同色异谱：不同光谱，同一颜色Metamerism: different spectra, the same color
人眼把无穷维光谱压缩成三维 XYZ。所以无穷多种"不一样的光"，可以在眼里变成"同一个颜色"。这是色彩管理一切难题的根，也是为什么手机屏看色卡总差一点、印刷打样到办公室就变色、不同人看同一台 OLED 颜色不一致。The eye compresses an infinite-dimensional spectrum into three-dimensional XYZ. So infinitely many "different lights" can become "the same color" to the eye. This is the root of every color-management problem — and why a color chart never quite looks right on a phone, why a print proof shifts color once it reaches the office, and why different people see the same OLED differently.
1. 问题场景：你已经被它影响了1. Problem scenarios: it's already affecting you
下面四个困惑，本质都是同色异谱的不同侧面。先认出它们，再回头看原理就会顺很多。The four confusions below are all different faces of metamerism. Recognize them first, and the theory afterward goes down much easier.
跨屏不一致Cross-screen inconsistency
同一张图在 iPhone OLED 和 MacBook LCD 上，颜色就是不一样——而且不是简单的色温偏移。The same image just looks different on an iPhone OLED versus a MacBook LCD — and it isn't a simple color-temperature shift.
打样变色Proof shifts color
印刷打样在 D50 灯箱里完美还原，搬到办公室白炽灯下样张就和原稿对不上了。A print proof matches perfectly in a D50 light box, then no longer matches the original once you move it under office incandescent light.
校色仪 vs 眼睛Meter vs. eye
校色仪测出屏幕白点是 D65，同事看觉得偏粉、自己看觉得偏绿，没有人是错的。A colorimeter reads the screen's white point as D65; a colleague sees it as pinkish, you see it as greenish, and nobody is wrong.
相机色卡校不齐Camera chart won't line up
用 ColorChecker 拉 IDT 后，色卡内的方块都还原得很准，拍其他物体却总有些颜色对不上。After building an IDT from a ColorChecker, the chart's own patches reproduce accurately, yet shooting other objects, some colors are always off.
2. 核心直觉：颜色是光谱的简化结果2. Core intuition: color is a simplification of a spectrum
同色异谱并不是一个孤立术语。它来自一条很简单的链路：物体或屏幕发出一段光谱，人眼只保留其中能刺激三类视锥细胞的部分，最后大脑得到一个颜色感受。Metamerism isn't an isolated term. It comes from a simple chain: an object or screen emits a spectrum, the eye keeps only the part that stimulates its three cone types, and the brain ends up with a single color sensation.
光谱 A：连续、宽峰Spectrum A: continuous, broad peaks
例如日光、荧光粉背光或某些反射材料。它在很多波长上都有能量，形状比较平滑。Like daylight, a phosphor backlight, or some reflective materials. It has energy across many wavelengths and a fairly smooth shape.
光谱 B：窄峰、离散Spectrum B: narrow, discrete peaks
例如 OLED、量子点或激光原色组合。它只在少数波长附近有明显能量。Like an OLED, quantum dots, or a laser primary combination. It only has notable energy near a few wavelengths.
这两条光谱在物理上并不相同，但如果它们让 L、M、S 三类视锥细胞产生相同响应，人眼就会把它们归为同一个颜色。色彩管理里的 XYZ，就是把这种三类响应整理成标准化的 3 个数。These two spectra aren't physically the same, but if they make the L, M, and S cones respond identically, the eye treats them as one color. XYZ in color management is just those three responses tidied into three standardized numbers.
真实世界先给出光谱The real world gives a spectrum
光谱可以很复杂：可能来自太阳、LED、OLED 像素、油墨反射或金属漆反射。The spectrum can be complex: from the sun, an LED, an OLED pixel, ink reflection, or metallic-paint reflection.
眼睛只采样三类响应The eye samples just three responses
L、M、S 三类视锥对不同波长敏感。它们不会记录完整光谱，只输出三类响应强度。The L, M, and S cones are sensitive to different wavelengths. They don't record the full spectrum — only three response strengths.
