A Possible Explanation of Oriented Bar Filling-in at the Blind- Spot in the light of Hierarchical Prediction Mechanism

Abstract

Despite the absence of retinal input, the blind-spot gets filled up with surrounding visual attributes. This phenomenon is called the perceptual filling-in or completion. Various possible neural mechanisms have been proposed but most of these are not entirely consistent with the findings of brain activation and brain organization. In one study, recently, it has been shown that the filling-in of the shifting bar and the anisotropy in the filling-in of misaligned bars could be explained by incorporating a common general principle called hierarchical predictive coding (HPC). In this report, we have extended this proposition to explain the filling-in of oriented bars. We have considered a three level (LGN-V1-V2) HPC model network in which, the blind-spot was emulated by removing the corresponding feed-forward (LGN-V1) connection. We simulated the responses of predictive estimator (PE) neurons at blind-spot while stimulating the network with oriented bar stimuli. Results show that the filling-in is best for aligned bars but it faded away with increasing orientation and moreover, corners were not predicted in any orientation. Qualitatively, these results are consistent with the findings of psychophysical experiments. We discussed this phenomenon in the HPC framework and argue that in the absence feed-forward connections, the best prediction is dominated by learned statistical regularities that include an abundance of bar and similar structures. These results suggest that the filling-in process could be a manifestation of HPC.