Color Difference in Practice: Tolerances and Pass/Fail QC

Every brand red, every batch of paint, every run of printed packaging faces the same question: is this close enough to the standard? Human "good enough" is unreliable - it drifts with the lighting, the observer, the time of day, and who is arguing. So the color industry replaced opinion with measurement: instruments report each color as CIELAB coordinates, and a color difference formula condenses the gap between standard and sample into one number, ΔE.

A number alone isn't a decision, though. You still need a tolerance - the largest ΔE you'll accept - and that tolerance has to reflect how people actually see, which is not uniform across colors. The story of practical color difference is the story of making ΔE and its tolerance match the eye well enough to bet a production run on.

ΔE measures the perceived distance between two colors in a perceptual space. The original ΔE76 is just the straight-line (Euclidean) distance in CIELAB - simple, but it over- and under-states differences because Lab isn't perfectly uniform. ΔE2000 fixes this with weighting terms for lightness, chroma, and hue, plus a rotation that handles the tricky blue region. The three formulas can disagree sharply - compare them on your own pair.

QC is one comparison repeated forever: measure the sample, compute ΔE2000 against the standard, and compare it to the agreed tolerance. Under tolerance, it ships; over, it's rejected or reworked. The whole game is setting that threshold sensibly - tight enough to protect the brand, loose enough that production can actually hit it. Set a tolerance and test a sample.

Two different tolerances hide inside "close enough." Perceptibility is the smallest difference you can see - around ΔE 1, the just-noticeable difference (JND). Acceptability is the largest difference a customer will tolerate - often ΔE 2-3 or more, because a visible difference can still be commercially fine. QC almost always uses acceptability, set by agreement for the product. Slide the difference and find your own thresholds.

Here is the subtle part. If ΔE were a true Euclidean distance, an acceptance region would be a circle (a sphere in 3D) around the standard. But we are not equally sensitive in every direction: we tolerate more difference in chroma and lightness than in hue, and sensitivity shrinks as colors get more saturated. ΔE2000 builds those weights in, so its equal-ΔE boundary is a tilted ellipse. A naive ΔE76 circle passes colors ΔE2000 would fail, and vice versa. See the two regions.

A working color-control program is more than a number and a threshold. The pieces that make pass/fail meaningful and repeatable:

Six questions on ΔE, tolerances, perceptibility, and QC. Instant feedback, no scores recorded - a wrong answer comes with a short explanation.

Color difference sits on the space it's measured in, the instruments that read it, and the workflow that acts on it - here's where to go next.