Light at the Specimen Surface: Geometric Optics of Petri Dish Imaging

Adam Murray / June 14, 2026

A geometric optics framework for predicting and avoiding specular artifacts in routine biological photography, derived from first principles and illustrated through an interactive ray-tracing simulation of a layered petri dish system.

OpticsMedical PhysicsQuantitative MethodsImage and Video ProcessingBiological Physics

For A. Rapolu, whose experimental initiative motivated this analysis.

Abstract

Routine biological photography of petri dishes under typical laboratory lighting frequently produces images with specular artifacts that confound downstream image-analysis pipelines. This article develops the geometric optics of the petri dish — a polystyrene shell containing MRS agar gel, an optional aqueous liquid layer, and an optional lid — as a layered optical system, and uses the resulting framework to predict which imaging configurations produce artifacts and which avoid them. We derive the meniscus profile from the Young–Laplace equation, establish a central lemma (Lemma 3.1) showing that any non-flat liquid interface above the specimen produces specular paths from a diffuse overhead source to a centered overhead camera, and use the Fresnel curve to quantify the brightness of those paths. From this analysis we derive a catalog of interventions — directional off-axis illumination, spread plates and cover slips, cross-polarized photography, and imaging from below through the dish bottom — and the quantitative regimes under which each works. The article is accompanied by an interactive ray-tracing simulator that implements the full physics. The framework was originally motivated by the cranial-window intrinsic signal imaging configuration of mouse visual cortex, and we close by discussing the cross-modality applicability of the same geometric analysis to slide microscopy, fluorescence microscopy, microfluidic device photography, cross-polarized art photography, and satellite imaging.

Interactive Demonstration

Adjust the geometry, lighting, and camera position below to explore how light interacts with the layered optical system of a petri dish. Toggling the liquid layer and lid on or off, and switching between overhead and directional illumination, reveals which configurations produce specular artifacts at the camera and which avoid them.

Petri Dish Optics

Cross-section of a petri dish with adjustable meniscus, optional liquid layer, optional lid, and adjustable lighting. Rays refract at every interface, Fresnel-split into reflected (dashed) and transmitted (solid) branches, and attenuate via Beer-Lambert through absorbing media. Camera arrivals are color-coded: green = signal (the ray scattered diffusely off the bench and carries an image of the dish) and red = glare (purely specular, no information about the dish contents). Toggle and slide to explore the configurations; the physics and the imaging-configuration catalogue are developed in detail in sections 2 through 5.

Agar meniscus (mm)1.00

Rise of the agar surface at the dish wall

Liquid pool height (mm)2.50

Base level of the liquid pool above the agar floor

2.34.0

Camera height (mm)180

Height above the dish floor; camera looks straight down

100260

Lamp 1 angle (°)30°

Off-vertical angle (right of camera = positive); ±90° is fully grazing

-90°+90°

Lamp 2 angle (°)-30°

-90°+90°

Primary rays

756

Reaching bench

829

Camera: signal

106

Camera: glare

Signal: by count

78%

Signal: by energy

86%

Agar (MRS)Liquid layerLid (polystyrene)LampCamera lens (⌀ 16 mm)Camera hit: signalCamera hit: glare

1. Introduction

1.1 Imaging Artifacts in Biological Photography

Photographing a petri dish is one of the most routine tasks in a working microbiology lab. A bacterial sample is poured or spread onto an agar plate, incubated overnight, and the resulting colonies are counted, photographed, and analyzed. The photographs feed into image-analysis pipelines — Fiji segmentation routines, dedicated tools like OpenCFU, automated commercial colony counters — that classify pixels as colony or background and report a colony-forming-unit count. The pipeline is calibrated on clean images; it requires colonies to be visually distinct from the agar around them, with no spurious bright regions to confuse the segmentation.

This works well when the image is well-formed. It fails — sometimes spectacularly, sometimes subtly — when the image carries specular artifacts.

The most common artifacts are bright reflections off the curved liquid surfaces inside the dish. A bright crescent at the dish edge from the meniscus of a poured suspension; an off-center glare spot from a ceiling light; a thin bright ring at the inside wall where the agar itself curves upward. These are not features of the specimen. They are accurate reflections of the lighting in the room, projected onto the curved liquid surfaces of the dish and stretched across their curvature. Segmentation pipelines do not know to ignore them; they segment the bright regions as colony pixels and report inflated counts. Manual review can correct the resulting errors, but at the cost of the throughput that automation was supposed to provide.

Practical responses to this problem in laboratory practice are largely folkloric. “Use a different lamp.” “Turn off the room lights.” “Put a polarizer on the camera.” “Image from below through the dish bottom.” Each of these is sometimes effective and sometimes not, with the difference depending on details of the dish, sample, and lighting that the folklore does not specify. This article aims to replace the folklore with a derivation: the geometric optics of the layered petri-dish system is simple enough to work out from first principles, and the conditions under which each intervention works (and fails) follow from the derivation directly.

1.2 The Cross-Domain Geometry of the Specimen Surface

The petri dish is one instance of a recurring optical structure across imaging modalities. The structure is:

A partially-diffuse specimen — a colony, a section of tissue, a cell culture, a microfluidic channel — that the imaging system is trying to characterize.
A curved or layered fluid interface above the specimen — a meniscus of bacterial suspension above agar, a film of mounting medium beneath a cover slip, a layer of PBS above an exposed cortex, a varnish coat above a painting.
A camera positioned somewhere above (or below, or to one side of) the specimen, with some illumination source elsewhere in the geometry.

Each of these settings produces variants of the same artifacts — specular reflections from the curved fluid interface contaminate the image of the diffuse specimen below — and each settles, in practice, on variants of the same interventions: flatten the interface (cover slips and spread plates), restrict the lighting direction (off-axis lamps), or filter the polarization (cross-polarized photography). What unifies them is the underlying geometric optics: a small number of equations (Snell’s law, the law of reflection, the Fresnel equations, the Beer–Lambert law, and the Young–Laplace meniscus profile) that fully determine which configurations produce artifacts and which do not.

This article works out the framework in the petri-dish setting because it is the simplest of the recurring instances — the geometry is rotationally symmetric, the materials are well-characterized, the camera is a consumer smartphone, and an interactive simulator can be embedded directly in the page. The framework itself is broader; §6.1 returns to the cross-modality story and works out the correspondence with cranial-window intrinsic signal imaging, which was in fact the setting in which the analysis here was first useful.

1.3 Contributions

This article delivers five things:

A closed-form treatment of the petri dish as a layered optical system (§2), including the meniscus geometry derived from the Young–Laplace equation, the material properties of every interface and bulk medium, and the coordinate system used throughout.
An explicit analysis of the curved-surface specular problem (§3), establishing the central geometric claim — that any non-flat liquid interface above the specimen produces specular paths to a centered overhead camera from anywhere in the upper hemisphere of source positions (Lemma 3.1, the mirror-family existence statement) — and tracing its immediate consequences for petri-dish photography.
A complete treatment of refraction and Fresnel reflectance at the smooth dielectric interfaces of the system (§4), including Snell’s law in vector form, the critical angle for total internal reflection, the Fresnel reflectance coefficients for s- and p-polarized light, Brewster’s angle, and the Beer–Lambert attenuation law.
A catalog of imaging configurations and their failure modes (§5), including the overhead-lighting failure mode (§5.1), directional off-axis illumination as a geometric fix (§5.2), surface-flattening interventions via spread plates and cover slips (§5.3), polarization-based suppression (§5.4), and imaging from below (§5.5). Each subsection includes the quantitative analysis that determines when the intervention works and the simulator-toggle steps a reader can use to verify it.
An interactive ray-tracing simulator embedded at the top of the page, which traces real rays through the system with proper Snell’s law refraction, Fresnel splitting at every interface, and Beer–Lambert attenuation through every absorbing medium, allowing the reader to vary any geometric or lighting parameter and see the consequence in real time.

