Predictive coding: A Possible Explanation of Filling-in at the blind spot

TL;DR

This paper proposes that hierarchical predictive coding can explain the filling-in phenomenon at the blind spot, supported by neural response simulations aligning with physiological data.

Contribution

It introduces a computational model using hierarchical predictive coding to explain the neural basis of filling-in at the blind spot.

Findings

Neural responses at the blind spot match physiological data.

Predictive coding accounts for filling-in phenomena.

Model demonstrates filling-in as a natural outcome of hierarchical prediction.

Abstract

Filling-in at the blind-spot is a perceptual phenomenon in which the visual system fills the informational void, which arises due to the absence of retinal input corresponding to the optic disc, with surrounding visual attributes. Though there are enough evidence to conclude that some kind of neural computation is involved in filling-in at the blind spot especially in the early visual cortex, the knowledge of the actual computational mechanism is far from complete. We have investigated the bar experiments and the associated filling-in phenomenon in the light of the hierarchical predictive coding framework, where the blind-spot was represented by the absence of early feed-forward connection. We recorded the responses of predictive estimator neurons at the blind-spot region in the V1 area of our three level (LGN-V1-V2) model network. These responses are in agreement with the results of earlier physiological studies and using the generative model we also showed that these response profiles indeed represent the filling-in completion. These demonstrate that predictive coding framework could account for the filling-in phenomena observed in several psychophysical and physiological experiments involving bar stimuli. These results suggest that the filling-in could naturally arise from the computational principle of hierarchical predictive coding (HPC) of natural images.