Perceptual Filling-in: From Experimental Data to Neural Network Modeling

Abstract

Recognizing objects, performing goal-directed actions, and navigating the environment are capabilities that crucially depend on the ability to correctly segregate and perceive surfaces. Surface perception results from computations that involve multiple processing levels in the visual system. The nature of these computations has been a matter of persistent debate. The contribution of this chapter to this debate is threefold. In the first part (“Reconstructive Processes Contributing to Surface Perception”), an overview is given of empirical studies that inform and constrain computational models of surface reconstruction. In the second part (“A Computational Model for Modal Texture Filling-in”), empirically supported principles of surface perception and known architecture of early visual areas are used to build a computational model of neural activity corresponding to visual filling-in of surface texture in early visual areas. The model explicitly simulates subthreshold and suprathreshold activity and therefore generates predictions not only for the activity distributions of spiking neurons, but also for functional magnetic resonance imaging (fMRI). The third part, “Insights, Limitations, and Future Research Directions,” provides a discussion of the main insights following from the modeling results. The model demonstrates the consequences of subthreshold neural spread of the BOLD signal for the activity distribution obtained with stimuli typically used to investigate perceptual filling-in, and it provides insight into the divergent data from human fMRI and neurophysiological experiments in animals. The model's architecture, in which surface-related activity in low-level regions is validated by recurrent loops involving higher-order areas, is in line with theories of conscious perception. Limitations of the model will be discussed, and future research directions will be proposed.