely low levels.
Spectral Mismatch Correction Factor#
For measured light sources with known spectral distributions, systematic errors caused by $f_1'$ can also be corrected by calculating the Spectral Mismatch Correction Factor (SMCF). This method requires prior knowledge of the spectral distribution of the measured light source and the actual spectral response of each instrument channel. Correction coefficients are derived through numerical calculation and applied to compensate for the results after measurement.
Conclusion: Hardware Accuracy is the Upper Limit for Algorithmic Accuracy#
In the overall accuracy chain of an imaging colorimeter, the spectral matching quality of tristimulus filters is at the most fundamental and critical position. Subsequent calibration algorithms—whether linear matrices, polynomial regression, or four-color matrices—are essentially “refining” and “compensating” on top of the accuracy foundation provided by the filter hardware.
A filter system with an $f_1'$ value of 1% and one with an $f_1'$ value of 10% will have an order of magnitude difference in final colorimetric accuracy even with identical calibration algorithms—because algorithms can only correct systematic deviations under known conditions and cannot compensate for information loss inherent in the hardware (especially unpredictable errors caused by metamerism).
Understanding this basic principle that “hardware accuracy is the upper limit for algorithmic accuracy” is instructive for correctly selecting imaging colorimeters, setting reasonable accuracy expectations, and designing effective calibration solutions in engineering practice.
FAQ#
Q1: Why is the f1’ value particularly important when measuring OLEDs and LEDs?#
Because OLEDs and LEDs are narrowband light sources with spectral energy concentrated in narrow wavelength bands. Matching deviations of the filter at those specific bands will be completely “exposed” and cannot be compensated for by the positive and negative deviation cancellation effect in broadband spectral integration. Therefore, the f1’ value is decisive for the measurement accuracy of narrowband light sources. For OLED and LED display inspection, high-precision filter systems with f1’ < 1.5% are recommended.
Q2: What is the Four-Color Matrix method? How does it differ from factory calibration?#
The Four-Color Matrix method is a user-level correction method for specific display types. The display under test shows four pure color screens—red, green, blue, and white—which are measured by both an imaging colorimeter and a high-precision spectroradiometer to calculate a 3×3 correction matrix dedicated to that display model. Unlike the “universal benchmark” of factory Illuminant A calibration, the Four-Color Matrix method sacrifices universal accuracy to achieve extremely high measurement consistency for specific spectral types, making it particularly suitable for batch inspection of the same panel model on production lines.
Q3: Can a calibrated RGB industrial camera replace a tristimulus filter-based imaging colorimeter?#
In some scenarios, yes, but there are fundamental limitations. RGB cameras map RGB values to CIE XYZ through mathematical calibration, achieving acceptable accuracy when calibration conditions match. However, since RGB filters do not satisfy the Luther-Ives condition, metameric errors are unpredictable when facing spectral distributions outside the calibration training set. For high-precision colorimetric measurement requiring Δu’v’ < 0.005, or scenarios requiring measurement of many different spectral types of light sources, tristimulus filter-based imaging colorimeters remain the irreplaceable choice.
This article is part of the Imaging Colorimeter Technology Knowledge Base series.