Throughout, the framework is presented in a form that makes its cross-modality applicability explicit, and §6 discusses the application of the same analysis to cranial-window intrinsic signal imaging, slide microscopy, fluorescence microscopy, microfluidic device photography, cross-polarized art photography, and satellite imaging.

2. The Layered Optical System

2.1 Geometry of the Petri Dish, Agar, and Optional Liquid Layer

We work with the geometry of a standard 100 mm × 15 mm polystyrene petri dish, the workhorse of microbiological practice. The dish is a shallow cylinder of inner radius $R = 50\,\text{mm}$ and inner wall height $15\,\text{mm}$ , manufactured with a polystyrene shell approximately $1\,\text{mm}$ thick on the floor and the side walls. A typical fill of MRS (de Man, Rogosa, and Sharpe) agar — the medium used for cultivating Lactobacillus species, and the one to which the simulator at the top of the page is calibrated — uses $18\,\text{mL}$ of medium, poured at $\sim 55°\text{C}$ and allowed to gel as it cools. The resulting agar layer fills the dish to a depth of approximately

$\Delta_{\text{agar}} = \frac{18\,\text{mL}}{\pi R^2} \approx 2.3\,\text{mm}$

above the dish floor.

The dish may carry an optional aqueous liquid layer above the agar — a bacterial suspension, a soaking solution, or simply water — poured to a thickness on the order of $2$ – $5\,\text{mm}$ . This liquid layer has its own free surface and its own meniscus at the dish wall, separate from the (frozen) meniscus of the agar surface beneath it.

The dish lid (when present) is a separate polystyrene disc, $\sim 1\,\text{mm}$ thick, that rests on the rim of the dish with a small clearance between its underside and the agar (or liquid) surface. The lid is offset slightly outward from the rim to allow gas exchange while still preventing dust ingress; in our model we treat the lid as a parallel-faced slab held at a fixed height above the dish floor.

For the geometric optics treatment, we work in the rotationally symmetric vertical cross-section through the dish center. Because the dish, the agar layer, the liquid layer, and the lid all share rotational symmetry about the vertical axis through the dish center, a 2D treatment captures all of the relevant geometric structure; skew rays through the volume can be reconstructed by rotation, and the simulator at the top of the page traces rays in exactly this 2D plane.

We adopt the coordinate system in which $x$ denotes the horizontal coordinate (across the dish) and $y$ denotes the vertical (upward) coordinate. The dish floor sits at $y = 0$ , the dish rim at $y = 15\,\text{mm}$ , the dish center at $x = 0$ , and the dish walls at $x = \pm R = \pm 50\,\text{mm}$ .

2.2 Meniscus Formation and the Young–Laplace Equation

Liquids in containers form curved free surfaces — menisci — wherever surface tension and gravity compete. The petri dish carries two such surfaces:

The agar meniscus, set in place at the moment the agar solidified at $\sim 35°\text{C}$ and frozen in that geometry ever after by the gel’s elastic strength. Its shape was determined by the surface tension of liquid agar at the time of solidification — broadly similar to water, $\gamma \approx 0.06$ – $0.07\,\text{N/m}$ — and the contact angle of the molten agar against the dish material.
The liquid-pool meniscus (when a liquid layer is present), set continuously by the balance between the layer’s surface tension and its own weight.

The shape of a static meniscus near a vertical wall is given by the Young–Laplace equation, which states that the pressure jump across a curved interface is proportional to the local mean curvature:

$\Delta P = \gamma \,\kappa = \gamma \left( \frac{1}{R_1} + \frac{1}{R_2} \right),$

where $\gamma$ is the surface tension and $R_1, R_2$ are the two principal radii of curvature of the interface. In our rotationally symmetric geometry, the relevant local cross-section is that of a curve in a vertical plane: one principal radius is along the meniscus profile in that plane, and the other tends to infinity in the planar-wall limit, since the dish radius $R$ is much larger than the capillary length introduced below.

For a 2D profile $y(x)$ , with $y$ the height of the interface above the bulk level and $x$ a horizontal coordinate, the curvature is $\kappa = y'' / (1 + y'^2)^{3/2}$ and the pressure jump across the interface arises from the hydrostatic head difference, $\Delta P = \rho g y$ , where $\rho$ is the liquid density and $g$ is the gravitational acceleration. Substituting:

Theorem 2.1 (Young–Laplace equation, planar 2D meniscus)

The shape $y(x)$ of a static liquid meniscus in 2D against a vertical wall, with surface tension $\gamma$ , density $\rho$ , in a gravitational field of strength $g$ , satisfies

$\gamma \,\frac{y''}{(1 + y'^2)^{3/2}} = \rho g y.$

The equation contains a natural length scale.

Definition 2.1 (Capillary length)

The capillary length $\lambda_c$ associated with a liquid of surface tension $\gamma$ and density $\rho$ in a gravitational field of strength $g$ is

$\lambda_c = \sqrt{\gamma / (\rho g)}.$

It is the length scale below which surface tension dominates the meniscus shape and above which gravity does. For water at room temperature, $\lambda_c \approx 2.71\,\text{mm}$ .

The meniscus is essentially confined to a band of width $\sim \lambda_c$ adjacent to the wall.

Lemma 2.1 (Linearised meniscus profile)

In the small-slope limit $|y'| \ll 1$ , the Young–Laplace equation linearises to $y'' = y/\lambda_c^2$ . For a dish of inner radius $R \gg \lambda_c$ with walls at $x = \pm R$ and a meniscus magnitude $m$ at each wall, the linearised solution is

$y(x) \approx m \cdot \exp\!\left(-\frac{R - |x|}{\lambda_c}\right),$

accurate to corrections of order $\exp(-2R/\lambda_c)$ .

Proof

In the small-slope limit, $(1 + y'^2)^{3/2} \approx 1$ and Theorem 2.1 reduces to $\gamma\,y'' = \rho g y$ , i.e. $y'' = y/\lambda_c^2$ by Definition 2.1. The general solution on the half-line is a linear combination of $\exp(\pm x/\lambda_c)$ ; selecting the branch that decays away from the wall and matching the boundary value $y(\text{wall}) = m$ gives the stated form for each wall. Since $R / \lambda_c \approx 18$ for a 100 mm dish containing water, the contribution of one wall’s meniscus to the other is of order $\exp(-2R/\lambda_c) \approx 10^{-16}$ , and the two wall contributions can be superposed without correction. $\quad \square$

The meniscus magnitude $m$ depends on the contact angle $\theta_c$ between the liquid and the dish material at the triple line — the line of contact between liquid, air, and solid. An integration of the linearised equation under the contact-angle boundary condition (see, e.g., de Gennes, Brochard-Wyart, and Quéré §2.4) yields

$m \approx \lambda_c\,\sqrt{2(1 - \sin \theta_c)}.$

For water against clean polystyrene, $\theta_c \approx 87°$ , giving a meniscus magnitude of only $m \approx 0.14\,\text{mm}$ — a barely-perceptible curvature at the wall. For surfactant-loaded liquids — bacterial suspensions in MRS broth typically contain biological surfactants from the cells themselves, and ingredients such as Tween 80 are often added explicitly to the medium — the effective contact angle drops substantially. At $\theta_c = 60°$ , $m$ rises to $\approx 1.4\,\text{mm}$ ; at $\theta_c = 30°$ , to $\approx 2.7\,\text{mm}$ . The representative value $m \approx 0.5\,\text{mm}$ used in §3.3 and §5.1 corresponds to a contact angle near $80°$ , intermediate between a clean dry polystyrene surface and a heavily surfactant-loaded liquid. The simulator allows the user to vary $m$ via the agar-meniscus and liquid-pool-meniscus sliders to explore the dependence directly.

For the agar surface, the same equation applies at the moment of solidification. Once the gel sets, the meniscus shape is locked in by the gel’s elastic strength and is essentially constant on experimental timescales. For poured molten MRS agar in a polystyrene dish, the contact angle at gel-point is approximately that of dilute aqueous solutions on warm polystyrene — roughly $80°$ – $85°$ — giving an agar meniscus magnitude of $\sim 0.3$ – $0.5\,\text{mm}$ .

2.3 Material Properties at the Wavelengths of Interest

The geometric optics framework of §3–§5 requires three material properties at every interface or volume:

The refractive index $n$ , which sets refraction angles via Snell’s law (Theorem 4.1) and Fresnel reflectances via the Fresnel equations (Definition 4.2).
The absorption coefficient $\alpha$ , which sets the attenuation rate of light passing through a transmitting medium via the Beer–Lambert law (Theorem 4.2).
The specular-versus-diffuse character of any reflecting surface, which sets whether reflected light carries directional information about the source (§3.1).

The values used by the simulator are summarized below.

Medium / surface	$n$	$\alpha$ (mm⁻¹)	Reflection character
Air	$1.0003 \approx 1.00$	$\sim 0$	(transparent)
Water	$1.333$	$\sim 10^{-3}$	smooth-specular at free surface
MRS agar	$1.34$	$\sim 0.35$	bulk-scattering (opalescent); smooth-specular at gel surface
Polystyrene	$1.59$	$\sim 10^{-3}$	smooth-specular at both faces
Benchtop (laminate)	(opaque)	(opaque)	Lambertian, albedo $\sim 0.55$

The refractive indices are nearly constant across the visible spectrum for these materials, with variations of order $0.005$ – $0.01$ between blue and red. For our purposes this is small enough that we treat each medium as having a single representative index across the visible range.

The absorption coefficients show stronger wavelength dependence. MRS agar absorbs preferentially in the blue end of the visible spectrum, giving the gel its characteristic warm amber color: white light passing through has its blue component preferentially attenuated, leaving the transmitted light yellow-amber. The absorption arises from a mixture of pigmented organic compounds in the medium (peptone, beef extract, yeast extract, and products of Maillard reactions during medium preparation). The Beer–Lambert formula of §4.4 applies separately to each spectral component, and the perceived color of light emerging from the agar is the wavelength-weighted result. For the simulator we use a single representative value $\alpha = 0.35\,\text{mm}^{-1}$ averaged across the visible range; this captures the dominant absorption budget while treating color as a secondary effect.

Polystyrene and water are essentially transparent across the visible spectrum, with absorption coefficients of order $10^{-3}\,\text{mm}^{-1}$ that yield negligible attenuation over the $1$ – $5\,\text{mm}$ path lengths encountered in the dish.

The specular-versus-diffuse character of each surface determines the kind of information it carries to the camera (§3.1). The smooth dielectric interfaces — air-water, water-agar (small Fresnel coefficient because the indices are close, but still Fresnel-character), air-polystyrene, and agar-polystyrene — are essentially pure specular, with reflectance governed by the Fresnel equations of §4.3. The agar bulk, in contrast, exhibits significant volume scattering: light entering the gel scatters off small density inhomogeneities and refractive-index variations in the gel matrix, and the scattered light eventually exits the gel diffusely. This volume scattering is what gives the gel its slightly opalescent appearance and what allows a camera to see colonies and surface features embedded in or on the agar. The simulator approximates the agar’s combined transmission and volume scattering with the absorption coefficient above; a fully physically faithful treatment would require Monte Carlo volume scattering and is beyond the scope of this work.

The benchtop beneath the dish — typically a painted or laminate countertop in a working lab — is opaque and Lambertian, with an albedo of order $0.5$ – $0.6$ . This is the surface that the simulator models as its “bench” (§5.1), and the diffuse light it scatters back up through the dish is the dominant source of image-forming light in a benchtop imaging setup (§5.5).

3. Reflection at Smooth and Curved Surfaces

3.1 Specular and Diffuse Reflection: A Material Property Distinction

Light reflecting off a surface follows one of two distinct angular patterns, and most real materials produce a blend of the two.

Definition 3.1 (Specular reflection)

A reflection in which incident light is sent in a single, geometrically determined outgoing direction — the one symmetric to the incoming direction across the surface normal at the point of incidence. The angular distribution of the reflected light is a delta function in this direction.

Definition 3.2 (Diffuse reflection, Lambertian limit)

A reflection in which incident light is scattered across the entire outward hemisphere with an angular distribution proportional to the cosine of the angle from the surface normal. The angular pattern of the outgoing light carries no memory of the incident direction; it depends only on the surface’s optical properties at the point of reflection.

Real surfaces sit on a continuum between these idealized limits. A polished sheet of metal is nearly pure specular, a fresh sheet of chalk is nearly pure Lambertian, and most everyday materials produce some mix. A glossy printed page, for instance, gives a strong specular reflection at certain viewing angles and a weaker diffuse component that reveals the ink underneath. The petri dish system contains both kinds at once: the agar surface scatters diffusely (the gel is slightly opalescent, with scattering centers embedded throughout its volume), while the air–liquid and air–polystyrene interfaces are smooth dielectrics that produce sharp specular reflections.

The distinction matters here because the two components encode different information. A diffuse reflection from the agar surface carries information about the agar itself — its color, its opacity, and any colonies, sample, or markings on its surface — because the light it sends to the camera was reorganized by the surface, with the source direction discarded. A specular reflection from any of the smooth interfaces carries information about the light source: its position, its color, its angular profile. Specular reflections from surfaces above the dish therefore project lighting features onto the image as bright regions that have nothing to do with the specimen. Those regions are what we call artifacts in this article: they are accurate reports about something the photographer is not trying to photograph.

3.2 The Law of Reflection

Specular reflection obeys a single geometric rule, derivable from Fermat’s principle of least time.

Theorem 3.1 (Law of reflection)

The angle of incidence equals the angle of reflection, both measured from the surface normal, and the incident ray, the reflected ray, and the surface normal lie in a single plane.

In vector notation, if the incident direction is $\mathbf{d}_{\text{in}}$ (a unit vector pointing along the ray’s direction of travel) and the outward-pointing unit normal at the reflection point is $\hat{\mathbf{n}}$ , the reflected direction is

$\mathbf{d}_{\text{out}} = \mathbf{d}_{\text{in}} - 2\,(\mathbf{d}_{\text{in}} \cdot \hat{\mathbf{n}})\,\hat{\mathbf{n}}.$

The geometric content is straightforward: subtract twice the component of the incoming direction that points into the surface. This flips the normal-aligned component of $\mathbf{d}_{\text{in}}$ while leaving its tangential component unchanged. The simulator at the top of the page implements this formula at every Fresnel interaction; each dashed reflected branch shows a ray that obeyed the law of reflection at its parent surface.

Fermat’s derivation, which we omit, observes that the total optical path length $|\mathrm{SP}| + |\mathrm{PD}|$ from a source point $\mathrm{S}$ via a reflection point $\mathrm{P}$ on the surface to a destination point $\mathrm{D}$ is stationary with respect to small variations of $\mathrm{P}$ along the surface exactly when the local geometry symmetrises around the surface normal — which is the equal-angles condition above. Hecht §4.2 gives a clean one-page treatment.

