The most intriguing aspect of Grossberg's vision model is the notion of feedback, and the role that it plays in visual processing, both for boundary completion, and also for surface brightness filling-in. I began to see a more general concept of feedback processing embodied in those two different aspects of Grossberg's model, which relates to the more general principle of reification in perception, the constructive, or generative aspect of visual processing which plays such a central role in Gestalt theory. Before I could submit my paper on Directed Diffusion again, I felt I had to write a paper just on this issue of feedback by itself, as a prerequisite for the way I wanted to present the Directed Diffusion model. The result was the two part paper below.
Computational Implications of Gestalt Theory I: A Multi-Level Reciprocal Feedback (MLRF) to Model Emergence and Reification in Visual Processing
Computational Implications of Gestalt Theory II: A Directed Diffusion to Model Collinear Illusory Contour Formation
This is a two-part paper that discusses the computational implications of Gestalt theory. I propose a perceptual modeling approach, i.e. to model the subjective experience of vision rather than the corresponding neurophysiological state. I propose specific computational interactions to account for the Gestalt properties of perception, and to explain the role of feedback in vision with a specific computational model that replicates a number of visual illusory phenomena. Part II of the paper demonstrates how the more subtle second order properties of illusory contour formation can be modeled computationally with a dynamic feedback model, as an alternative to the hard-wired receptive field embodied in the neural network approach. Submitted to Perception & Psychophysics June 1999, rejected October 1999. Resubmitted to Cognitive Psychology November 1999. Due to bureaucratic oversight, each of the two papers were sent to a different reviewer as copies of the same paper. rejected May 2000.
This concept then feeds directly into my Directional Harmonic theory, that generalizes the process of directed diffusion to account not only for collinear completion as seen in the Kanizsa figure, but also for the perceptual completion through illusory vertices and intersections involving the intersection of two, three, four, or more illusory contours that meet at a sharp vertex.
Directional Harmonic Theory: A Computational Gestalt Model to Account for Illusory Contour and Vertex Formation.
This is an updated and thoroughly revised version of my first paper with all new computer simulations using a completely different approach, and a more general discussion of Gestalt principles. Submitted to Perception August 2001. Revise & Resubmit November 2001. Resubmitted January 2002