Perception & Vision User Guide

• Tool URL: open Perception Vision.
• Default state: the built-in test card, colour vision at protanopia, severity 1.00, split view, D65 — that is, the scene as authored — and the GPU renderer.
• Six tabs: Lab is the workbench. Simulation shows where a colour goes on its way through the projection. Modes shows every entry over the same source. Data is the register of what each figure rests on. Export takes a list of colours, the frame, the settings or a link. Reference holds the standards, the formulas, the citations and the research notes.
• Processing: everything runs on your machine. Without WebGL2 the page runs the same simulations on the CPU and says so — it used to show nothing at all.

Nothing is uploaded. Files and camera frames are decoded into a local canvas and handed to your own GPU, and nothing leaves the page until you press an export button.

The CSF overlay that used to be on this list is gone. It multiplied the image by the Mannos-Sakrison contrast sensitivity curve and handed that curve the pixel's distance from the centre where it takes spatial frequency, so at its own defaults it returned a black frame. The Modes tab shows what it did and says what a real one would take.

• Severity: 0 leaves the source alone, 1.00 is full dichromacy. In between it is a linear blend toward it — a convention of this tool, not a model of anomalous trichromacy.
• Type: protan, deutan or tritan. Daltonise pushes what the projection lost back into the channels that survive, instead of simulating the loss.
• Adaptation level: where the mesopic crossfade and the light-adaptation gain sit.
• Age: 20 to 80, for lens yellowing.
• Scotoma radius: the size of the central-loss disc.
• Cataract: scatter, yellowing and contrast-loss strength together.
• Acuity: the denominator of a Snellen fraction, 20/20 to 20/200.
• Compare: split, simulated only, or difference. Both halves of a split get the same adjustments, so the only thing that differs across the divider is the simulation. Drag on the image to move the divider.
• Renderer: GPU or CPU. The same maths either way — the colour transform composes into one matrix that both use — and the Lab checks the two against each other after every change.
• Before the eye: the illuminant, D65, D50 or A, with a strength. Bradford adaptation is applied to the stimulus before the visual system, because that is the order light arrives in. Under illuminant A the scene leaves sRGB by a wide margin; that is the correct answer and the readout says how much.
• Adjustments: contrast, gamma, blur, glare and noise. These are not simulations and the page does not count them as ones.

• The grey ramp: nine steps with the channel spread the running simulation introduces into each, in 8-bit code values. Under colour vision it has to be zero all the way along, because a dichromacy projection collapses one hue axis and leaves the achromatic one alone. Under scotopic a non-zero row is the effect, not a defect, and the note under the strip says which case you are in.
• Neutral drift: the worst of those nine figures.
• Colours merged: how many of the frame's own distinct colours the simulation brought to the same place. This is the thing a colour-vision simulation is for.
• Out of sRGB: how much of the frame the projection pushed outside the gamut, rather than letting the clamp hide it.
• GPU and CPU agree: the two renderers are run over a grid of 216 colours and compared. It reads n/a — spatial for a mode that reads its neighbours or its position, because a strip of flat colours cannot check one.
• Frame luminance: mean, minimum and maximum of what is on screen.
• Enlarge: double-click the viewport.

The Export tab takes hex colours, one per line, and runs each through the simulation the viewport is showing — the same code, not a second one that drifted from it. The table gives the simulated hex, the LMS cone responses, ΔE₀₀ against the input, and whether the result stayed inside sRGB.

Put a grey in any list you check. It is the row that catches a simulation tinting neutrals: a dichromacy projection has to leave it exactly where it was, at ΔE 0.0.

Library registers Perception Vision as a Perception tool with preset capture. When a canvas is present, the first tool canvas is stored as a perception snapshot. The shared runtime hook reads window.AAPerceptionVision.getState() and restores through window.AAPerceptionVision.restoreState(saved).

Use Library saves for repeatable accessibility review cases, UI audit snapshots, design-system comparisons and teaching demos.

• Accessibility review: upload a screenshot, run protan, deutan and tritan at severity 1.00, watch colours merged, and export the CSV for the palette.
• Palette check: paste your brand hexes into the Export tab with a grey among them and read the ΔE₀₀ column — two rows that arrive at the same hex are two colours your reader cannot tell apart.
• Low-light check: scotopic or mesopic, lower the contrast, add a little glare, and compare in split view.
• Ageing review: lens at 70 and cataract, to test blue-dependent labels and low-contrast affordances.
• Acuity check: the Snellen chart as the source, then walk the acuity slider from 20/20 to 20/200.

• This is an educational and design-review simulator, not a clinical diagnostic tool. Nothing it prints is a measurement of anybody's vision.
• The dichromacy matrices are the published Viénot 1999 values and are not individualised to an observer. Severity between 0 and 1 is a blend, not a model of anomalous trichromacy; Machado 2009 is what that would take.
• The rod weighting is this tool's own and is not V′(λ). No scotopic luminance is reported from it.
• The lens, cataract, acuity and light-adaptation models carry this tool's own constants. They show that an effect exists; they carry no clinical meaning.
• There is no contrast sensitivity function here. The mode that claimed one was withdrawn, and nothing on the page may be reported as a contrast-sensitivity result.
• Camera access depends on browser permission and a secure context. WebGL2 depends on the device; without it the page runs the same simulations on the CPU.
• Uploaded image content is session-local; Library restores settings, not file bytes.

The Data tab carries all of this as a register, entry by entry, with a status of verbatim, computed, stand-in or absent, and says what each absent entry would take. For architecture and implementation details, see the Perception Vision Developer Reference.