Abstract
Percepts under stressed visual attention (e.g., brief peripheral presentations) contain systematic errors that suggest perceived objects correspond to independently sampled features (Vul & Rich, 2010). Here we investigate the possibility of a hierarchy of binding: are colours free-floating features within a spatial region, or are they associated to some attribute of the objects? To test if object parts are mapped to colour before features are bound to complete objects, we asked subjects to report the two colours of a multi-part object. We examined several object types, such as a bulls-eye with one central colour and another as the surround, and a cross with differently coloured vertical and horizontal bars. A cued target object was presented in a circular array of similar distractors arranged around fixation. The whole array was briefly presented and participants were asked to report both colours of the target (e.g., the colour of the inner and outer circles). The colours of object parts near the target object were chosen to be unique, so incorrect reports of target letters could be identified as intrusions from a particular part of a particular object. In this task, binding can break down at different levels: features, parts, or objects, each predicting different patterns of error responses. We find that across a large number of object types, binding does appear to be hierarchical. Errors occur as entire misreported objects (10% of the time on average), as object parts with correct binding of colours to substructure (12%), as colour intrusions from adjacent objects (67%), and as guesses (.11%). However, for objects with more complex part arrangements, intrusions of bound sub-parts reduce to zero. Together, these results suggest that the visual system binds features into objects hierarchically and that conscious perception amounts to sampling at a number of different levels of object representation.
Meeting abstract presented at VSS 2013