Horizontal stripes
Reflective display · color science · software
Rendering color with six e-paper states
How can a display reproduce photographs when each pixel has only six color choices?
I measured the panel's color, tested an experimental ICC workflow and built an image renderer.
The hardware: Pimoroni Inky Impression 7.3-inch Spectra 6, an 800 × 480-pixel reflective display.
How this e-paper differs from an RGB screen
R 64
G 160
B 208
Light from the screen · fast updates
BlackWhiteRedYellowGreenBlue
→Reflected light · slower updates · holds its image without power
To suggest intermediate colors, the renderer spreads the six states across neighboring pixels. This is halftoning, or dithering. The arrangement matters, which is why I measured patterns as well as individual colors.
Schematic controls and enlarged pixels, not a simulation of the panel's appearance.
An image renderer guided by measured colors
My renderer uses measured references for the six native states and distributes color error between neighboring pixels to approximate the source image.
Software previews, not panel photographs. Both routes receive the same interpreted, fitted source and use one visualization palette based on measured native-state colors scaled to panel white. These show pixel choices rather than predicting reflected color.
Katsushika Hokusai, Under the Wave off Kanagawa (The Great Wave), ca. 1830–32. The Metropolitan Museum of Art, JP1847. Public domain. Only side padding is hidden; the artwork is not cropped.
Across eight development images, the renderer reduced mean calculated reference error on every image compared with the matched Pimoroni route, although some local regions worsened. Physical color comparison is pending.
Same color counts, different measured colors
Both patterns use 50% black and 50% yellow pixels. I changed the stripe direction and measured the display. Horizontal stripes measured lighter; vertical stripes measured darker.
Vertical stripes
Same amounts of black and yellow. Different measured colors. The two mean colors were 18.24 ΔE00 apart, so color counts alone could not describe both results.
Left in each pair: enlarged digital pixels. Right: approximate sRGB previews from measured XYZ, not panel photographs. These show the direction of the color difference; your screen cannot reproduce the display's exact appearance.
Measurement details and limits
In this one-pixel black/yellow comparison, four i1Pro 3 readings were taken, two per pattern; D50, CIE 1931 2° observer and nominal perfect-diffuser D50 Lab. ΔE00 expresses the distance between the two measured mean colors. The comparison does not isolate a physical cause, and other color pairs showed smaller differences.
Color management reduced error on a measured grid
I used an experimental ICC profile to convert image colors before the same Pimoroni renderer. On a separate 133-color grid, mean measured error fell from 19.99 to 11.20 ΔE00, using panel white as the Lab reference.
Digital pixel maps visualized with measured native-state colors scaled to panel white. The numerical result comes from the color grid. Vincent van Gogh, Cypresses, 1889. The Metropolitan Museum of Art, 49.30, Rogers Fund, 1949. Public domain.
Prediction improved in some tests but did not transfer reliably
Forecasts saved before measurement succeeded for selected exact patterns. For six regions in three new images, a learned model was closer than the additive baseline on all 12 readings, but neither met the chosen engineering limit of 1 ΔE00. Reliable prediction across new colors, returns and image regions remains unresolved.