The Science of Light — How Cameras Capture Photons

Before we discuss apertures, megapixels, or film grain, we need to start at the deepest level: the nature of light itself. This is not an abstraction. Understanding what light actually is — physically, mathematically — will change how you think about every photograph you make or view for the rest of your life.

Light is electromagnetic radiation. It travels as a wave, oscillating through electric and magnetic fields simultaneously, at the universe's speed limit: approximately 299,792 kilometers per second in a vacuum. But here is where things get strange, and where photography's entire scientific foundation rests: light also behaves as a particle. These particles are called photons — discrete, indivisible packets of energy with zero mass.

This dual wave-particle nature of light is one of the most experimentally verified and philosophically unsettling facts in all of science. For photography, the particle nature is what matters most. When light strikes a sensor or a film emulsion, it does not arrive as a smooth, continuous stream — it arrives as a hail of individual photons, each carrying a specific amount of energy determined by its wavelength.

This brings us to one of the most important discoveries in the history of science — and the one that makes every digital photograph possible. In 1905, a 26-year-old patent clerk named Albert Einstein published a paper explaining a puzzling experimental result called the photoelectric effect: when light strikes certain metals, it ejects electrons. Classical wave theory predicted that any light, given enough time, could dislodge electrons. But experiments showed that only light above a certain frequency — regardless of its brightness — could do so. A dim blue light could free electrons; a blinding red floodlight could not.

Einstein's explanation was radical: light must arrive in discrete packets (photons), and each photon either has enough energy to free an electron or it doesn't. There is no accumulation over time. This paper — not his work on relativity — is what earned Einstein the 1921 Nobel Prize in Physics. And it is the exact physical mechanism that every CCD and CMOS camera sensor in the world exploits, billions of times per second, every time you press a shutter button.

Electromagnetic radiation spans an enormous range of wavelengths and frequencies: from radio waves hundreds of meters long, to microwaves, to infrared, to visible light, to ultraviolet, to X-rays, to gamma rays shorter than an atomic nucleus. This entire range is the electromagnetic spectrum, and all of it travels at the same speed — the speed of light — but each band interacts with matter very differently.

Cameras and human eyes are both instruments tuned to a specific, narrow portion of this spectrum. ✓ Established The human eye detects wavelengths from approximately 380 to 700 nanometers — a range we call visible light, where violet sits at the short-wavelength end and deep red at the long-wavelength end. [2] One nanometer is one billionth of a meter. To put that in perspective, the entire range of colors we can perceive spans about 320 billionths of a meter — a sliver of a sliver of the full electromagnetic spectrum.

Within this visible band, different wavelengths produce different color perceptions: violet (~380–450nm), blue (~450–495nm), green (~495–570nm), yellow (~570–590nm), orange (~590–620nm), and red (~620–700nm). These are not arbitrary human categories — they correspond to distinct physical wavelengths that interact differently with pigments, materials, and photoreceptors.

Here is something crucial for photographers: silicon — the material used in virtually all digital camera sensors — is sensitive to a broader range of wavelengths than the human eye. Silicon responds to light from roughly 200nm (deep ultraviolet) all the way to 1100nm (near-infrared). Left unfiltered, a digital sensor would produce images dramatically different from what we see — foliage would appear pale, skies would look strange, and skin tones would be distorted. This is why every digital camera contains an infrared cut filter (also called a hot mirror) placed in front of the sensor — to restrict the sensor's response to approximately match human visual perception.

This has interesting implications. Photographers who use modified cameras with the IR-cut filter removed can photograph in near-infrared light, producing dreamlike images where green foliage glows white and blue skies turn nearly black. The sensor's native capability was always there — it was just filtered away.