XYZ 是标准化记录XYZ is the standardized record
CIE 色匹配函数把标准观察者的三类响应整理成 X、Y、Z，方便设备和软件计算颜色。The CIE color-matching functions turn the standard observer's three responses into X, Y, Z so devices and software can compute color.
信息丢失导致同色异谱Lost information causes metamerism
从完整光谱到 3 个数的过程中，很多细节被丢掉。因此不同光谱可以得到同一组 XYZ。Going from a full spectrum to three numbers throws away a lot of detail. So different spectra can yield the same XYZ.
3. 基础定义3. Basic definitions
四个术语撑起了同色异谱的整个理论框架。先把它们的边界划清楚，下面所有讨论都建立在这之上。Four terms hold up the whole theoretical frame of metamerism. Pin down their boundaries first; everything below builds on them.
三刺激值与色匹配函数Tristimulus values and color-matching functions
三刺激值 (X, Y, Z) 是把光谱 S(λ) 与三条标准函数 x̄(λ)、ȳ(λ)、z̄(λ) 逐波长相乘再积分得到的 3 个数。这三条函数叫色匹配函数 (CMF)，是 1931 年 CIE 根据多人色觉实验拟合出来的"标准人眼"。The tristimulus values (X, Y, Z) are three numbers obtained by multiplying the spectrum S(λ) by three standard functions x̄(λ), ȳ(λ), z̄(λ) wavelength by wavelength and integrating. Those three functions are the color-matching functions (CMF) — the "standard human eye" the CIE fit from many people's color-vision experiments in 1931.
常用观察者：CIE 1931 2°（小色块匹配，目前 RGB/印刷链路默认）、CIE 1964 10°（大色块匹配，工业色差常用）、CIE 2015 生理观察者（按年龄/视场参数化）。Common observers: CIE 1931 2° (small-field matching, today's default in RGB/print chains), CIE 1964 10° (large-field matching, common for industrial color difference), and the CIE 2015 physiological observer (parameterized by age/field size).
同色异谱对Metameric pair
当两条光谱形状不同但计算出的 XYZ 完全相等时，它们就是一对同色异谱（metameric pair）。在指定观察者和指定光源下，眼睛把它们视作同一颜色。When two spectra have different shapes but compute to exactly equal XYZ, they form a metameric pair. Under a given observer and a given illuminant, the eye treats them as one color.
关键限定：同一观察者 + 同一光源。只要其中一个条件改变，原本的匹配就可能失效——这就是后面要讲的"失败"。The key qualifier: same observer + same illuminant. Change either condition and the match can break — that's the "failure" discussed later.
同色异谱黑Metameric black
一种非零光谱，但它的 X、Y、Z 三个积分都精确为零。把它叠加到任何光谱上，颜色不变——这就是为什么同色异谱必然存在：CMF 的"零空间"里有无穷多种可加光谱。A non-zero spectrum whose X, Y, and Z integrals are all exactly zero. Add it to any spectrum and the color is unchanged — which is why metamerism must exist: the "null space" of the CMF holds infinitely many spectra you can add.
直观理解：metameric black 必然有正有负的波段（物理光谱非负，所以它只能以两条光谱的"差"的形式存在）。任何 XYZ 相同的两条光谱，差值就是一条 metameric black。Intuitively, a metameric black must have both positive and negative bands (physical spectra are non-negative, so it can only exist as the "difference" of two spectra). The difference of any two spectra with equal XYZ is a metameric black.
Luther 条件The Luther condition
一台三通道设备（相机、扫描仪、滤色片式色度计）要做到colorimetric——即"和人眼完全一致地判断颜色相同"——它的光谱灵敏度必须是 CMF 的线性变换。换言之：每条 RGB 通道响应，要能写成 x̄/ȳ/z̄ 的某种加权和。For a three-channel device (camera, scanner, filter-based colorimeter) to be colorimetric — i.e. to judge colors identical exactly as the eye does — its spectral sensitivities must be a linear transform of the CMF. In other words, each RGB channel response must be expressible as some weighted sum of x̄/ȳ/z̄.
