A computational model of recurrent, colinear long-range interaction in V1 for contour enhancement and junction detection

Publisher Website
Google Scholar
Abstract
Physiological and psychophysical studies have demonstrated the importance of colinearity in visual processing. Motivated by these empirical findings we present a novel computational model of recurrent long-range processing in V1. Unlike other models which employ cocircular connection patterns, we restrict the long-range interaction to cells with colinear aligned RFs in accordance with empirical findings (e.g., Bosking et al., 1997). Besides colinear excitatory long-range interaction, the model uses isotropic inhibitory short-range interaction and modulating feedback. Self-normalizing shunting equations guarantee the saturation of activities after a few recurrent cycles. The primary computational goal of the model is to evaluate local, often noisy orientation measurements within a more global context and to selectively enhance coherent activity by excitatory, modulating feedback. In a first study, the model quantitatively reproduces response facilitation and suppression to a single bar element depending on the local surround outside the classical RF (Kapadia et al., 1995). With same parameters, we evaluate the competencies of the model for the processing of artificial and natural images. We show that the model robustly increases the contour saliency (Li, 1999). Further, circular variance within a model hypercolumn is decreased along contours, but preserved at points of intrinsically 2D signal variations such as corners and junctions. Junctions can thus be robustly extracted based on a distributed, hypercolumnar representation. We show for a number of generic junction configurations (T, L, X, Y, W, Psi) and various natural images that junctions can be accurately and robustly detected. Moreover, localization is better and positive correctness higher compared to a detection scheme based on a purely feedforward representation. To conclude, the model realizes basic tasks of early and midlevel vision within a single, biologically plausible architecture.

This publication has 0 references indexed in Scilit: