Neural model of nonlinear subfield integration in cortical simple cells

Most approaches that model biological early vision systems perform, at the cortical level of simple cells, a linear integration of the activity from visual ON and OFF pathways which are separated at the retinal level. Based on empirical as well as theoretical investigations we propose a nonlinear neural network model that is selectively responsive to contrast magnitude as well as to the sharpness of luminance transition. The nonlinear circuit allows for accurate and reliable detection of contrast changes even in noisy images. Simulations with artificial and camera images show a higher positional selectivity for local contrasts than an equivalent linear device. Furthermore, in a multiscale hierarchy the nonlinear circuit produces a unique maximum response in scale-space where scale directly relates to the width of the luminance transition