The Science of Seeing

The Opportunity

A huge economy runs on how things look, and almost no student studies it as a scientist. The global art market trades tens of billions of dollars a year, benchmarked annually by the Art Basel and UBS Art Market Report. Around it sit far larger industries built on light and color: paints and coatings, the color-driven cosmetics market, the optics, sensor, and display in every phone camera, color management for print and packaging, and the deliberate look of every film and video game.

All of that runs on people who understand the physics of seeing: optical engineers, color scientists, camera and display designers, paint and pigment chemists, museum conservators, and the appearance specialists who make a rendered surface in a film or game read as wet, metallic, or soft. Almost no one reaches those fields having done real optics or perception research as a teenager.

And the open questions are close at hand. The forward physics is settled: how a wavelength of light meets a pigment, how a lens forms an image, how the eye's three cone types sample what reaches them. What is not settled is the harder half: predicting what a person will actually see. The same gray looks light or dark depending on what surrounds it. One photograph splits a room into people who see a blue dress and people who see a gold one. The work runs the whole chain, from the instrument that forms the image to the eye that reads it, and most of it is measurement.

You are well placed for it. You can make the exact image on demand, the same swatch, the same composition, the same lit surface, which makes you both the experimenter and the source. You build and calibrate the instruments, measure the light, color, and surface, and, where the question calls for it, run the perception tests your trained eye is built to judge. And your material is everywhere: any museum or gallery near you is full of real artworks to study, and the phone in your pocket is a camera, a screen, and the start of an optical instrument.

What You Get

Standing on real science. You ground your own question in published work and push it one honest step further. Research advances not by inventing from nothing, but by building on what is known.

A scientist's question, not a critic's. “Is this painting better than that one” has no answer key. “What physically differs between a painting and a photograph of it, and which of those differences can a viewer actually detect” is science: you can measure it, it has a mechanism, it can be wrong. The question is yours to find; the mentor helps you sharpen it into the second kind.

Doable without a dedicated lab. A phone camera, a printed color chart, free software (ImageJ, GIMP, Python), a cheap diffraction-grating spectrometer, lenses you build from a water droplet, a glass bead, or a 3D printer, the paints and screens you already use, and a museum or gallery you can visit. Optics and color are hands-on experimental science you can do properly this way.

A real publication and presentation pathway. A genuine result is something you can submit to the SoTS journal, present at the conference, and talk about honestly in a college application.

What Students Do

The question is yours, and finding it is most of the work. It starts from your own eye: something you noticed mixing a color, printing a photo, or standing in front of a painting a screen never quite captures. These questions don't have answer keys. The mentor helps you sharpen a question you care about and connect it to what is known. From there the work is hands-on and yours: you build optical instruments, measure light, color, and surface, photograph and analyze, and, where it fits, run perception tests, then choose your methods, write it up, and aim to publish.

Some Examples of Studies in This Area

The settled physics is not the research, and “which painting is better” is not science. Each one below pairs a real, recent, open-access study with the next question its own authors leave open. They are examples, not assignments, grouped by the instruments you build, the light and color you can measure, how the eye reads an image, and how a real artwork differs from its copy, each sized to what an advanced high schooler can do with simple gear and a museum nearby. Your own question will be narrower, and it will be yours.

Optics: the instruments you build

• A microscope you fold or cast. A microscope folded from paper around a glass bead costs about a dollar and resolves close to a micron (Cybulski and colleagues, 2014), and a hanging drop of cured silicone makes a phone lens for under a cent (Lee and colleagues, 2014). Both papers hit the same wall: aberration grows toward the edge of a simple lens. How far can you push a cheap lens before the edges blur, and what design wins at a fixed cost?
• A spectrometer from a phone. A roughly $30 open-source spectrophotometer uses a phone camera as its detector and tracks a lab instrument well enough to measure water quality (Feng and colleagues, 2021). The authors flag that its readings drift as the battery drains and differ across phone cameras, not yet pinned down. Build one, then measure exactly how much those things move the numbers.
• A microscope that out-resolves its lens. A roughly $150 Raspberry Pi microscope takes hundreds of cheap, low-resolution photos under light from different angles and computes one image sharper than its optics should allow (Aidukas and colleagues, 2019). The authors point to what still wastes resolution and time. How few lighting angles can you use before the trick breaks?

