We tested the hypothesis that object recognition is an active search of complicated fragments in the visual image. This search is performed in accordance with criteria based on invariant descriptions of an object's perceptual class. The basic strategy is to activate these descriptions during parallel search. ‘Upper’ segments search for appropriate fragments of the picture. ‘Subordinate’ segments are included by request of the ‘upper’ segments. Description segments include three types of records: integral (whole) characteristics of some fragment; characteristics of fragments which are members of this fragment; and characteristics of relations between the fragments. This structure of perceptual description permits parallel analysis of the visual scene by different segments by the ‘autonomy’ principle and permits the use of incomplete sets of segments for recognition by the ‘quorum’ principle. Different ways of forming connections between segment records may be considered as ‘thinking’ components of visual perception. The main points of our model follow from results of our tachistoscopic experiments. We measured thresholds for the recognition of test figures. Different levels of figure complexity were used: parallel lines and strips, geometric figures, schematic faces, textures, etc. It was found that the stages of the recognition process are connected with types of operations described in our model. These results give rise to the possibility that the properties of the neurons involved in visual search might be identified.
Hemispheric Specialization Effects in Visual Image Generation