现实里几乎没有相机精确满足。相机的 RGB 滤色片需要同时兼顾噪声、灵敏度、红外抑制和制造一致性，通常只能接近人眼三刺激值。白平衡和矩阵校正能修正灰轴与一部分色卡，但不能把所有光谱都变成人眼会看到的颜色。In reality almost no camera satisfies it exactly. A camera's RGB filters must balance noise, sensitivity, IR rejection, and manufacturing consistency, so they can only approximate the eye's tristimulus values. White balance and matrix correction can fix the gray axis and part of a chart, but can't turn every spectrum into the color the eye would see.
CIE 1931 和 CIE 2015 到底差在哪What actually differs between CIE 1931 and CIE 2015
先说结论：CIE 2015 不是说 1931 错了，也不是要把现有 RGB/XYZ 体系推翻。它更像是把“标准观察者”从一个固定模板，推进成了一个更接近真实生理差异的模型，所以它对解释为什么同一台窄峰显示器会让不同人看到不同白点更有帮助。The takeaway first: CIE 2015 doesn't say 1931 was wrong, nor does it overturn the existing RGB/XYZ system. It pushes the "standard observer" from a fixed template toward a model closer to real physiological variation, so it's more useful for explaining why the same narrow-band display makes different people see different white points.
历史上最重要，也最方便工程统一Historically the most important, and the easiest to standardize on
1931 模型来自颜色匹配实验，核心价值是给行业一个统一坐标系。今天的 XYZ、xy、Lab、ICC、Rec.709/P3/Rec.2020 白点与色域表达，本质上都还建立在这个共同参考上。The 1931 model came from color-matching experiments; its core value is giving the industry one shared coordinate system. Today's XYZ, xy, Lab, ICC, and Rec.709/P3/Rec.2020 white-point and gamut notation are all still built on this common reference.
把“标准观察者”往真实人眼又推近了一步Nudging the "standard observer" one step closer to a real eye
2015 模型建立在锥细胞光谱响应、生理透镜与黄斑吸收数据之上，可以按视场大小、年龄等参数生成观察者函数。它不是只问“平均人怎么配色”，而是更接近“不同类型的人可能怎么响应”。The 2015 model is built on cone spectral responses and physiological lens/macular absorption data, and can generate observer functions parameterized by field size, age, and more. It asks not just "how does the average person match color" but closer to "how might different types of people respond".
4. 五种同色异谱失败4. Five kinds of metameric failure
同色异谱本身不是"故障"——它是色彩系统能存在的前提。真正的问题是条件改变时同色异谱匹配失效。CIE/ISO 把这种失效按触发条件分成 4 类，加上相机/扫描仪场景一共 5 类。Metamerism itself isn't a "fault" — it's the precondition that lets a color system exist at all. The real problem is a metameric match failing when conditions change. CIE/ISO sort this failure into 4 types by trigger, plus the camera/scanner case for 5 in total.
5. 为什么新显示设备更容易同色异谱失效5. Why newer displays fail metamerism more easily
现代显示设备为了更高亮度、更大色域和更高效率，越来越依赖窄峰或多峰光谱。问题在于：XYZ 可以被校准到同一个白点，但光谱形状越尖、越不连续，不同观察者之间的差异就越容易被放大。For higher brightness, wider gamut, and better efficiency, modern displays increasingly rely on narrow-band or multi-peak spectra. The catch: XYZ can be calibrated to the same white point, but the sharper and more discontinuous the spectrum, the more easily differences between observers are amplified.