3.3 The Curved-Surface Problem: Continuous Families of Mirror Orientations

The law of reflection is local. It tells us what a single point on a surface does to a single incident ray. A flat surface presents the same local geometry everywhere — one outward normal, one reflected direction for each incident direction. A curved surface presents a continuously varying local geometry — the normal rotates as you move along the surface, and what each point does to an incident ray depends on where the point is.

This shift, from a single normal to a continuous family of normals, is what produces the central problem this article addresses.

Lemma 3.1 (Mirror-family existence)

Let $\gamma : [a, b] \to \mathbb{R}^2$ be a smooth curve, and let $\theta(t)$ denote the outward normal angle at $\gamma(t)$ . Assume $\theta$ varies continuously and monotonically over $[a, b]$ . Let $\mathrm{S}, \mathrm{D}$ be points on the side of $\gamma$ that the outward normal points toward, and let $\beta(t)$ denote the bisector angle of $\angle \mathrm{S}\gamma(t)\mathrm{D}$ as a function of $t$ . If $\beta(a) \in [\theta(a), \theta(b)]$ and $\beta(b) \in [\theta(a), \theta(b)]$ , then there exists $t^* \in [a, b]$ at which the local outward normal coincides with $\beta(t^*)$ — producing a specular ray-path from $\mathrm{S}$ to $\mathrm{D}$ via $\gamma(t^*)$ .

Proof

Define $f : [a, b] \to \mathbb{R}$ by $f(t) := \theta(t) - \beta(t)$ . The angles $\theta$ and $\beta$ are continuous in $t$ , so $f$ is continuous. By hypothesis $\beta(a) \geq \theta(a)$ and $\beta(b) \leq \theta(b)$ , whence $f(a) \leq 0$ and $f(b) \geq 0$ . By the intermediate value theorem there exists $t^* \in [a, b]$ at which $f(t^*) = 0$ , i.e. $\theta(t^*) = \beta(t^*)$ . $\quad \square$

For the petri dish: the meniscus profile near the dish wall, derived in §2.2 from the Young–Laplace equation, has the approximate form

$y(x) \approx m \cdot \exp\!\left(-\frac{R - |x|}{\lambda_c}\right),$

where $m$ is the meniscus magnitude (in mm), $R$ is the dish radius, and $\lambda_c \approx 2.7$ mm is the capillary length of water. The local slope of the meniscus is

$\frac{dy}{dx} = \pm \frac{m}{\lambda_c} \exp\!\left(-\frac{R - |x|}{\lambda_c}\right),$

which ranges from approximately zero at the dish center to a maximum of $m / \lambda_c$ at the dish wall. For a typical meniscus magnitude $m = 0.5$ mm, this peak slope is approximately $0.18$ , corresponding to a local-normal tilt of approximately $\arctan(0.18) \approx 10°$ from vertical.

Lemma 3.1 then applies to any source–camera geometry whose required bisector angle falls within the range $[0°, 10°]$ swept by the meniscus normal. This range covers essentially every routine imaging configuration in which the camera sits directly above the dish and the lighting comes from anywhere within $\sim 20°$ of vertical — that is, the lighting condition of essentially every laboratory ceiling. The conclusion is geometric: for any non-zero meniscus, a directly-overhead camera under typical room lighting receives specular paths from a continuous arc of meniscus points. This is the geometric root of the overhead-lighting failure mode, which we analyze quantitatively in §5.1.

Remark

Picture the meniscus as a strip of tiny flat mirrors arranged in an arc around the inside of the dish, each tilted at a slightly different angle from the next. When the room is lit by a diffuse overhead source, every mirror reflects the bit of the ceiling directly opposite its own surface normal up into the camera — and across the strip, that means some mirror somewhere reflects each part of the ceiling. The bright artifact you see at the edge of the dish in such a photo is, quite literally, an image of the ceiling, projected onto the meniscus and stretched across its curvature.

Two practical interventions follow directly from this picture, and both work by removing one of the two continuous ranges that Lemma 3.1 relies on.

The first intervention removes the continuous range of surface normals. If the surface is flat, every point has the same normal, and the bisector condition either holds for every point simultaneously (producing a single intense glare spot when source and camera are positioned just so) or fails everywhere (producing no specular artifact). The configuration becomes binary rather than continuous. Spread-plate techniques (where bacterial suspension is absorbed into the agar rather than left as a curved liquid layer on top) and optical cover slips (where a flat plate of glass or polystyrene is laid across the dish, presenting a single horizontal normal everywhere) both implement this fix. We return to both in §5.3.

The second intervention removes the continuous range of incoming directions. If the lighting source is concentrated in a single direction rather than diffused across the upper hemisphere, the bisector condition is satisfied at a single point on the curved surface, not an arc — and choosing the source angle carefully can push the required bisector outside $[\theta(a), \theta(b)]$ entirely, in which case the hypothesis of Lemma 3.1 fails and no specular path exists. Directional off-axis illumination implements this fix. We return to it in §5.2.

Both interventions are interactive in the simulator at the top of the page. To see the failure mode, toggle “Overhead lighting” on with the lid off and the liquid layer on, and place the camera directly above the dish (camera height around 180 mm); red glare-hit dots accumulate on the camera and “Signal: by count” drops well below 50%. To see the first intervention, toggle off the liquid layer, collapsing the strongest meniscus contribution. To see the second, toggle “Overhead lighting” off and turn on Lamp 1 at an angle around $\pm 30°$ to $\pm 45°$ ; the glare collapses and “Signal: by count” climbs toward 100%.

4. Refraction and the Fresnel Split

4.1 Snell’s Law at Planar and Curved Interfaces

When a ray crosses an interface between two transparent media, it bends. The amount and direction of the bending are given by Snell’s law, derivable like the law of reflection from Fermat’s principle of least time.

Theorem 4.1 (Snell’s Law)

Let $n_1, n_2$ be the refractive indices of two transparent media meeting at a smooth interface, and let $\theta_1$ be the angle of incidence (measured from the surface normal) of a ray crossing the interface. The refraction angle $\theta_2$ on the far side satisfies

$n_1 \sin \theta_1 = n_2 \sin \theta_2.$

In vector form, if $\mathbf{d}_{\text{in}}$ is the unit incident direction and $\hat{\mathbf{n}}$ is the outward-pointing unit surface normal on the incident side, the refracted direction is

$\mathbf{d}_{\text{out}} = \frac{n_1}{n_2}\, \mathbf{d}_{\text{in}} + \left(\frac{n_1}{n_2} \cos \theta_1 - \cos \theta_2 \right) \hat{\mathbf{n}},$

with $\cos \theta_1 = -\mathbf{d}_{\text{in}} \cdot \hat{\mathbf{n}}$ and $\cos \theta_2 = \sqrt{1 - (n_1/n_2)^2 \sin^2 \theta_1}$ . This vector form is valid whenever refraction is possible, i.e. when $(n_1/n_2)^2 \sin^2 \theta_1 \leq 1$ . The case where the square root is imaginary corresponds to total internal reflection, handled in §4.2.

For the petri dish, the interfaces of interest are:

Air ( $n = 1.00$ ) ↔ liquid pool ( $n = 1.33$ )
Liquid pool ↔ agar ( $n = 1.34$ )
Agar ↔ polystyrene floor ( $n = 1.59$ )
Air ↔ polystyrene lid ( $n = 1.59$ )

In each case the ray bends toward the surface normal when entering the denser medium (lower index to higher index), and away when entering the less dense one. The simulator at the top of the page implements this formula at every Fresnel interaction; the solid (transmitted) branch of every ray follows the refracted direction, with the dashed (reflected) branch obeying Theorem 3.1.

