Color-Motion Feature Binding Errors Across Qualitatively Different Motion Percepts

Our visual experience of the world is one where the color and motion of objects seem inextricably linked, a percept that seamlessly spans from edge-to-edge of our visual field. Considering that motion detection can be twice as fast as color detection in the central visual field, with magnitudes of even greater difference in the periphery, the aforementioned subjective visual coherence of an object’s color and motion is a neurally constructed one. Feature binding, which is separate from individual feature encoding, is a necessary step in bridging the relatively distinct neural pathways and temporal dynamics of color and motion perception.

Perceptual grouping based on similarity is a plausible mechanism for organizing feature binding, but complex stimuli can be related in a multitude of ways based on this principle. For example, the similarity could be based on low-level features of stimuli, such as their shape, size, color, speed, and motion direction, or based on a higher-level interpretation that integrates low-level stimulus features into coherent perceived objects. This dissertation investigates whether color-motion feature binding is driven by higher-level representations of an object and/or lower-level stimulus features of an object. Note that contributions from higher- and lower-level representations are not mutually exclusive.

The experimental approach here exploits a known procedure that causes steady and stable feature-binding errors in the peripheral visual field. Specifically, errors in feature binding can be induced in the periphery by presenting an observer with two overlaid sets of differently colored dots moving in opposite directions — for example red dots moving vertically upward and green dots downward in the central visual field, but with directions reversed in the periphery (red dots moving downward and green dots upward).  Often, all red dots are perceived to move in the same direction (upward, in this example) and all green dots downward. The experiments here measure the proportion of viewing time during a 15-sec test trial when these steady, stable feature-binding errors are perceived.

Across three experiments, peripheral feature-binding errors were measured using three distinct motion percepts — 2D vertical linear motion, 3D radial optic flow motion (as in looming), and 3D rotating structure-from-motion cylinders — in order to characterize how the visual system organizes feature binding when a great amount of low-level feature similarity conflicts with disssimilar higher-level object motion (2D planar, 3D optic flow, or 3D structure-from-motion).

Experiment 1 found that steady peripheral binding errors persisted even with unequal low-level features of the stimuli creating vertically perceived linear motion. Experiment 2 showed that qualitatively distinct high-level percepts of 2D radial motion together with 3D optic flow motion reduced peripheral binding errors despite the preponderance of identical low-level features for both types of motion stimuli. Experiment 3 replicated the influence of a high-level percept using a 3D structure-from-motion vertically rotating cylinder.

Taken together, the experiments show that color-motion feature binding errors are clearly affected by the similarity of higher-level object percepts, even in the presence of a large degree of similarity among lower-level stimulus features. The implicit perceptual ambiguity in an ensemble of low-level stimulus features suggests a perceptual advantage to rely, at least in part, on higher-level object representations for feature binding processes.