We can recognize rapidly and effortlessly complex visual scenes. Such amazing ability in visual recognition needs the effective processing of visual information along the multiple stages of visual pathways. Neurophysiological experiments have provided evidence for a "simple-to-complex" processing model based on a hierarchy of increasing complex image features, performed along the feedforward pathway of the ventral visual system. On the other hand, visual system has abundant feedback connections, whose number is even larger than the feedforward ones. Li et al. [1] showed that top-down signals allowed the V1 network to engage stimulus components that are relevant to a perceptional task and to discard influences from components that are irrelevant to the task [1]. They showed that V1 neurons exhibited characteristic response patterns depending on the array of stimulus components. However, it remains unclear how top-down signals reflecting task behaviors generate such response patterns. To address this issue, we present a model of visual system that consists of the primary visual cortex and the higher area V4. We consider the two types of perceptual tasks used by Li et al., bisection task and vernier one. V4 neurons receive the top-down signal relevant to the task, as well as the feedforward inputs from V1 neuron, and feed the outputs back to V1 neurons. We show here that top-down signals modulate the feedback from V4 to V1 neurons, depending on which stimulus components are relevant or irrelevant to the task being executed. We also show that dynamical property of V1 neurons is caused by the feedback signals from V4 and the modulation of lateral connections of V1 neurons. Our model provides the results on the firing properties of V1 neurons that are compatible with the experimental results by Li et al.. This study provides a useful insight to understanding the role of V1 area as an adaptive processor in visual processing.

[1] Li eta al. Nat Neurosci, 7651-7657 (2004).