Example (A typical ray through the layered system)

Consider a ray entering the dish from above at $10°$ from vertical — a representative angle for a ceiling-lit configuration. The ray crosses three interfaces on its way to the dish floor, and Snell’s law applies at each.

At the air–water interface, $\theta_1 = 10°$ so $\sin \theta_1 = 0.174$ , and

$\sin \theta_2 = \frac{n_{\text{air}}}{n_{\text{water}}} \sin \theta_1 = \frac{1.00}{1.33}(0.174) = 0.130 \quad \Rightarrow \quad \theta_2 \approx 7.5°.$

The ray has bent $\approx 2.5°$ closer to the surface normal.

At the water–agar interface, the two media have nearly identical refractive indices, so

$\sin \theta_3 = \frac{n_{\text{water}}}{n_{\text{agar}}} \sin \theta_2 = \frac{1.33}{1.34}(0.130) \approx 0.130 \quad \Rightarrow \quad \theta_3 \approx 7.4°.$

Almost no bending at all. Optically, the water–agar interface is nearly invisible.

At the agar–polystyrene interface,

$\sin \theta_4 = \frac{n_{\text{agar}}}{n_{\text{poly}}} \sin \theta_3 = \frac{1.34}{1.59}(0.130) \approx 0.109 \quad \Rightarrow \quad \theta_4 \approx 6.2°.$

The ray has bent another $\approx 1.2°$ toward the normal.

The total deflection across all three interfaces is $10° \to 6.2°$ , a net change of just under $4°$ . This is a small effect — the path through the petri dish is almost a straight line, with the dominant geometric distortion coming from the curvature of the meniscus rather than from refraction at the flat layer interfaces. But “almost straight” is not the same as “straight,” and the residual bending is enough to displace specular paths by several millimeters at the dish floor for typical imaging geometries.

4.2 Total Internal Reflection

Snell’s law has a kinematic obstruction. When light passes from a denser medium ( $n_1$ ) to a less dense medium ( $n_2 < n_1$ ), there exists a critical angle beyond which the equation $n_1 \sin \theta_1 = n_2 \sin \theta_2$ has no real solution: the right-hand side would require $\sin \theta_2 > 1$ . Past this angle the light cannot escape the denser medium and is reflected entirely back into it.

Definition 4.1 (Critical angle, total internal reflection)

The critical angle $\theta_c$ for an interface between media of refractive indices $n_1 > n_2$ , viewed from the denser side, satisfies

$\sin \theta_c = \frac{n_2}{n_1}.$

For incidence angles $\theta_1 > \theta_c$ , no light is transmitted into the less dense medium; the ray undergoes total internal reflection off the interface, obeying the law of reflection (Theorem 3.1) with all of its energy retained on the incident side.

For the air–water interface, the critical angle viewed from the water side is

$\theta_c = \arcsin\!\left(\frac{1.00}{1.33}\right) \approx 48.8°.$

This matters for any configuration in which light enters the dish through the lid, refracts down into a liquid layer, and is then incident on the water-air interface from below at a steep angle — for example, when the lid is on and ambient room light enters through the lid edges. Light trapped inside the liquid layer at angles exceeding the critical angle will bounce repeatedly between the air–water interface above and the agar surface below, slowly attenuated by Beer–Lambert losses in the agar (§4.4), before eventually escaping or being absorbed entirely. The simulator’s maxDepth = 7 setting truncates these paths after seven Fresnel splits, which captures the dominant contributions; longer paths exist but carry negligible flux.

4.3 The Fresnel Equations: Energy Partition at an Interface

Snell’s law specifies the direction of the transmitted and reflected rays, but it says nothing about how the incident energy is partitioned between them. That information comes from Maxwell’s equations applied at the boundary, and the result is the Fresnel equations.

Definition 4.2 (Fresnel reflectance coefficients)

For a ray of unit incident amplitude crossing an interface from medium $n_1$ to medium $n_2$ at incidence angle $\theta_1$ and refraction angle $\theta_2$ , the amplitude reflectances for s-polarized (perpendicular to the plane of incidence) and p-polarized (parallel) light are

$r_s = \frac{n_1 \cos \theta_1 - n_2 \cos \theta_2}{n_1 \cos \theta_1 + n_2 \cos \theta_2}, \qquad r_p = \frac{n_2 \cos \theta_1 - n_1 \cos \theta_2}{n_2 \cos \theta_1 + n_1 \cos \theta_2}.$

The corresponding energy reflectances are $R_s = |r_s|^2$ and $R_p = |r_p|^2$ . For unpolarized light, the energy reflectance is the average of the two components,

$R(\theta_1) = \tfrac{1}{2}(R_s + R_p),$

and the transmittance is $T(\theta_1) = 1 - R(\theta_1)$ by conservation of energy.

At normal incidence ( $\theta_1 = 0$ ) both polarization components reduce to the same value,

$R(0) = \left(\frac{n_1 - n_2}{n_1 + n_2}\right)^2,$

which evaluates to approximately $0.020$ ( $2.0\%$ ) for the air–water interface, $0.052$ ( $5.2\%$ ) for air–polystyrene, and a vanishingly small $7 \times 10^{-6}$ (less than $0.001\%$ ) for water–agar — the indices are so close that the interface is nearly invisible optically. Most light passes through these interfaces; only a few percent reflects back at normal incidence. But as $\theta_1$ grows, $R(\theta_1)$ rises monotonically, approaching $1$ as $\theta_1 \to 90°$ . At grazing incidence the surface acts as a near-perfect mirror.

The p-component has additional structure. There exists an angle at which $r_p$ vanishes identically — Brewster’s angle.

Definition 4.3 (Brewster’s angle)

For an interface between media $n_1$ and $n_2$ , Brewster’s angle $\theta_B$ is the angle of incidence at which $r_p = 0$ , so that the reflected light is purely s-polarized. It satisfies

$\tan \theta_B = \frac{n_2}{n_1}.$

For air–water this is $\theta_B = \arctan(1.33) \approx 53.1°$ ; for air–polystyrene, $\theta_B \approx 57.8°$ . At Brewster’s angle, a linear polarizer aligned with the p-direction will eliminate the reflected light entirely while leaving the diffuse component largely intact. This is the geometric basis of the polarization-based artifact suppression discussed in §5.4.

Remark

The Fresnel curve, $R(\theta)$ , has a shape that anyone who has driven on a wet road already knows by intuition. Looking down at the surface from directly above — nearly normal incidence — you see right through the water film to the asphalt; only a few percent of the light is reflected back. But viewing the same wet road from a low angle far ahead of the car, the reflection becomes mirror-bright; oncoming headlights appear as sharp specular highlights, sometimes brighter than the headlights themselves. The same thing happens at the inside edge of the meniscus in a petri dish photo: the edge points have higher local incidence angles, because the meniscus slope is steepest there, and the Fresnel coefficient at those angles is correspondingly larger. The bright crescent you see is amplified twice — first by the existence of a specular path (Lemma 3.1), and second by the larger reflectance at the steeper edge points.

4.4 Beer–Lambert Attenuation Through Absorbing Media

Even between interface crossings, a ray traveling through a non-perfectly-transparent medium loses energy to absorption. The intensity decays exponentially with path length.