光谱较宽，观察者差异相对温和Broader spectrum, milder observer differences
白 LED + 彩色滤光片或荧光粉背光的发射峰通常较宽。它不一定色域最大，但不同人的锥细胞差异被宽光谱“平均”了一部分。A white LED + color filters, or a phosphor backlight, usually has broad emission peaks. The gamut may not be the largest, but a broad spectrum partly "averages out" different people's cone differences.
宽峰 / 中等色域Broad peaks / medium gamut量子点把红绿推成窄峰Quantum dots push red and green into narrow peaks
量子点能做出很纯的红、绿原色，提升 P3 / Rec.2020 覆盖率。但窄峰刚好落在 L/M 锥响应变化最敏感的区域，不同观察者的白点和饱和色感知会更容易分开。Quantum dots make very pure red and green primaries, boosting P3 / Rec.2020 coverage. But those narrow peaks land right where the L/M cone responses change most steeply, so different observers' white-point and saturated-color perceptions split apart more easily.
窄峰 / 高色域 / 高风险Narrow peaks / wide gamut / high risk越接近单色光，越考验“标准观察者”The closer to monochromatic, the more it tests the "standard observer"
激光和部分窄带 OLED 的原色更接近线状光谱。校准仪可以把白点调到 D65，但真实观众不是同一个标准观察者，所以有人看偏绿、有人看偏粉。The primaries of lasers and some narrow-band OLEDs are closer to line spectra. A meter can set the white point to D65, but real viewers aren't one single standard observer, so some see it green-ish and others pink-ish.
极窄峰 / 观察者同色异谱Very narrow peaks / observer metamerism为什么 HDR 会把问题放大Why HDR amplifies the problem
HDR 不只是亮度更高，也常常伴随更广色域、更高饱和度和更强的显示原色纯度。饱和色越靠近显示器原色，颜色越依赖那几个窄峰，观察者同色异谱就越明显。HDR isn't just higher brightness — it usually comes with wider gamut, higher saturation, and purer display primaries. The closer a saturated color is to a display primary, the more it depends on those few narrow peaks, and the more visible observer metamerism becomes.
这也是为什么同一个 HDR 片段在不同 OLED、Mini LED + 量子点电视和参考监视器上，即使测量白点相近，肉眼观感仍可能不一致。This is also why the same HDR clip can look inconsistent to the eye across different OLEDs, Mini-LED + quantum-dot TVs, and reference monitors, even when their measured white points are close.
校准为什么不能彻底解决Why calibration can't fully fix it
常规校准把显示器对齐到一个标准观察者和一个目标白点，例如 CIE 1931 2° 的 D65。它能减少设备误差，但不能消除人眼个体差异。Routine calibration aligns a display to one standard observer and one target white point, e.g. D65 under CIE 1931 2°. It reduces device error but can't eliminate individual differences between eyes.
更严谨的做法是：关键监看使用同型号参考设备；避免用单一消费屏判断最终颜色；对品牌色、肤色、霓虹色等敏感内容做多设备和多人交叉检查。CIE 2015 这类生理观察者模型的价值，不是替代校准，而是帮助我们更诚实地描述“校准后的剩余差异”。A more rigorous practice: use the same reference model for critical monitoring; avoid judging final color on a single consumer screen; cross-check sensitive content like brand colors, skin tones, and neon across multiple devices and multiple people. The value of physiological-observer models like CIE 2015 isn't to replace calibration, but to help us describe the "residual difference after calibration" more honestly.
为什么相机拍屏和人眼观察会差很多Why photographing a screen differs so much from viewing it
显示器校准的目标是让人眼看到目标颜色：在指定观察者模型下得到正确的 XYZ / Lab。相机拍屏时，光没有先经过人眼，而是进入相机自己的 RGB 滤色片。只要相机的三条光谱灵敏度不满足 Luther 条件，它就会把同一条显示光谱压缩成另一组 RGB。The goal of display calibration is for the eye to see the target color: correct XYZ / Lab under a given observer model. When a camera shoots a screen, the light doesn't pass through an eye first — it enters the camera's own RGB filters. As long as the camera's three spectral sensitivities don't satisfy the Luther condition, it collapses the same display spectrum into a different set of RGB.
