Visual similarity effects in categorical search

Area of Science:

• Cognitive Psychology
• Computer Vision
• Human-Computer Interaction

Background:

• Visual search is fundamental to daily tasks.
• Understanding how visual similarity influences search is crucial for designing effective interfaces and information retrieval systems.

Purpose of the Study:

• To investigate the impact of visual similarity on search guidance for categorically defined targets without visual previews.
• To determine if purely visual similarity, independent of category knowledge, can guide search behavior.
• To compare human similarity judgments with computer vision model outputs in predicting search performance.

Main Methods:

• Collected human visual similarity rankings for target categories (teddy bears, butterflies) and distractor objects.
• Created search displays with varying levels of distractor similarity (high, low, mixed).
• Analyzed manual response times and fixation patterns on target-absent trials.
• Validated findings using similarity estimates from a computer vision model based on color, texture, and shape.

Main Results:

• Search displays with low-similarity distractors led to faster responses and fewer fixated distractors than high-similarity displays.
• In mixed displays, initial fixations were disproportionately drawn to high-similarity distractors.
• Results were consistent whether using human similarity judgments or computer vision model outputs.
• Search effects were best predicted when human and model similarity rankings agreed.

Conclusions:

• Visual similarity, independent of explicit category information, effectively guides visual search.
• Both human perception and computer vision models capture relevant visual features for guiding categorical search.
• The findings have implications for optimizing visual search interfaces and understanding human visual processing.