Theorem 4.2 (Beer–Lambert law)

A ray of initial intensity $I_0$ traveling a distance $d$ through a homogeneous absorbing medium with absorption coefficient $\alpha$ (in inverse length) emerges with intensity

$I(d) = I_0 \exp(-\alpha d).$

The absorption coefficient is wavelength-dependent — the medium absorbs different colors at different rates. For MRS agar in the visible range, absorption is strongest in the blue and weakest in the red, which is what gives the gel its characteristic warm amber color: white light passing through has its blue component preferentially attenuated, leaving the transmitted light yellow-amber. The simulator uses a single representative value $\alpha = 0.35\,\text{mm}^{-1}$ for the agar, which over a typical $2.3\,\text{mm}$ fill depth gives a single-pass transmittance of

$\frac{I}{I_0} = \exp(-0.35 \times 2.3) = \exp(-0.81) \approx 0.45.$

Roughly $55\%$ of the light entering a single agar pass is absorbed before reaching the other side. The polystyrene lid and dish floor have absorption coefficients on the order of $10^{-3}\,\text{mm}^{-1}$ and contribute negligibly over their $1\,\text{mm}$ thickness, and pure water has $\alpha \approx 10^{-3}\,\text{mm}^{-1}$ over the same range. The dominant absorption budget therefore belongs to the agar.

5. Imaging Configurations

5.1 The Overhead-Lighting Failure Mode

The configuration most experimentalists default to — dish on the benchtop, camera directly overhead, ambient ceiling lighting — is also the configuration in which the geometry of §3 and the energetics of §4 conspire most efficiently to produce artifacts. We work out the failure mode quantitatively here.

Consider the scenario in concrete terms. A 100 mm petri dish sits on a benchtop. A clear aqueous bacterial suspension has been poured over the agar, producing a liquid layer with a meniscus of magnitude $m \approx 0.5\,\text{mm}$ near the dish wall (per the Young–Laplace equation, §2.2). The camera — a smartphone, a webcam, a USB microscope — is mounted directly above the dish at a height of order $100$ – $300\,\text{mm}$ . Ceiling lighting illuminates the scene as a near-uniform-radiance upper hemisphere.

By Lemma 3.1, every point $P$ on the meniscus produces a specular path to the camera from some source position $S$ in the upper hemisphere. Explicitly: at a meniscus point with local-normal angle $\theta_n(P) \in [0°, 10°]$ (measured from vertical, per §3.3), the source $S$ that produces a specular ray-path $S \to P \to C$ to a camera $C$ directly overhead must lie at an angle of approximately $2\theta_n(P) - \theta_{PC}$ from vertical (where $\theta_{PC}$ is the angle from $P$ to the camera, of order $\pm 15°$ at the meniscus for $h_C \approx 180\,\text{mm}$ ). For $\theta_n \in [0°, 10°]$ and the meniscus on both sides of the dish, this required source position ranges over $\alpha \in [-15°, +15°]$ from vertical — comfortably inside any ambient ceiling-lighting field. The meniscus is fully populated with specular contributors.

The brightness contributed by each point is governed by the Fresnel reflectance $R(\theta_i)$ (§4.3), where $\theta_i$ is the local incidence angle at $P$ . For the small meniscus normal angles in this scenario, $\theta_i$ is also small — only a few degrees from normal — so $R \approx R(0) \approx 0.020$ at the air–water interface. Two percent of the source radiance is reflected per unit area of the meniscus. Across the full meniscus arc, integrating the contributions weighted by the local Fresnel coefficient and the differential area gives a total reflected flux at the camera that is small in absolute terms but substantial compared to the diffuse signal carried back by light scattered off the agar and the bench beneath it. The result is a high-contrast artifact superimposed on a low-contrast signal — exactly the regime where automated colony-segmentation pipelines fail.

The visual signature is a bright crescent inside the dish rim — quite literally an image of the ceiling, projected onto the meniscus and stretched across its curvature (Remark §3.3). The crescent is brightest at the wall, where the meniscus is steepest, and fades inward toward the center, where it flattens.

Example (The failure mode in concrete terms)

Pour 18 mL of bacterial suspension onto a fresh agar plate in a standard 100 mm dish. Place the dish on a typical lab bench, mount a smartphone $\approx 180\,\text{mm}$ above the dish on a tripod or copy stand, and switch on the ceiling lights. This is the default configuration — no special setup. Take a photo.

The image will show:

A bright crescent or full ring inside the dish rim, where the curved meniscus has reflected a portion of the ceiling into the camera. The width of the bright region scales with the meniscus magnitude; for a typical $m = 0.5\,\text{mm}$ meniscus, the bright crescent occupies the outer $\sim 5\,\text{mm}$ of the dish in the image.
Structured intensity within the crescent, because the meniscus stretches the ceiling image radially. Bright lights, ceiling tiles, anything with spatial structure projects onto the meniscus as a distorted reflection.
Saturated or near-saturated pixels at the crescent’s brightest points, because the reflectance there is amplified both by Lemma 3.1 (specular path existence) and by the higher local incidence angle at the steeper edge of the meniscus (Remark §4.3, the wet-road effect).

In the simulator at the top of the page, this configuration corresponds to: lid off, liquid layer on, bench scatter on, overhead lighting on, both lamps off, camera height $\approx 180\,\text{mm}$ . The reported statistics will show a substantial glare fraction — “Signal: by count” drops well below $50\%$ , and the red glare-hit dots accumulate at the camera lens. Toggling the liquid layer off (collapsing the strongest meniscus contribution) is the cleanest way to isolate the meniscus contribution: signal recovers, glare drops dramatically. The same artifact in the same place, on demand.

5.2 Directional Off-Axis Illumination

A diffuse upper-hemisphere source (room ceiling lighting) gives the geometry of §3.3 maximum opportunity to produce artifacts: for every point on the meniscus, some source in the upper hemisphere can produce a specular path to the camera. The most direct way to defeat this is to remove the upper hemisphere and replace it with a single concentrated direction.

For a source positioned at angle $\alpha$ from vertical and a camera at angle $\theta_{PC}$ from vertical as seen from a meniscus point $P$ , the specular condition (Theorem 3.1) is

$2\theta_n(P) - \alpha = \theta_{PC},$

i.e. the local normal at $P$ must bisect the directions from $P$ to source and from $P$ to camera. For a camera directly above the center of a 100 mm dish at height $\sim 180\,\text{mm}$ , $|\theta_{PC}|$ at the meniscus is approximately $15°$ (depending on which side of the dish the meniscus point sits on). With the meniscus normal range $\theta_n \in \pm[0°, 10°]$ (§3.3), the specular condition is satisfiable at some meniscus point only when

$\alpha \in [-15°, +15°] \text{ from vertical}.$

Lamp angles within $\pm 15°$ of vertical (equivalently, within $\sim 75°$ of horizontal) produce specular meniscus paths to a centered overhead camera. Lamp angles outside this range — anything substantially below $75°$ from horizontal — do not. The hypothesis of Lemma 3.1 fails, and no specular meniscus path exists.

The simulator implements this configuration. Setting “Overhead lighting” off and turning on Lamp 1 at $\pm 30°$ to $\pm 45°$ from vertical (well outside the $\pm 15°$ failure window) reduces the specular meniscus contribution to essentially zero — visible immediately in the canvas as the collapse of red glare-hit dots and the climb of “Signal: by count” toward $100\%$ .

Remark

The practical recipe: turn off the room lights; place a single lamp at roughly the height of the dish on the bench, aimed inward at the dish at an angle of $30°$ to $45°$ above the horizontal; shoot from directly above with the camera centered over the dish. The geometric argument above guarantees that no specular path connects any meniscus point to the camera, regardless of the meniscus magnitude. The image you get is a clean darkfield rendering: the diffuse signal from the agar, the bench under it, and any sample sitting on the agar reaches the camera unobscured by mirror reflections of the lamp itself.