传统宽峰 LCD 的光谱较平滑，相机误差有时不明显；OLED、激光投影和 RGB mini LED 的原色更窄，光谱能量集中在少数波段。窄峰越多，相机响应与人眼 CMF 的偏差越容易被放大，所以拍出来可能偏绿、偏粉、肤色不稳，或者不同屏幕在照片里白点差异很大。Broad-peak LCDs have smoother spectra, so camera error is sometimes negligible; OLED, laser projection, and RGB mini-LED have narrower primaries with energy concentrated in a few bands. The more narrow peaks there are, the more the deviation between the camera response and the eye's CMF is amplified — so shots can look green-ish, pink-ish, have unstable skin tones, or show big white-point differences between screens.
判断显示器颜色时应以人眼观察和仪器测量为准。相机画面只是“这台相机对这台显示器光谱的反应”，不是显示器视觉效果本身。除非为特定相机、镜头、曝光和显示器光谱建立专门校正，否则拍屏不能作为颜色准确性的依据。Judge display color by eye observation and instrument measurement. A camera image is only "this camera's reaction to this display's spectrum", not the display's visual result itself. Unless you build a dedicated correction for a specific camera, lens, exposure, and display spectrum, a screen photo can't be evidence of color accuracy.
三个典型设备案例Three typical device cases
- 工程取舍：Engineering trade-off: 更窄的光谱峰带来更大色域和更高效率，但也更容易暴露“标准观察者 ≠ 每个真实观察者”的问题。narrower spectral peaks bring wider gamut and higher efficiency, but also more readily expose the problem that "the standard observer ≠ every real observer".
- 内容制作：Content production: 不要把“校色仪读数一致”直接等同于“所有人看起来一致”。HDR 广色域母版尤其需要参考环境和参考显示器。don't equate "the meter reads the same" with "everyone sees the same". Wide-gamut HDR masters especially need a reference environment and a reference display.
- 未来方向：Future directions: 多原色显示、CIE 2015 生理观察者、按年龄/视场建模的观察者补偿，都在试图降低这种差异。multi-primary displays, the CIE 2015 physiological observer, and observer compensation modeled by age/field size are all attempts to reduce this difference.
6. 演示 A：两条不同光谱，同一个颜色6. Demo A: two different spectra, one color
左边是窄峰光谱（类似 OLED / 激光投影原色），右边是宽峰光谱（类似传统 LCD 荧光粉）。调整两组峰的强度，让两侧的 XYZ 三刺激值尽可能接近。当 ΔE 落到 1 以下，人眼就会把两者视为同一颜色——尽管它们的光谱形状完全不一样。On the left is a narrow-band spectrum (like OLED / laser-projection primaries); on the right a broad-band spectrum (like traditional LCD phosphors). Adjust the peak strengths to bring the two sides' XYZ tristimulus values as close as possible. Once ΔE drops below 1, the eye treats them as one color — even though their spectral shapes are completely different.
光谱 A · 窄峰Spectrum A · narrow peaks
光谱 B · 宽峰Spectrum B · broad peaks
你刚才看到了什么What you just saw
左右两条光谱在 380–780 nm 上形状完全不同——一边是三个尖锐的窄峰，一边是三个平滑的宽峰。但当 ΔE 小于 1 时，人眼无法区分这两种光。把这两块色卡贴在一起，看到的是同一种颜色。The two spectra have completely different shapes across 380–780 nm — three sharp narrow peaks on one side, three smooth broad peaks on the other. But once ΔE is below 1, the eye can't tell them apart. Place the two swatches side by side and you see one color.
匹配按钮不是保证成功的“自动调色”。它只允许三个峰做 0–100 的非负加法；如果精确解需要负峰值，或需要把某个峰推到 100 以上，就只能给出最接近的结果。The match button isn't a guaranteed "auto-grade". It only allows non-negative addition of the three peaks over 0–100; if the exact solution needs a negative peak, or a peak pushed above 100, it can only give the closest result.
