Perceptual Science Series
Studying color constancy using natural tasks
Dr. Ana Radonjic
Monday, November 03, 2014, 12:00pm - 07:00pm
Post-doctoral Fellow, University of Pennsylvania, Department of Psychology
In everyday life we rely on color information to select objects as the targets of our actions (e.g., the ripest fruit, the freshest fish). To be useful for such tasks, our perception of object color needs to correspond to object surface reflectance across a wide range of contexts in which the object is viewed. It is well known that the visual system partially compensates for the variable effects of context (such as changes in illumination) to maintain certain degree of stability of object color appearance, but whether this color constancy supports accurate object selection is not understood.
To study how color constancy guides real-life object selection we developed a paradigm in which we asked subjects to select objects based on color across an illumination change. In a basic color selection task, the subjects selected which of the two test objects presented under one illumination appeared closer in color to the target object presented under a different illumination. From subjects’ choices, we inferred a cross-illumination selection-based match for the target via a variant of the maximum likelihood difference scaling method, and used the inferred match to quantify constancy. Our results show that selection-based constancy is good when measured using naturalistic stimuli, but is dramatically reduced when the stimuli are simplified, indicating that a certain degree of stimulus naturalness is critical for good constancy.
We used our basic color-selection task as a building block in the development of more complex goal-directed tasks for which color is instrumental. In a blocks-copying task, for example, we asked subjects to recreate an arrangement of colored blocks across a change in illumination. Our results show that color constancy mechanisms support accurate cross-illumination object selection in this task.
Our current studies aim to extend our understanding of how constancy operates in real-life tasks by probing how variations in stimulus context modulate selection-based constancy and whether naturalistic appearance-based feedback can affect subjects’ performance.