Abstract
Color is widely recognized as a key guiding feature during search for objects defined by one or two colors. However, real-world objects are oftentimes defined by many colors in varying proportions. In three experiments, we examined how color proportions directly affect guidance. Experiment 1 compared guidance for two colors utilizing color conjunction stimuli distributed as 70%/30% or 90%/10% with the predominant color denoted as primary and the second predominant color denoted as secondary. We presented three key distractor types: swapping primary and secondary (inverted), primary only (primary), and non-target colors (neutral). For the 70/30 conjunction, proportionally more first saccades were directed to inverted over primary distractors (p = .02). Conversely, for the 90/10 conjunction, more first saccades were directed to primary compared to inverted distractors (p < .001). Overall, there were significantly more first saccades to color-matching distractors than neutral distractors (all ps < .001). Experiment 1’s results indicate that when the color proportion within an object is sufficient, both colors guide attention; when the proportion is insufficient, the predominant color guides attention. Experiments 2 and 3 examined the effect for real-world objects by using a color histogram to quantify the color proportions and exact color coordinates for each object. To generate color-matching distractors, we grouped distractors with targets by color via a nearest neighbor machine learning algorithm. Keeping lightness constant, we transformed the distractor colors to the target’s primary or secondary color. To prevent guidance by shape, a recolored target object replaced the inverted distractor. In Experiment 2, we confirmed that search was guided by the primary color for real-world objects (p < .001). In Experiment 3, proportionally more first saccades were directed to primary over secondary over neutral distractors (all ps < .001). Overall, the data suggest that color proportions affect guidance in real-world objects.