为什么会这样Why this happens
视网膜上只有 3 种锥细胞（L、M、S），任何光谱进入眼睛后都会被压缩成 3 个数字（XYZ 三刺激值）。无穷多种光谱可以产生同一组三刺激值——它们就是同色异谱对（metameric pair）。这是色彩管理的起点。The retina has only 3 cone types (L, M, S), so any spectrum entering the eye is compressed into 3 numbers (the XYZ tristimulus values). Infinitely many spectra can produce the same tristimulus values — they form a metameric pair. This is the starting point of color management.
7. 演示 B：换光源，匹配失效7. Demo B: change the illuminant, the match breaks
两个反射率不同的样本，可以被设计成在 D65 下完全匹配。但只要换一个光源（白炽灯 / 荧光 / LED），同色异谱匹配就会失效——这就是光源同色异谱失败，也是印刷打样中最常见的问题之一。Two samples with different reflectances can be designed to match exactly under D65. But change the illuminant (incandescent / fluorescent / LED) and the metameric match breaks — this is illuminant metameric failure, one of the most common problems in print proofing.
反射率 + 光源 SPDReflectance + illuminant SPD
D65 日光虚线 = 光源 SPD（按峰值归一化）；实线 = 两个样本的反射率曲线。眼睛看到的 = SPD × 反射率，再经 CMF 积分。Dashed = illuminant SPD (normalized to its peak); solid = the two samples' reflectance curves. What the eye sees = SPD × reflectance, then integrated against the CMF.
样本 B 反射率峰值Sample B reflectance peaks
3 个高斯峰（FWHM ≈ 65 nm）3 Gaussian peaks (FWHM ≈ 65 nm)典型操作序列A typical sequence
1. 点击"在 D65 下匹配 A"——样本 B 的三个峰自动调整，让两块色卡在 D65 下完全一致（ΔE < 1）。2. 依次切换到 D50、A、F2、LED——观察 ΔE 如何变化。3. 在 A 和 F2 下 ΔE 会显著上升，因为这些光源的 SPD 形状差异很大。1. Click "Match A under D65" — sample B's three peaks auto-adjust so the two swatches are identical under D65 (ΔE < 1). 2. Switch through D50, A, F2, LED in turn and watch how ΔE changes. 3. Under A and F2, ΔE rises sharply, because those illuminants have very different SPD shapes.
为什么 F2 特别凶猛Why F2 is especially brutal
D65 是平滑日光，F2 是带尖锐汞蒸气发射线的荧光（405/436/546 nm 等强线）。F2 在不同波长上能量分布完全不均，对样本反射率的某些波段格外敏感。任何在 D65 下匹配但依赖不同波段反射的两个样本，在 F2 下都很容易产生可见差异。D65 is smooth daylight; F2 is fluorescent with sharp mercury emission lines (strong lines at 405/436/546 nm, etc.). F2's energy is very unevenly distributed across wavelengths, making it especially sensitive to certain bands of a sample's reflectance. Any two samples that match under D65 but rely on different reflectance bands easily show a visible difference under F2.
8. 工作流落地8. Putting it into the workflow
同色异谱不是抽象概念——它每天都在你的工作链路里制造问题。下面四个场景是高频实例。Metamerism isn't an abstract idea — it causes problems in your workflow every day. The four scenarios below are frequent, concrete examples.
相机 IDT 永远不完美A camera IDT is never perfect
相机用自己的光谱灵敏度替代了人眼 CMF——而几乎没有相机精确满足 Luther 条件。所以 ARRI / Sony / Canon 的 IDT（Input Device Transform）只是最小二乘最优解，不是数学完美匹配。A camera substitutes its own spectral sensitivities for the eye's CMF — and almost no camera satisfies the Luther condition exactly. So an ARRI / Sony / Canon IDT (Input Device Transform) is only a least-squares best fit, not a mathematically perfect match.
