Use discrete clicks for blurred-image attention maps

Collect discrete clicks instead of continuous mouse movements when you want an importance map from a blurred image. For example, use click-to-reveal bubbles rather than recording every mouse path so the data captures deliberate points of interest instead of transitions between them.

Continuous movement records both destinations and the paths between them. Clicks remove most of that transition noise and turn each sample into an explicit choice.

Mechanism: Clicking adds a small effort cost. That cost makes viewers more selective, so the collected samples align better with the regions they consciously choose to inspect.

Evidence: BubbleView clicks matched or exceeded continuous mouse-movement approaches at approximating eye fixations on natural images for feasible participant counts, and the paper shows that movement data includes noisy transition traces that require post-processing (Kim et al., 2017).

Notes: The paper also reports lower computational and post-processing cost for click collection.

• User Goal: Recover the most important regions of a static image or approximate eye fixations remotely.
• Task: Run a mouse-contingent blurred-image study online.
• Chart Setting: Click-to-reveal or move-to-reveal interface on a static image.
• Success Criterion: The map highlights deliberate regions of interest with fewer participants and less post-processing.

Break it when: You need broader sampling of image regions rather than only the most important ones. Why: Clicking is more selective, so some regions that would receive a quick glance or pass of the mouse may never be sampled.

Sacrifice: Clicking is slower than moving the mouse. Risk: Fewer regions may be explored in the same amount of time. Mitigation: Give participants more time per image when broad coverage still matters.

Mistake: Treating all continuous mouse samples as equally meaningful points of interest. Why it fails: The raw path contains transitions between destinations, not just the destinations themselves.

Failure Sign: Raw traces form long arcs and loops between a small number of hotspots. Quick Check: Visually inspect a few raw sessions and look for path-like traces between meaningful regions. Stronger Test: Compare how many participants are needed before the hotspot map stabilizes under clicks versus movements.

• Replace continuous reveal with click-to-reveal bubbles.
• If you must keep movement data, convert the trajectories into discrete points of interest before analysis.
• Increase the viewing time to offset the slower pace of clicking.
• Rebuild the map from the discrete selections rather than from the full paths.