This is the same principle that distinguishes brightfield from darkfield illumination in microscopy. In a darkfield microscope, the condenser is positioned so that no direct light reaches the objective; only light scattered by the sample is collected. The result is a high-contrast image of the sample against a dark background. The off-axis lamp here implements the same idea at the macroscopic scale of a petri dish.

5.3 Flattening the Surface: Spread Plates and Cover Slips

A complementary intervention follows from removing the other continuous range that Lemma 3.1 relies on — the range of local normal orientations swept by the curved meniscus. If the imaging surface is flat (or close to it), every point has the same outward normal, and the geometry of the lemma collapses to a single-point condition: either the source position $S$ happens to satisfy the specular condition at the surface (producing a single bright glare spot at the geometric reflection of $S$ ), or no specular path exists at all. The configuration becomes binary rather than continuous.

Two methods implement this fix, drawn from different traditions.

The first is the spread plate technique from microbiology. Rather than leaving a poured liquid suspension as a separate layer on top of the agar — where it forms a curved meniscus and is the dominant artifact source — the suspension is spread thinly over the agar surface (with a glass rod, a sterile spreader, or sterile glass beads) and allowed to absorb into the gel before incubation. The bacterial cells are distributed across the agar surface itself, which is nearly flat (set in place when the agar solidified, with negligible residual meniscus for a properly poured plate). The dominant curved interface is removed entirely.

The second is the optical cover slip, a thin transparent plate (glass, polystyrene, or PMMA) laid horizontally over the dish, in optical contact with whatever liquid is below or with the air above. The cover slip’s upper and lower surfaces are flat to optical tolerances, presenting a single horizontal normal everywhere. The same intervention is used routinely in slide microscopy (mounting a specimen between a microscope slide and a thin cover glass), in cranial-window intrinsic signal imaging (a cover slip seated on agar above the exposed cortex stabilizes the optical path and suppresses stimulus-monitor glare — the configuration this article was motivated by), and in microfluidic-device photography (where the device is bonded to a flat substrate before imaging).

Example (Cover slips, spread plates, and cranial windows)

The same geometric trick — replacing a continuous family of surface normals with a single horizontal one — appears in three otherwise disconnected experimental settings:

Microbiology (the spread plate). A bacterial suspension poured onto an agar plate forms a curved liquid pool with a $\sim 0.5\,\text{mm}$ meniscus at the wall. Spreading the suspension with a glass rod or sterile spreader, then allowing absorption into the agar (typically a $15$ – $30$ minute incubation), reduces the liquid layer to a thin damp film and eventually to nothing. The resulting agar surface is nearly flat across the dish, and the failure mode of §5.1 is suppressed.

Light microscopy. A specimen mounted between a microscope slide (bottom) and a cover slip (top), in a mounting medium that fills the gap with matched refractive index, presents two parallel flat interfaces to the objective. Without the cover slip, the objective looks through a curved fluid drop with the same problem the petri-dish meniscus has, at higher magnification. The cover slip is so universal in microscopy that the design of nearly every microscope objective assumes its presence — the corrected working distance is computed for a specific cover slip thickness (commonly No. 1.5, $0.17\,\text{mm}$ ).

Cranial-window intrinsic signal imaging. In wide-field optical imaging of mouse visual cortex — used routinely in retinotopic mapping — a circle of skull is removed to expose the cortex, kept moist with phosphate-buffered saline (PBS) during the experiment. A camera observes the cortex from above while a stimulus monitor displays visual patterns to the mouse’s eye. The PBS forms a curved liquid layer at the edge of the cranial window that acts as a lens: it differentially refracts and magnifies the cortical signal across the imaging field, reflects parts of the stimulus monitor back into the camera as glare (by Lemma 3.1, the curved meniscus always has a point whose local normal bisects the monitor-to-camera geometry), and — because the PBS evaporates over the course of the experimental session — changes all of these effects continuously over hours of recording. The fix, due to Peichao Li (浙江大学 / Zhejiang University) and the motivating insight for this article, is to fill the cranial window with agar and seat a glass cover slip on top. The agar approximately matches the refractive index of the cortical tissue, so the agar-cortex interface contributes minimally to the optical path; the cover slip presents a single horizontal upper interface; and the configuration is geometrically stable for the duration of the session. Specular reflections of the stimulus monitor now require one specific source-monitor-camera geometric arrangement (the cover slip’s single normal must bisect monitor-to-camera, which is a single point of failure rather than a continuous arc), and that arrangement is controlled away from by the experimentalist. The same intervention has now transferred — in a different scientific domain, with bacterial samples replacing cortical tissue and a phone camera replacing the rig camera — to the petri-dish work described in this article.

In all three cases the fix is geometric and identical: replace a curved liquid interface with a flat solid one. The simulator demonstrates this most directly via the “Liquid layer” toggle. With the liquid layer on, the agar’s flat surface is overlain by a curved liquid meniscus and the failure mode of §5.1 appears. Toggling the liquid layer off collapses the geometry to just the agar surface (slightly meniscused at the wall but much flatter than a fresh liquid pool), and the artifact largely disappears.

5.4 Polarization-Based Specular Suppression

When light reflects off a smooth dielectric interface, the reflected component is partially polarized. The p-component vanishes at Brewster’s angle (§4.3, Definition 4.3) and is weaker than the s-component at every angle. Reflected light from any of the smooth dielectric interfaces in the dish system — the meniscus, the lid surfaces, the polystyrene floor — is therefore biased toward s-polarization.

A linear polarizer oriented to block the s-component (transmission axis parallel to the plane of incidence) attenuates the specular reflection substantially while passing the largely unpolarized diffuse signal with the customary $\sim 50\%$ loss. The net result is improved signal-to-glare contrast.

In practice, snap-on polarizing filters for smartphone cameras cost a few dollars and rotate manually on the lens housing; the user aligns the polarizer by sight while observing the image, rotating until the bright specular region dims maximally. For a single fixed-angle specular reflection (e.g., one off-axis lamp at $\pm 30°$ ), one polarizer orientation removes the reflection essentially entirely. For diffuse overhead illumination — where the specular reflections from different meniscus points lie in different planes of incidence — no single polarizer orientation can suppress the full crescent, but the worst contributions are still significantly reduced.

The polarization suppression compounds with the geometric interventions of §5.2 and §5.3 rather than replacing them. The cleanest practical configuration is: spread the suspension into the agar (or place a cover slip), illuminate with a single off-axis lamp at $30°$ – $45°$ , and orient a polarizer to suppress whatever specular component remains. The diffuse signal from the agar, the bench, and the sample then dominates the image; specular contributions reduce to acceptable noise.

5.5 Imaging from Below

A final intervention inverts the imaging geometry entirely. The dish bottom — a flat sheet of polystyrene, manufactured with optical tolerances much tighter than the curvature of any liquid meniscus on top — is by far the flattest surface in the system. Imaging through the bottom bypasses the meniscus and any liquid pool above it.

Two specific configurations exploit this. The first is backlit transmission imaging: a diffuse light source illuminates the dish from below, light passes upward through the polystyrene floor, agar, and any colonies, and is collected by a camera above the dish. Colonies (which scatter and partially absorb the transmitted light) appear as darker regions against a bright agar background. This is the imaging geometry used by automated colony counters (the SCAN series from Interscience, the ProtoCOL from Synbiosis, and most clinical microbiology counting systems). The light path is geometrically simple — near-vertical transmission through a stack of flat layers — and the failure mode of §5.1 does not apply.