用 ColorChecker 校准只优化了色卡内的 24 块，对色卡光谱之外的物体不保证。荧光物体、窄峰染料（如紫色染料、某些霓虹色）尤其容易出现明显偏差。Calibrating with a ColorChecker only optimizes the chart's 24 patches and guarantees nothing for objects outside the chart's spectra. Fluorescent objects and narrow-band dyes (purple dyes, some neon colors) are especially prone to visible error.
拍摄显示器也是同一个问题：OLED、激光投影、RGB mini LED 等窄峰光源可能让相机拍出明显偏色，但这不等于人眼看到的显示效果错误。对策：拍屏只作记录，颜色判断以受控环境下的人眼观察、分光测量和目标交付标准为准。Shooting a display is the same problem: narrow-band sources like OLED, laser projection, and RGB mini-LED can make the camera record an obvious cast, but that doesn't mean the display looks wrong to the eye. The fix: treat screen photos as records only, and base color judgments on eye observation in a controlled environment, spectral measurement, and the target delivery standard.
D50 标准观察箱为什么必须用Why the D50 standard viewing booth is mandatory
印刷颜色由油墨光谱 × 光源 SPD 决定。墨厂调出的"匹配色"是在指定光源下匹配。如果你在普通办公室灯下看打样，看到的就是不同光源下的同色异谱失败。Print color is set by ink spectrum × illuminant SPD. A "matched color" mixed at the ink house matches under a specified illuminant. View the proof under ordinary office lighting and what you see is illuminant metameric failure.
图文印刷国际标准光源是 D50（ISO 3664：2009）。墨样、打样、看样都必须在 D50 灯箱里进行。带荧光增白剂（OBA）的纸张还要考虑 UV 容差。The international standard illuminant for graphic-arts printing is D50 (ISO 3664:2009). Ink draw-downs, proofs, and press checks must all happen in a D50 light box. Papers with optical brightening agents (OBA) also need UV tolerance considered.
电视/视频制作改用 D65，是因为目标显示器是 D65 白点——保持评估光源与最终显示光源一致。TV/video production uses D65 instead, because the target display has a D65 white point — keeping the evaluation light consistent with the final display light.
OLED / QD-OLED 显示中的观察者差异Observer differences on OLED / QD-OLED displays
传统 LCD 用宽峰荧光粉，不同观察者看到的颜色差异较小。OLED 和 QD-OLED 用窄峰发光体（OLED 的窄带蓝、量子点的极窄绿/红），同样的 XYZ 在不同人眼里能产生 ΔE 5–18 的差距（Optics Express 2020 等多项实测）。Traditional LCDs use broad-peak phosphors, so color differences between observers are small. OLED and QD-OLED use narrow-band emitters (OLED's narrow-band blue, quantum dots' very narrow green/red), and the same XYZ can produce ΔE differences of 5–18 between different eyes (measured in Optics Express 2020 and others).
这就是为什么"有人觉得新 iPhone 偏绿、有人觉得正常"——校色仪用的是 CIE 1931 标准观察者，你的眼睛不是。色域越宽、原色越窄，问题越严重。This is why "some people think the new iPhone looks green and others think it's fine" — the meter uses the CIE 1931 standard observer, and your eye isn't it. The wider the gamut and the narrower the primaries, the worse it gets.
方向：CIE 2015 生理观察者模型、Asano 八参数个体观察者、显示器自适应观察者补偿——业界还在探索。Directions: the CIE 2015 physiological observer, Asano's eight-parameter individual observer, and display-adaptive observer compensation — the industry is still exploring.