Light, color, and surface, measured

• What makes a surface look glossy. A neural network taught only to compress and predict images of surfaces, never told what gloss is, ends up reproducing both when people see gloss and when they misjudge it (Storrs and colleagues, 2021), and the authors note no model yet explains every case. Photograph a glossy object, move its highlight to where the shape and lighting say it should not be, and find where it stops looking shiny.
• Color from structure, not pigment. Some color comes not from dye but from microscopic structure, and on a butterfly wing it is the ordinary ridges on the scales, not the fancier nanostructures usually credited, that drive how the color shifts as you turn it (Zobl and colleagues, 2020). Mount an iridescent surface, a CD, a soap film, or an oil slick on a turntable and chart how its hue shifts with angle, then test what changes the curve.
• A phone that names a pigment. With a calibration step, an ordinary phone camera becomes a colorimeter accurate enough to identify historical pigments in painting replicas (Sáez-Hernández and colleagues, 2024). The authors validated it under one lamp on one phone. How fast does the accuracy fall apart when you change the light, the room, or the phone?

How the eye reads the image

• One photo, two viewers. “The dress” splits people into white-gold and blue-black because their brains assume different lighting (Lafer-Sousa and colleagues, 2015), and a later study found such images share a recipe: two washed-out, opposite colors (Jeong, 2021). Can you build a brand-new image that splits people the way the dress did?
• When color constancy breaks. A surface usually keeps its color as the light changes, but that should fail most for the most colorful surfaces (Foster and Reeves, 2022). The authors modeled it; no one has shown how big the failure looks to a human eye. Photograph pastel and saturated objects under different lights and ask people which ones changed color.
• Where one color becomes another. A name boundary between two colors makes them look more different, and that effect grows the larger the color difference (Li and Nagai, 2024). Build color pairs that straddle a boundary and pairs that sit within one name, at matched step sizes, and measure which pairs people call more different.
• Shadows that should not work. People are surprisingly bad at spotting physically impossible shadows and reflections: only about 60% right on shadows, and near chance on reflections until the error is large (Nightingale and colleagues, 2019). Measure that threshold in your peers, then go hunting for the impossible shadows painters have quietly gotten away with for centuries.

The original and its copy

• The painting versus the photograph. People looked longer at real artworks in a museum and rated them higher than the same works shown as sharp digital images in a lab (Brieber and colleagues, 2014). The authors used psychology students of uncertain motivation. Measure the gap with real gallery visitors, then try to isolate what the copy loses: size, surface, or context.
• Why you step back from a big canvas. In a real exhibition, the larger the painting, the farther back people stood, a tight linear law (Carbon, 2017). Does it hold across a mixed collection, and does it collapse when the same painting becomes a postcard or a screen?
• Direct beats mediated. People say they would rather see an artwork as a physical print than an indirect live view, and through a mirror than on an equivalent monitor, even when the images match (Bertamini and Blakemore, 2019). They only asked hypothetically. Show peers one artwork as a print, in a mirror, and on a screen, and measure what they actually prefer.

Mentor

Dr. Sergey Samsonau

• Trained 100+ students in research methodology at NYU
• Built and directed research labs program at PRISMS (top USA high school)
• Led AI projects at financial, medical, and educational institutions
• Conceptualized and organized AI Meets Science NYC-metro area conference
• Developer of original research education methodologies

View Full Profile

Prerequisites

• You make or study visual art seriously: painting, drawing, photography, digital art, design, printmaking, animation, or film all count
• Curious about why and how an image looks the way it does, not just how to make it
• Willing to read into a real literature and design your own study, not follow pre-made assignments
• Commitment to weekly meetings and work between sessions

No prior research experience required. A year of physics is useful preparation but not required: you pick up the optics, measurement, and analysis your project needs as you go.

General Details on SoTS Research Labs

• About
• 10 Reasons
• Research & College
• How It Works
• Certificate
• Expectations
• Refund Policy

Enrollment and Cost

Available in In-person and Hybrid formats, in Fall, Spring, and Summer semesters. See Princeton Labs enrollment for full details and the schedule call.

You already make the image. Now ask the question no one has answered yet.

Outside the Princeton, NJ area? Hybrid lets students anywhere participate in authentic weekly research: Zoom sessions plus a few half-day in-person visits per semester. See Princeton Labs enrollment for details.