Beyond the visible spectrum, cameras have been engineered to work in other bands for specialized applications. X-ray detectors in medicine use phosphor screens that convert X-ray photons to visible light. Radio telescopes use antenna arrays tuned to centimeter-scale wavelengths. Thermal cameras detect mid-infrared radiation emitted by warm objects. All of these are, fundamentally, the same exercise: tuning a detector to a specific band of the electromagnetic spectrum and converting the incoming photons into a recordable signal. The physical principles are universal — only the wavelength window changes.

A modern camera sensor looks, at macroscopic scale, like a small rectangle of gray silicon. But zoom in, and you find an extraordinarily precise grid of millions of individual light detectors called photosites (or photodiodes). Each photosite is a tiny well of silicon, and its job is elegantly simple: count how many photons hit it during the exposure, and report that number as an electrical charge.

The mechanism is the photoelectric effect, operating at microscopic scale. ✓ Established When a photon strikes the silicon in a photosite, it transfers its energy to an electron in the silicon's crystal lattice. If the photon carries enough energy (i.e., its wavelength is short enough), it frees that electron from its bound state, creating what is called an electron-hole pair. The freed electron is then collected and held in the photosite's potential well. At the end of the exposure, the accumulated charge in each photosite is read out, converted to a digital number by an analog-to-digital converter (ADC), and stored as the pixel value for that location. [1]

That number — typically represented in 12 or 14 bits in a RAW file — represents the intensity of light at that point in the scene. Modern digital interchangeable-lens cameras record 14-bit RAW data, allowing theoretically 16,384 distinct levels of tonal expression per channel. [3]

There are two major sensor architectures in use today: CCD (Charge-Coupled Device) and CMOS (Complementary Metal-Oxide-Semiconductor). Both are built on metal-oxide-semiconductor (MOS) technology, and both exploit the photoelectric effect identically at the individual photosite level. [5] Where they differ is in how they read out the accumulated charge.

In a CCD, the charge from each row of photosites is shifted — like a bucket brigade — along the sensor to a single output amplifier at the edge, which reads them sequentially. This produces very uniform, low-noise output, but it is slow, power-hungry, and complex to manufacture. CCDs dominated professional and scientific imaging from the 1970s through the early 2000s.

CMOS sensors, by contrast, have an amplifier directly at each photosite, allowing any pixel to be read out independently and in parallel. This makes CMOS sensors dramatically faster, less power-hungry, and cheaper to manufacture. For many years, CMOS sensors had worse noise performance than CCDs — but relentless engineering improvements, including back-side illumination (BSI) and stacked sensor designs, have largely closed that gap.

One critical — and frequently misunderstood — point: sensor pixels detect light intensity, not color. A silicon photosite is fundamentally colorblind. It counts photons without any inherent ability to distinguish their wavelengths. Color information must be added by placing colored filters over the photosites — which leads us to one of the most consequential engineering compromises in the history of digital photography.

In 1976, Kodak engineer Bryce Bayer invented a solution to the colorblind sensor problem that became so dominant it is now almost synonymous with digital imaging itself: the Bayer color filter array (CFA). The idea is straightforward: place a mosaic of tiny colored filters — red, green, and blue — over the photosite array so that each photosite records only one color of light. Then use software algorithms to reconstruct the full color image from the incomplete data.

The specific pattern Bayer chose — and that virtually every digital camera still uses today — is a 2×2 repeating grid with two green filters, one red, and one blue. ✓ Established The Bayer array uses twice as many green sensors as red or blue because the human eye is far more sensitive to green wavelengths, and maximizing green resolution produces the most perceptually accurate luminance (brightness) detail. [7]

The reconstruction process — inferring the missing two color values at each pixel location from the surrounding filtered photosites — is called demosaicing (sometimes spelled demosaicking). It is a fundamentally interpolative process: your camera is making educated guesses about what color two-thirds of its pixels actually saw, based on neighbors. Sophisticated demosaicing algorithms do this remarkably well in most situations, but they are the reason digital images can show artifacts like color moiré on fine fabric patterns or false color fringing on high-contrast edges.