Rec.2020 / 广色域的光谱限制The spectral limits of Rec.2020 / wide gamut
Rec.2020 用接近单色光的 RGB 原色（630 / 532 / 467 nm），xy 色域面积约为 Rec.709 的 1.9 倍。但单色光原色 = 极窄峰——意味着观察者同色异谱效应被推到极限。Rec.2020 uses near-monochromatic RGB primaries (630 / 532 / 467 nm), with an xy gamut area about 1.9× that of Rec.709. But monochromatic primaries = extremely narrow peaks — pushing the observer-metamerism effect to its limit.
实务后果：HDR 母带在不同型号 OLED 上、不同观看角度、不同观察者眼里的偏色比 SDR 显示更明显。这不一定是"屏幕没校准"，也可能来自显示光谱与观察者差异。Practical consequence: an HDR master shows more color shift across different OLED models, viewing angles, and observers than an SDR display does. This isn't necessarily "an uncalibrated screen" — it can come from the interaction of display spectrum and observer differences.
BT.2100 没有规定补偿方案，目前只能靠：1) 使用相同型号监视器交付；2) 多观察者交叉评估；3) 关键颜色避免落在窄峰原色附近。BT.2100 specifies no compensation scheme, so for now you can only: 1) deliver on the same monitor model; 2) cross-evaluate with multiple observers; 3) keep critical colors away from the narrow-band primaries.
9. 常见误区9. Common misconceptions
这四句话在论坛/教程/营销材料里出现频率很高，但每一条都漏掉了同色异谱给出的限定。These four lines turn up constantly in forums, tutorials, and marketing — and each one drops the qualifier that metamerism imposes.
❌ "ΔE = 0 就是颜色相同""ΔE = 0 means the colors are identical"
ΔE = 0 只在指定光源和观察者下成立。ΔE = 0 only holds under a specified illuminant and observer.换光源或换人，ΔE 就可能不再是 0。色差报告必须标注光源（D65/D50/A...）和观察者（2°/10°），否则不可比。 Change the illuminant or the person and ΔE may no longer be 0. A color-difference report must state the illuminant (D65/D50/A…) and observer (2°/10°), or it isn't comparable.
❌ "校色仪准就是眼睛看也准""If the meter is accurate, the eye sees it accurately too"
校色仪是另一个观察者。A colorimeter is just another observer.它的读数按 CIE 1931 标准观察者计算（滤色片式仪器甚至只是近似这组 CMF）。你的眼睛不是 CIE 1931——尤其在 OLED 窄峰前差异最大。校准只是"对仪器对准"，不保证"你看上去对"。 Its readings are computed against the CIE 1931 standard observer (filter-based instruments only approximate that CMF). Your eye isn't CIE 1931 — the gap is largest in front of OLED narrow peaks. Calibration only "aligns to the instrument"; it doesn't guarantee "it looks right to you".
❌ "广色域屏幕颜色更准""A wide-gamut screen is more accurate"
色域大 ≠ 颜色准A bigger gamut ≠ accurate color，反而更易触发观察者同色异谱失败。窄峰原色（Rec.2020、QD-OLED）让不同人看同一台屏的偏色更明显。广色域适合表达，不直接等于"准确"。 — if anything it more easily triggers observer metameric failure. Narrow-band primaries (Rec.2020, QD-OLED) make the cast more obvious between people viewing the same screen. Wide gamut is good for expression; it doesn't directly equal "accurate".
❌ "白平衡到 D65 就是还原真实""White-balancing to D65 restores the truth"
白平衡补偿的是相机响应，不是 Luther 条件不满足。White balance compensates the camera's response, not the unmet Luther condition.白平衡只让灰卡在相机里变成 (R=G=B)，并不能让所有物体的颜色与人眼一致——后者需要光谱级匹配，物理上几乎不可达。 White balance only makes the gray card read (R=G=B) in the camera; it can't make every object's color match the eye — that would require spectral-level matching, which is physically almost unreachable.
10. 资料来源10. Sources
权威标准 + 学术综述 + 实务参考。按照可获取性和实用性排序，引用时请回到原文核实。Authoritative standards + academic reviews + practical references, ordered by accessibility and usefulness. Always go back to the originals when citing.