The second is inverted microscopy: the dish is placed on the stage of an inverted microscope, with the objective looking up through the dish floor. The objective focuses on the agar surface (or any cells growing on it) from below. The curved liquid meniscus above the agar is now on the far side of the specimen from the objective, contributing nothing to the imaging path. Inverted microscopes are used routinely in cell culture, where the curved meniscus of the culture medium above the cells would otherwise frustrate observation.

Both configurations sacrifice the easy top-down view that smartphone photography provides, but in exchange they eliminate the curved-interface problem at its root. For research and clinical settings where this is acceptable, imaging from below is the most reliable solution.

6. Discussion

6.1 Cross-Modality Applications

The framework developed in this article has its origin in a different optical problem, in a different scientific domain. The cranial-window intrinsic signal imaging configuration introduced in the §5.3 Example — bacterial samples replaced by mouse visual cortex, phone camera replaced by a CCD rig, ceiling lighting replaced by a stimulus monitor presenting drifting visual patterns to a mouse — was the setting in which the geometric analysis here was first useful.

For some time, we had been focused on optimizing other parameters of that rig — camera exposure, stimulus pattern design, illumination wavelength, hemodynamic post-processing pipelines — on the assumption that the dominant sources of noise lay elsewhere. A recent conversation with Peichao Li redirected that attention sharply. He pointed out that the PBS layer above the cortex, which we had been treating as essentially invisible, was producing both differential refraction of the underlying cortical signal (a wavefront-shape problem) and specular reflections of the stimulus monitor back into the camera (a glare problem), and that both could be eliminated by replacing the PBS with agar and seating a glass cover slip on top. The intervention was tested almost immediately, and the noise floor of the imaging dropped substantially.

The same geometric reasoning then applied, verbatim, to a separate experimental setting: routine petri-dish photography for bacterial colony counting. Bacterial suspensions poured onto MRS agar form curved liquid layers with menisci on the order of $0.5$ – $1\,\text{mm}$ ; the ambient lighting in any standard lab is approximately the upper-hemisphere diffuse source modeled in §5.1; the cameras (smartphones rather than imaging rigs) are positioned directly above the dish. Every element of the failure-mode analysis transfers without modification. The interventions transfer with the same one-to-one correspondence — spread plates and cover slips for flatness (§5.3), off-axis lamps for source-direction restriction (§5.2), polarizers for the residual specular component (§5.4) — and the underlying geometric reasoning, encoded in Lemma 3.1 and the Fresnel curve, is identical across the two settings. Writing the geometry down explicitly is what made the cross-domain transfer obvious in retrospect; without that, the cranial-window insight might have stayed in the cranial-window domain.

The same structure recurs in a wider class of imaging modalities involving curved or layered liquid interfaces. Slide microscopy through cover slips is the everyday version: a thin glass plate placed over a wet specimen to convert the curved free surface of the mounting medium into a flat one, with the design of nearly every microscope objective assuming the cover slip’s presence. Fluorescence microscopy of cells in aqueous media depends on the same flatness for the same reason. Microfluidic device photography routinely bonds devices to flat glass or PMMA substrates before imaging through them. Cross-polarized photography of art behind varnish, used routinely in museum imaging, is the polarization-based suppression of §5.4 deployed at scale — a polarizer on the source and a second on the camera, oriented perpendicularly, suppress reflections off the glossy varnish coat while leaving the diffuse signal from the painting underneath largely intact. And satellite imaging through the atmosphere is the same geometry in another guise, on a vastly different scale: the atmosphere is a (relatively) flat layered medium between an opaque specimen (Earth’s surface) and a camera at altitude, and constraints on lighting and viewing angle that govern petri-dish imaging have direct analogs in remote sensing.

What unifies these settings is the recurring structure of a curved or layered fluid (or fluid-like) interface between a partially-diffuse specimen and a camera. The geometric analysis here — Lemma 3.1 for the existence of artifact-producing specular paths, the Fresnel curve for their brightness, and the catalog of interventions in §5 for their suppression — applies across the lot, with only the material parameters and length scales changing from one setting to the next.

6.2 Limitations of the Geometric Optics Framework

The geometric optics framework adopted throughout this article — light treated as rays traveling in straight lines through homogeneous media, refracting and reflecting at smooth interfaces according to Snell’s law and the Fresnel equations — is the macroscopic limit of a more complete physical theory. For the imaging problems considered here that limit is more than adequate, but several of the phenomena it neglects are worth naming explicitly.

Diffraction is the most fundamental omission. Light is a wave, and at features whose size approaches the wavelength (~500 nm in the visible) the ray approximation breaks down: edges produce diffraction patterns, finite apertures impose finite resolution, and the angular spread of “specular” reflections from a roughened surface depends on surface roughness at this scale. For the macroscopic imaging considered here — dishes $100\,\text{mm}$ across, menisci $0.5\,\text{mm}$ wide, camera lenses of order $10\,\text{mm}$ — features are six to seven orders of magnitude larger than the wavelength, and diffraction effects are not visible at the millimeter scale that matters for our analysis. The same framework applied to high-resolution microscopy (sub-micron cell imaging, fluorescence localization) would need to incorporate diffraction explicitly.

Mie and Rayleigh scattering from suspended particles in a transmitting medium are second-order effects in clear MRS agar but become significant for densely-cultured plates, plates with heavy precipitate, or suspensions of optical-wavelength particles (bacterial cells in late stationary phase, for instance, often produce visibly turbid suspensions). When scattering becomes comparable to the absorption budget of the medium, the Beer–Lambert law of §4.4 must be replaced by a more complete radiative-transfer treatment that tracks both absorption and scattering simultaneously. We have not implemented this in the simulator; for clear MRS the approximation is reasonable, but heavily scattering plates would need a more careful analysis.

Birefringence and polarization rotation in structured biological materials (collagen fibers, muscle tissue, certain bacterial polymers, agar polymer matrices under residual mechanical stress) can rotate the polarization of transmitted light by appreciable angles. This affects the polarization-based suppression of §5.4: a polarizer orientation that perfectly suppresses a Fresnel reflection from a flat interface may not perfectly suppress one that has passed through a birefringent medium. In practice the rotations involved are small enough that the suppression is still substantial, but they place a floor on how well the polarization trick can perform.

Quantum-mechanical phenomena (single-photon counting, two-photon absorption, fluorescence, phosphorescence) are operative at all imaging scales but do not bear on the macroscopic geometry that determines the artifact patterns we have analyzed. They matter for very-low-light imaging and for specialized fluorescence techniques, but not for routine benchtop petri-dish photography.

For the macroscopic imaging problem treated here, the geometric optics framework is adequate, and the analysis transfers with full validity to any imaging system whose features are large compared to the wavelength of light, whose materials are predominantly absorbing rather than scattering, and whose photon counts are well above the shot-noise regime.

References

[1] E. Hecht, Optics, 5th ed. Boston: Pearson, 2017.

[2] M. Born and E. Wolf, Principles of Optics, 7th ed. Cambridge: Cambridge University Press, 1999.

[3] P.-G. de Gennes, F. Brochard-Wyart, and D. Quéré, Capillarity and Wetting Phenomena: Drops, Bubbles, Pearls, Waves. New York: Springer, 2004.

[4] Fisher Scientific, “Lactobacilli MRS Agar,” product information sheet.

[5] P. Li (李培超), Zhejiang University, personal communication, 2026.

The choices an experimentalist makes about lighting and surface flatness are not folklore but theorems — and the theorems transfer across imaging modalities without modification.