Abstract
Intro. The visual system's modular structure raises the question of how cues, which are analyzed in separate modules / brain regions, are brought back together (the binding problem). Previous research has shown that contour binding across attributes occurs in parallel, within attribute maps, throughout the visual field, and with an initial “default” bias towards integration (Poirier & Frost, 1998–2002).
Model. The default integration bias suggests that assemblies (and stimulus features) are integrated via fairly “automatic” facilitory connections, of low spatiotemporal resolutions. In addition, segregation occurs via “intelligent” inhibitory connections, focusing on local specific differences of higher contrast and information value. Binding is dynamic, flexible, uses multiple cues simultaneously, and propagates across assemblies and brain areas. In this model, binding influences tasks because: (1) integrated items are processed as one, reducing visual scene complexity, (2) segregated items “pop-out”, and (3) two different items erroneously bound together will share properties, leading to illusions and possibly difficult search.
Implications. This theoretical framework requires re-interpretation of data for many tasks. (1) Visual search measures the efficiency with which a target segregates while distractors integrate together. (2) Redundancy effects measure a stimulus' ability to stimulate the integration mechanisms. (3) Illusory conjunctions are found when integration occurs but segregation fails, despite spatial separation of items. (4) Texture segregation differs from single item similarity judgments because the former also uses spatial effects like collinearity (integration) & edge detection (segregation).
Discussion. The model provides a unifying framework that reconciles different results and research areas. Thus, a meta-analysis of several different tasks should reveal many specific integration & segregation cues, along with their spatiotemporal properties.