✓ Established Virtually all digital cameras capture only one of three primary colors per pixel cavity, discarding roughly two-thirds of incoming light in the process. [7] This is not a flaw in any particular camera — it is a structural property of Bayer-filtered sensors.

Alternatives exist. Sigma's Foveon sensor stacks three layers of silicon at different depths, exploiting the fact that different wavelengths of light penetrate silicon to different depths — red penetrates deepest, blue least. This captures all three colors at every pixel location with no interpolation required, theoretically delivering superior color accuracy. In practice, Foveon sensors have struggled with noise performance, low-light capability, and processing complexity, limiting their commercial adoption.

At the high end, medium format cameras and studio systems sometimes use multi-shot capture — physically shifting the sensor or using a rotating filter wheel to capture red, green, and blue data at every photosite location in three separate exposures. This delivers perfect, interpolation-free color data, but requires a perfectly stationary subject and controlled lighting — practical only in studio still-life photography.

Photographic film is, at its heart, a photochemical system. A film emulsion consists of microscopic crystals of silver halide salts — typically silver bromide, silver chloride, or silver iodide — suspended in a gelatin layer coated on a transparent base (originally glass, then nitrocellulose, then polyester). These crystals are the film's light-sensitive elements, analogous to the photosites of a digital sensor.

When a photon strikes a silver halide crystal, it initiates a photochemical chain reaction. The photon's energy frees an electron from a halide ion, which then migrates through the crystal lattice to a sensitivity speck — a tiny impurity or structural defect deliberately engineered into the crystal. At the sensitivity speck, the free electron reduces a silver ion to a neutral silver atom. Repeat this process a handful of times on the same crystal, and you have created a stable cluster of silver atoms: the latent image. This is invisible to the naked eye — it is a chemical potential, not a visible mark.

✓ Established Film records light through photochemical reactions with silver-halide crystals, while digital sensors convert photons directly to electrical signals — a fundamentally different physical process. [8] The latent image is made visible through chemical development, where a reducing agent (the developer) amplifies the silver clusters by converting all the silver ions in the exposed crystals to metallic silver — a process of enormous amplification. The unexposed crystals are then dissolved away by the fixer (sodium thiosulfate), leaving a permanent image.

The key structural difference from digital imaging is that film's tonal response is analog and continuous. Rather than discrete integer values (0 to 16,383 in a 14-bit file), film's density varies smoothly across the scene. Crucially, the characteristic curve of film — the relationship between exposure and resulting density — has a gentle S-shaped roll-off at the highlight end. As you push into overexposure, film's response compresses gradually and gracefully rather than clipping abruptly. Many photographers and cinematographers argue this makes film's handling of highlights more aesthetically forgiving and visually pleasing.

Film speed — the ISO rating — reflects how readily the silver halide crystals respond to light. Faster films use larger crystals, which have a greater cross-section for photon capture but produce coarser, more visible grain. Slower films use smaller, more tightly packed crystals, producing finer grain and superior resolving power but requiring more light. The lowest commercially sold photographic film as of 2022 was ISO 0.8 (FPP Super Positive) [8] — a film requiring extraordinarily bright conditions or very long exposures.

Color film adds layers of complexity. Color negative film uses three emulsion layers, each sensitized to a different part of the spectrum, with color coupler chemicals that produce dye clouds in complementary colors during development. The spectral response of each layer is determined by the sensitizing dyes added to the silver halide crystals — and these response curves are broad, overlapping, and carefully tuned to balance color accuracy against sensitivity.

The comparison between film and digital photography is one of the most reliably contentious topics in visual media. Part of the reason the debate persists — even as digital has clearly won the commercial war — is that the two technologies differ in ways that are not simply better or worse, but different in kind. Understanding the physics clarifies which claims are objective and which are matters of aesthetics.

