A neural network model is proposed to explain how different patterns of blood-oxygen-level-dependent (BOLD) signals arise during the perceptual organization of visual stimuli. The model implements object-based selection by global inhibition applied to all background spatial locations and local excitation that enables activity to spread along the locations of the figure. In the model, the state of perceptual grouping is associated with an elevated firing rate for all nodes encoding the figure and simultaneous reduction in the strength of dendritic processing. It is assumed that computation in the dendritic trees provides a major contribution to the BOLD signal. Computer simulations show that increased perceptual complexity of the stimulus leads to a reduction of the BOLD signal in V1. On the other hand, the lateral occipital complex (LOC) exhibits an opposite pattern with the strongest BOLD signal for images with a 3-D interpretation. This pattern is observed when there are no distractors present in the visual field. In the task of collinear contour grouping (with many distracting Gabor patches) both the V1 and the LOC showed an elevated BOLD signal when a perceptual group (contour) is present compared to the control condition. Also, the model explicates how global or local orientation of attention contributes to the perceptual grouping through differential neural activations in the parietal cortex. The proposed model is consistent with the concepts of base and incremental grouping derived from electrophysiological studies. |