Change Blindness Test

What the test asks

Two versions of a drawn scene take turns on the screen with a short grey blank between them. One thing differs: its colour changes, it is there and then gone, or it moves. Without the blank the change would jump out at once. With it the whole picture flickers and you find the change only when your attention gets to it.

This is the flicker task of Rensink, O'Regan and Clark (1997), with their timing: a picture for 240 ms, grey for 80 ms, the same picture again, grey, then the changed picture twice in the same way, over and over, for up to 60 seconds. Before each scene the screen stays grey for 1 second (the original added a 3-second grey field with a white rectangle before it).

Main objects and background details

Rensink and colleagues asked five people to describe each photo. A part that three or more of them mentioned was a central interest, a part nobody mentioned a marginal interest. Changes to central interests were found much faster, although the marginal changes were on average more than 20% larger.

This page draws its own scenes and sets the roles by a rule, without asking anyone. The main object is the one thing of its kind, large and in front: the car on a street, the kettle on a kitchen shelf, the person in a park, the monitor on a desk. A background detail is one of several similar things: a window, a mug, a tree on the horizon, a book on a shelf. The rule is a guess about what draws attention, not a measurement of it.

How the scenes are kept fair

• A run has 18 scenes: 3 kinds of change × 2 roles × 3, in random order, at most 3 of one role in a row.
• The part of the picture that changes covers 3 to 5% of the scene, measured by comparing the two pictures point by point. In each pair of scenes with the same kind of change, the detail's change is 1.0 to 1.25 times as large as the main object's, so a slower detail is not explained by a smaller change.
• 9 changes are in the left half of the scene and 9 in the right, and for each role 3 are near the middle, 3 in a ring around it and 3 near the edges.
• A colour change keeps the lightness: the colours differ by at least 0.15 in the OKLab colour space, their lightness (0 black, 1 white) by at most 0.05.
• A moving object goes sideways by at least the width of its bounding box and at least 4% of the scene width. No other object's box touches its box before or after the change, and it is never the only object in its ninth of the scene.
• Each scene grows from a random seed. The page keeps the seeds of your last 200 scenes on this device and does not use them again.

How the result is calculated

The time runs from the first picture of the flicker to your press of I see it. A scene counts as found only when you then click inside the changing object, before or after the change, or within a margin of 3% of the scene width around it. A wrong place, or 60 seconds without an answer, counts as not found.

An alternation is one swap of the pictures, every 640 ms in the original. Here pictures last whole frames of your screen: at 60 Hz a picture is 14 frames and a blank 5, so they swap every 633 ms.

For each role the page takes the median of the 9 times; scenes not found count as longer than 60 seconds. Rensink and colleagues averaged only correct answers, but leaving misses out makes the slower role look faster. The main number is the median for background details minus the median for main objects, with a 95% percentile bootstrap interval from 2000 new sets of 9 and 9 times drawn from yours.

With 9 scenes per role the interval is wide. For a simulated participant whose times had medians of 4 s for main objects and 9 s for details, the interval was typically about 15 s wide and held the true difference in about 95% of 2000 simulated runs (simulation of 1 November 2026). One run shows whether the difference is large, not its exact size.

Why the grey blank matters

Rensink and colleagues concluded that we do not hold a detailed picture of the whole scene: a change is seen only when attention is on the object while it changes. Normally the change makes a local jump that pulls attention to it; the blank hides that jump in a flicker of the whole screen.

With flickering rectangles, Rensink (2000) found that blanks of 80, 120 and 160 ms gave about the same results, and he cites earlier work in which 67 ms may already be enough. Here a blank lasts about 80 ms (never under 67 ms) and at least 3 frames. If a skipped frame cuts a blank between the two different pictures below 67 ms in the 2 seconds before you press I see it, the scene is not counted and a new one of the same kind comes later.

Compared with the original study

10 observers per experiment, 48 colour photos of real scenes 27° wide and 18° high, a key press followed by a spoken description, 1.2% errors. Without the blanks, central and marginal changes did not differ. Drawn scenes, roles set by a rule and a different screen make these values landmarks, not norms.

Things to try

• Choose No blank in Settings: the pictures swap directly, each for about 640 ms. In the original the changes were then found in 1.4 alternations.
• Compare the two lines in the chart of This run. How far apart are they at 10 s?
• Next time, search the background on purpose. Does the difference shrink?
• In Show the numbers, compare colour, there and gone, and position.
• Repeat the test another day and see in History how much your difference moves.

Why there is no percentile

In a study by Andermane and colleagues (2019), 59 people did a flicker task with 30 indoor scenes twice, 1 to 4 weeks apart, with other scenes the second time. Their scores correlated only r = 0.45 and were lower the second time, which the authors put down to harder scenes. So the page compares you only with your own runs.

Limits

• Drawings, not photos. Flat colours, few objects and no texture make the scenes simpler than photos, so your times can differ a lot from the published ones.
• The roles are a rule. Main object and background detail are this page's choice, not what people actually look at first.
• Screen size. The original scenes were 27° × 18°, about 29 cm wide at 60 cm. Compare only runs on the same screen.
• Colour vision. A colour change keeps the lightness, so with red-green colour blindness some colour changes can be very hard to see.
• Learning. Once you know that background details change too, you may search them on purpose, and the difference shrinks.
• Your screen. Below 50 Hz or with uneven frames the page warns you and offers Run anyway; the run is marked less precise and kept apart in History. Below 30 Hz it does not run.
• A hidden tab. The scene in progress is discarded and the run waits for Resume.
• Flicker. The whole scene alternates with a grey field about 3 times a second, for up to a minute per scene. If flickering bothers you, choose No blank or stop with Esc.
• Not a medical test. It does not test eyesight, attention problems or fitness to drive.

Controls

• Keyboard: Space or Enter = I see it. On the still picture the arrows or Tab move a frame over the objects, and Enter or Space picks the framed one. Enter on Next scene goes on; Esc stops the run.
• Mouse and touch: I see it, or click or tap the scene while it flickers; then click or tap the place that changed.
• Pauses: for 0.7 s after a result or Stop the buttons are dimmed and ignore presses.
• Language: the link at the end of History switches the app to Slovak.

Sources

• Rensink, R. A., O'Regan, J. K., & Clark, J. J. (1997). To see or not to see: The need for attention to perceive changes in scenes. Psychological Science, 8(5), 368–373. doi:10.1111/j.1467-9280.1997.tb00427.x
• Rensink, R. A. (2000). Visual search for change: A probe into the nature of attentional processing. Visual Cognition, 7(1–3), 345–376. doi:10.1080/135062800394847
• Andermane, N., Bosten, J. M., Seth, A. K., & Ward, J. (2019). Individual differences in change blindness are predicted by the strength and stability of visual representations. Neuroscience of Consciousness, 2019(1), niy010. Full text (PMC6345093)