The ISO comparison deserves special emphasis. ✓ Established Digital cameras have achieved ISO equivalent speeds up to 4,560,000 — a sensitivity level that would be physically impossible with conventional silver-halide film chemistry. [9] The fastest professional film ever widely available, Kodak P3200 or Ilford Delta 3200, topped out at around ISO 3200 pushed in development. The gap is not incremental — it represents roughly three orders of magnitude of sensitivity advantage for digital in extreme low-light conditions.

Yet the debate about highlight handling and tonal roll-off is not purely aesthetic. It reflects a real physical difference in the response curves of the two systems. A digital sensor has a hard saturation point — when a photosite collects its maximum number of electrons, adding more photons produces no more signal, and the highlight is clipped to pure white. Film's photochemical response compresses progressively, producing highlight detail across a wider range of overexposure. Modern digital post-processing and HDR techniques can partially emulate this roll-off, but the underlying sensor physics remains binary at saturation.

Exposure — the total amount of light that reaches your film or sensor during a single photograph — is controlled by three variables that photographers call the exposure triangle: aperture, shutter speed, and ISO. Understanding these controls at the physics level, not just as knobs to turn, transforms how you make creative decisions.

Aperture: The Light Funnel

The aperture is an adjustable opening inside the lens, formed by overlapping metal blades (the diaphragm). Its diameter controls how much of the available light cone passing through the lens is actually admitted to the camera. A larger opening admits more light; a smaller opening admits less.

Aperture is expressed in f-numbers (or f-stops): f/1.4, f/2, f/2.8, f/4, f/5.6, f/8, and so on. The f-number is the ratio of the lens's focal length to the aperture diameter. Crucially, the relationship between f-number and light is inverse and squared: because aperture area scales with the square of the diameter, each full stop in the f-number sequence (e.g., f/2.8 to f/4) halves the amount of light reaching the sensor. Going from f/1.4 to f/8 — a difference of five stops — reduces incoming light by a factor of 32.

But aperture also controls depth of field — the range of distances in the scene that appear acceptably sharp. A large aperture (small f-number) produces shallow depth of field, with the background blurring into smooth bokeh. A small aperture (large f-number) produces deep depth of field, keeping far and near objects simultaneously sharp. This is not a side effect — it is a consequence of the optics of lens focusing, and it is one of the most powerful compositional tools in photography.

Shutter Speed: Time as a Dimension

The shutter speed controls how long the sensor is exposed to light — typically ranging from seconds-long exposures in night photography to 1/8000th of a second or faster in sports cameras. Shutter speed and aperture are directly reciprocal: doubling your shutter speed (e.g., from 1/125s to 1/250s) halves the exposure, exactly as closing one f-stop does.

But shutter speed also controls how motion is recorded. A fast shutter speed freezes motion — water droplets, athletes, birds in flight — capturing them as sharp instants. A slow shutter speed allows moving subjects to blur across the frame during the exposure, producing the streak of a waterfall or the light trails of city traffic. This is physics, not processing: the sensor is recording photons for the entire duration of the exposure, and a moving subject deposits its photons across a trail of pixel locations rather than a single sharp point.

ISO: Amplification and Its Cost

ISO is the most misunderstood of the three controls. ✓ Established Increasing ISO in a digital camera does not make the sensor more physically sensitive to photons. It amplifies the electrical signal from the sensor after the photons have been counted — like turning up the volume on a quiet recording. [10] The same number of photons hit the sensor at ISO 100 and ISO 6400 for a given exposure — but at ISO 6400, the resulting signal is amplified 64 times more.

The cost of this amplification is noise. Every sensor has a baseline level of random electrical variation — thermal noise, read noise, shot noise. At low ISO, this noise is overwhelmed by the signal from the photons. At high ISO, the amplification raises both the signal and the noise equally, but crucially, the photon signal was small to begin with (because there was not much light), so the noise becomes a larger fraction of the total. This is why high-ISO images look grainy.

The comparison between the human eye and a camera is one of the oldest in optics, and it is genuinely illuminating — but it needs to be handled carefully, because the eye differs from any camera in ways that are as important as the similarities.

