Semi-transparency worsens depth ambiguity
Aliases: see-through depth · x-ray UI · transparent panel ambiguity
What it is
An opaque panel covering a lamp at least still casts one vote: whatever is hidden is behind. Make the same panel semi-transparent and the lamp’s contour leaks through the buttons, and that vote is gone. Transparency-induced depth ambiguity is two surfaces stacked along one line of sight: vision sees both “the panel is in front” and “the lamp is still intact,” and cannot say which layer the interface sits on. This leaf is how translucency blurs the depth reading, not translucency as a look, and not a verdict on occlusion rules.
Why it happens
Occlusion is a binary cue: if A covers B’s contour, A is in front. Semi-transparency turns the cue into a blend: each pixel is panel × α + background × (1−α). As α drops, background contours are read at their own depth, while the panel’s stereo disparity still converges the eyes on a nearer plane. Two lawful depths occupy one visual direction; ambiguity is those two depths competing to explain the pixel.
A head movement adds a third reading: the background shears faster or slower in motion parallax. Transparency keeps all three readings online, so the panel is sometimes stained glass, sometimes a film on the lamp, sometimes two objects flickering in rivalry. The lower the α and the sharper the background contour, the more reliably ambiguity beats stereo.
Studying it
Hold geometry and stereo depth fixed and change only panel opacity. Ask people to place a marker “at the same depth as the panel.” Under transparency the placement drifts toward background depth, or is revised back and forth between the two. A forced choice also works: “panel nearer than the lamp / farther / cannot say.”
Independent variables: opacity, sharpness of background contours, stereo on or off. Dependent variables: standard deviation of placed depth, rate of “cannot say,” number of revisions.
A flat design-tool preview is read as layer order and will not show stereo and parallax pulling the transparent panel toward two depths. The background must be a real object or a stereo scene with a definite depth; a solid fill hides the ambiguity.
Where it stops holding
When the background has no readable contour (fog, sky, a blank wall), transparency barely adds depth ambiguity; it only fades the panel. An opaque stroke or a solid text block locally restores an occlusion cue and shuts the ambiguity outside that backing. Monocular viewing drops the stereo end of the tug, so transparency shows up more as contrast loss than as a depth fight. An interface that is meant to be looked through in order to operate what is behind depends on the ambiguity; then the acceptance test is no longer “depth is unique.”
Applying it
- When someone must be able to say “this panel is in front of the lamp,” do not lower opacity to “feel less blocking.” Stabilise the depth reading first, then decide how see-through it can be.
- If sharp contours — window frames, text, grids — sit behind a translucent panel, raise α or move the panel off that background.
- Do not treat “looks airy” as a pass. Airy is often the vernacular name for the ambiguity.
- How to check: ask someone to touch “the depth the panel occupies” with a virtual finger. If turning transparency on lands the touch on the background object, or on two different depths across attempts, ambiguity has been worsened. On the opaque control, the touch should land at panel depth.