Vinnie Garcia

Assist cam ๐Ÿ‘๏ธ

Phones have built-in color filters too (iOS: Accessibility, Display and Text Size. Android: Accessibility, Color correction). This is a demo, not a medical device. Tap the picture to hide or show these controls.
How it worksBack to the demo

Making differences easier to see

Assist Cam can change colors or draw stronger outlines around them. Simulate shows the output of a color vision deficiency model. Color assist tries to move lost color differences into other channels. Outlines works from brightness instead, putting bold marks along edges.

They answer different questions. A simulation can show which colors a model brings closer together; assistance tries to separate them again. Neither is a vision test, a medical device, or a promise that a setting will help every viewer.

Color lost, color moved

Protan, Deutan, and Tritan select models involving long-, medium-, and short-wavelength cone responses. The shader first converts the camera's sRGB colors to linear RGB, then applies a matrix from Machado, Oliveira, and Fernandes's A Physiologically-based Model for Simulation of Color Vision Deficiency (2009).

Those are fixed severity-1.0 matrices. Strength controls how much assistance is added, not the severity of the simulation.

For Color assist, the code subtracts the simulated color from the original. It moves some of that difference into the green and blue channels, adds it back, and clips anything outside the displayable range. This is the Fidaner-style error redistribution identified in the shader. All three types share the same redistribution rule, including Tritan; it isn't fitted to an individual viewer or separately tuned for every deficiency.

Compare neighboring colored objects as you change Strength. Making the whole image more vivid doesn't necessarily make the distinction you need clearer. For a baseline, use Normal with Contrast at 100%. Contrast also changes Normal.

Outlines have different limits

The outline pass smooths brightness and measures horizontal and vertical changes with Sobel filters. Sensitivity lets weaker changes through. Line weight widens the marks by taking the strongest nearby edge within a disk. Color, Dimmed, and Yellow on black change what sits behind those marks.

A shadow or reflection can get an outline. A low-contrast obstacle can get none. The code sees contrast, not objects, so don't rely on it for navigation or other safety-critical tasks. Lighting, screen calibration, and individual vision also limit the color comparisons.

The edge detector explainer lets you inspect that calculation pixel by pixel. For a simpler way to separate lightness from color, Value Finder groups a scene into a few gray tones.