The structural parallels are real. The eye has a lens that focuses incoming light onto a photosensitive surface at the back. It has an adjustable aperture — the iris, which controls the diameter of the pupil from approximately 2mm in bright sunlight to approximately 8mm in darkness. [11] The effective focal length of the human eye is approximately 22mm in 35mm equivalent terms (though the physical focal length is about 17mm). [11]

The photosensitive surface — the retina — contains two types of photoreceptor cells: rods and cones. Cones, concentrated in the central fovea, handle color vision in daylight conditions. There are approximately 6 million cones, with three subtypes sensitive to different wavelength ranges corresponding roughly to red, green, and blue — the same three-channel color architecture that the Bayer filter array mimics. Rods, numbering around 120 million, are distributed across the peripheral retina and are extraordinarily sensitive to light but provide no color information — which is why colors become desaturated in very dim conditions. [11]

Here is where the camera analogy begins to break down. The human eye does not capture a single, uniform image the way a camera sensor does. The fovea — the high-resolution color-sensitive center — covers only about 2 degrees of visual angle, roughly the size of your thumbnail at arm's length. The rest of the retina provides lower-resolution, motion-sensitive peripheral information. Your brain compensates for this extreme variation in resolution by constantly moving your eyes in rapid involuntary movements called saccades — several times per second — stitching together a composite picture of the scene from many high-resolution foveal samples.

This means the eye is fundamentally a scanning system rather than a single-shot imager. Some estimates of the eye's effective resolution account for this saccadic scanning and arrive at figures like 576 megapixels. ⚖ Contested This figure is highly debated — it conflates the scanning process with a static image capture in a way that does not reflect how cameras work, making direct megapixel comparisons between eyes and cameras not particularly meaningful. [11]

Perhaps the most remarkable — and under-appreciated — property of the retina's photoreceptors is their sensitivity at the absolute limit. ◈ Strong Evidence The human eye can detect a single photon. But there is a critical qualification: the eye does not transmit this single-photon event to the brain unless multiple photons are detected in a brief time window, due to signal-to-noise thresholding in the retinal neural circuitry. [4] This neural gating suppresses random thermal noise from triggering false visual percepts — a noise-reduction system built into the visual hardware at the hardware level, not in post-processing.

Ask most photographers how the human eye compares to a camera sensor in dynamic range, and you will hear: "the eye is far better — 20, 30 stops, whatever." This is true in one sense and misleading in another. Understanding the distinction is practically important for photographers working in high-contrast scenes.

Dynamic range is the ratio between the brightest and darkest parts of a scene that can be captured with useful detail simultaneously. It is measured in f-stops (or equivalently, exposure value steps or EV), where each stop represents a factor of two in brightness.

◈ Strong Evidence The human eye's instantaneous dynamic range — what you can see in a single fixation without adaptation — is approximately 10–14 f-stops. [11] This is quite close to modern high-end DSLR and mirrorless camera sensors, which achieve 8–14 stops of dynamic range. [11] In a direct instantaneous comparison, a modern full-frame sensor and a human eye are genuinely competitive.

The much larger figures — 20+ stops, sometimes cited as high as 24 stops — refer to the eye's total adaptive range: the full range of light levels across which the eye can operate, from starlight to bright sunlight, using both pupil constriction/dilation and photochemical adaptation (rod and cone sensitivity adjustment over time). This adaptation is not instantaneous — it takes seconds to minutes for the eye to fully adjust to a dramatic change in lighting. [11]

This distinction matters enormously for practical photography. When you stand in a doorway looking from a dark interior to a bright outdoor scene, you perceive detail in both — but this is your eye continuously adapting across the scene, not capturing it all at once. If you photograph the same scene with a single camera exposure, you must choose: expose for the interior and blow out the exterior, or expose for the exterior and lose the interior to darkness. This is the fundamental challenge of high-contrast photography.

