Dominating Opponent Inhibition of On and Off Pathways for Robust Contrast Detection

Recently, we have developed a nonlinear circuit for oriented contrast detection, which is motivated by main architectural stages in mammalian visual system. In this work we examine how on-and oo-contrast are combined before feeding into the nonlinear circuit. We propose a mechanism of dominating opponent inhibition (DOI), where each pathway receives stronger weighted inhibitory input from the opponent domain. We employ an analysis of a simpliied circuit accompanied by simulations with systematic parameter variations. Results show that DOI makes the circuit robust to noise, largely independent of the amount of noise added. Finally, we demonstrate the capabilities of the model by processing synthetic as well as natural images. Results are compared to a linear circuit, equivalent to a rst order Gaussian derivative and to the nonlinear model without DOI, showing that the new circuit largely suppresses spurious noise, while remaining sensitive to contrast variation at edges.