Mach banding in smooth gradients gets misread as color banding or posterization defects

What it is

Gradients in digital images and interfaces — background gradients, shadow transitions, data-visualization color scales — are stored as a finite number of quantized luminance levels. Even when the physical luminance difference between adjacent levels is already small enough to be nearly imperceptible on its own, the visual system still triggers edge enhancement at each level boundary, exaggerating these otherwise-subtle steps into visible narrow bands. Engineers and designers often misdiagnose this as "insufficient color depth," "rendering precision loss," or "a compression banding artifact," but in many cases the quantization is already dense enough and the rendering has lost no precision at all — the visible bands are purely a subjective amplification of an already-existing tiny step, a phenomenon called false contouring.

Why it happens

A gradient is effectively made up of many very narrow bands laid end to end, each one luminance step brighter than the last. As long as there's even a tiny luminance difference between adjacent bands, each boundary satisfies the condition for edge enhancement — a receptive field straddling a luminance step — so every boundary produces a slightly brightened fringe on one side and a slightly darkened fringe on the other. When the quantization step is small enough and the number of levels high enough, the physical size of any single step usually falls below the threshold ordinary contrast sensitivity could detect directly. But edge enhancement's response to a step doesn't simply scale down proportionally to zero as the step shrinks — it can still amplify a step that shouldn't be visible on its own into a subjectively clear, narrow fringe. The result is that an otherwise continuous, smooth gradient gets perceived as a series of countable stripes. This explains a counterintuitive fact encountered when troubleshooting the issue: the measured quantization step can be far below the normal luminance-discrimination threshold, and yet banding is still clearly visible — because what's being seen isn't the step itself, but the amplified result of edge enhancement acting on that step.

Studying it

Imaging and display engineering fields systematically determine the psychophysical threshold for "how many quantization levels, at what gradient slope, before false contouring becomes noticeable," by presenting gradient patterns at different bit depths or different spans/slopes and having participants judge whether stripes are visible, or adjust a parameter until the stripes just disappear — yielding a picture of how the minimum imperceptible bit depth varies with gradient slope, display dynamic range, and viewing distance. A key consideration in this line of research is that simply back-calculating the required bit depth from ordinary luminance-discrimination thresholds tends to underestimate what's actually needed, because edge enhancement amplifies a step that would otherwise sit below threshold up above it — so the bit depth that's empirically found to be "enough" typically ends up higher than one estimated from raw luminance discrimination alone, ignoring edge enhancement. Typical independent variables are bit depth/number of quantization levels, total gradient span and slope, and display dynamic range; typical dependent variables are detection rate for perceptible banding and participants' subjective acceptability ratings.

Where it stops holding

• This entry concerns the case where the quantization step itself is already fine, the physical step is below the ordinary discrimination threshold, and edge enhancement is what amplifies it into visible bands; if the number of quantization levels is genuinely too low and the physical step already exceeds the ordinary discrimination threshold, that's a real insufficient-bit-depth problem, not the false contouring discussed here.
• How noticeable false contouring is depends directly on the gradient's slope, span, and the dynamic range the display can actually render; the same bit depth can be clearly visible on a short, steep gradient and completely imperceptible on a long, gentle one — there is no single "safe" bit-depth figure that holds regardless of the gradient's specific parameters.
• Viewing distance and display size also change the outcome, since they determine the spatial frequency a gradient's steps correspond to on the retina; the same asset can show visibly different amounts of banding when viewed small and close versus large and far.

Applying it

• When a report of "I can see rings of stripes in this gradient" comes in, or you notice this yourself, first read the actual rendered or stored luminance/color values point by point along the gradient, rather than assuming it's a bit-depth shortfall and jumping to a higher bit depth or a different encoding. If the values themselves increase continuously and smoothly across the whole gradient with no large runs of repeated values, the quantization itself is fine and the stripes are false contouring from edge enhancement — increasing bit depth will often have limited or no effect.
• Only when the values genuinely show large flat runs that jump at just a few points is it a true quantization shortfall, which does need to be addressed by raising bit depth or switching to a finer encoding.
• Verification: produce several versions of the same gradient, holding bit depth constant while varying only the total span or slope, and compare how visible the stripes are as slope changes. If the stripes get more visible as the gradient steepens and disappear as it flattens, while the underlying numeric bit depth never changed, that confirms the stripes come from edge enhancement rather than quantization precision.

Related

• Same group: A1.31.1 The visual system actively boosts contrast at luminance step edges, producing Mach bands · A1.31.2 Edge enhancement comes from lateral inhibition, not a change in physical luminance · A1.31.4 Adding smoother transitions or dither noise weakens edge-enhancement artifacts
• Nearby: A1.04 Contrast Sensitivity and Spatial Frequency Channels
• Search terms: false contouring · posterization · Mach band · banding artifact