Modern cameras have developed several partial solutions. HDR (High Dynamic Range) imaging merges multiple exposures captured at different settings to extend the effective range. Computational HDR in smartphones captures multiple exposures in rapid succession and merges them algorithmically before displaying a single image — a direct technological attempt to replicate the eye's adaptive capability in a single instant. Log profiles in video cameras compress the tonal range to preserve highlight and shadow information for later expansion in post-production.

None of these fully replicate the eye's continuous, spatially adaptive response — but they narrow the gap significantly. The best current full-frame mirrorless sensors, in careful RAW post-processing with graduated adjustments, can produce results that closely approximate what a viewer with normal vision would perceive standing in the same scene.

Conventional camera sensors count photons in batches — each photosite accumulates thousands to millions of photons per exposure and reports a single average value. This works beautifully in normal conditions, but it discards information at the quantum level and imposes a noise floor that limits performance in extreme low light. The next frontier in imaging aims to push beyond this — to sensors that detect and record individual photon arrival events.

The first step toward this frontier is already in commercial production. In 2018, researchers at Dartmouth College announced the Quanta Image Sensor (QIS): a sensor architecture in which each sub-pixel element — called a jot — is so small and so sensitive that it can detect a single photon. ✓ Established [5] Rather than reporting a smooth gradient of charge like a conventional photosite, each jot outputs a binary signal: photon detected, or not. The full image is constructed from the statistical ensemble of billions of these binary events, aggregated across many jots and time steps. The result is an imaging system that operates at the quantum noise floor — the theoretical limit set by the particle nature of light itself.

At the far extreme of sensitivity are Superconducting Nanowire Single-Photon Detectors (SNSPDs), developed and refined at NIST (the National Institute of Standards and Technology). These devices consist of nanometer-scale wires of superconducting material — cooled to near absolute zero — that carry a persistent electrical current. ✓ Established When a single photon strikes the nanowire, it deposits enough energy to disrupt the superconducting state, triggering a measurable electrical pulse. The wire then rapidly returns to its superconducting state, ready for the next photon. [12]

The applications radiating outward from single-photon detection technology are wide and consequential. In medical imaging, photon-counting CT scanners can produce images with dramatically lower radiation doses because they extract maximum information from every X-ray photon rather than averaging across large detector elements. In LiDAR systems for autonomous vehicles and defense applications, single-photon detectors enable ranging and imaging at ranges and resolutions impossible with conventional detectors. In quantum cryptography, SPDs are essential for detecting individual photons encoding quantum key information — their detection events are, by the laws of physics, unforgeable.

For consumer photography, the trajectory is toward progressively larger sensors with progressively lower read noise, pushing the per-pixel noise floor toward the shot-noise limit — the irreducible quantum noise set by the probabilistic arrival of photons. Some current mirrorless cameras already achieve read noise below one electron in certain readout modes, meaning the dominant noise source is no longer the electronics but the photon statistics themselves. When that threshold is crossed universally, sensor engineering will have reached its fundamental physical limit.

The arc from Einstein's 1905 paper to today's quantum imaging sensors is one of the clearest examples in modern science of a fundamental discovery — pursued for purely theoretical reasons — eventually becoming the operational foundation of a global industry. Every photograph taken on every smartphone, every frame captured on every professional mirrorless camera, every security camera frame and every astronomical telescope image executes the photoelectric effect billions of times per second. The photon arrives. The electron is freed. The image begins.

Understanding this — not as metaphor but as actual physics — transforms your relationship with your camera. The exposure triangle is not a set of arbitrary dials: it is a three-variable control system for photon collection. Sensor noise is not a camera flaw: it is the quantum probabilistic nature of light, visible to the naked eye in your images. And the coming wave of single-photon sensors is not a marketing upgrade: it is imaging technology reaching the wall of physical possibility, where the only limit is the irreducible randomness built into light itself.