Logic in Visual Brain: Compute to Recognize Similarities-Formalized Anatomical and Neurophysiological Bases of Cognition

Object recognition is a complex neuronal process determined by interactions between several visual areas: from the retina, thalamus to the ventral visual pathway. These structures transform variable, single pixel signal in photoreceptors to a stable object representation in higher areas of the visual cortex. Neurons in macaque monkey area V4, midway in ventral stream, represent such stable shape detector. Traditionally these processes are described as feed forward hierarchy of increasing in size and complexity receptive fields (static spatiotemporal filters). A fundamental question in visual neuroscience is how these processes might identify an object or its elements in order to recognize it in new, unseen conditions? We propose a new approach to this problem by extending the classical definition of the receptive field (RF) to a fuzzy detector. Our RF modification is a consequence of the computational properties of the bottom-up and top-down pathways comparing stimulus with predictions. The “driver-type” logic (DTL) of bottom-up computations looks for large number of possible object parts (hypotheses), as object’s elements are similar to RF properties. The optimal combination is chosen, in unsupervised, parallel, multi-hierarchical pathways by the “modulator-type” logic (MTL) of top-down computations. The DTL is related to anatomic divergence of ascending pathways and represents a large assembly of possible combinations of elements. Anatomical convergences of descending pathways determine selective property of the MTL. Such interaction between DTL (hypotheses) and MTL (predictions) gives the visual system universality of processing (with a high resolution of lower areas) vast number of possible visual cues and flexibility to choose right one (in agreement with an individual experience).