Representation of edges of variable blur by neuronal responses in the lateral geniculate body and the visual cortex of cats: Limits of linear prediction

Abstract

We have measured the responses of cells in the cats lateral geniculate body and the visual cortex to edges which were blurred to various degrees (cosinusoidal blur). For the same cells also the responses were determined to sinusoidal gratings of various fundamental frequency and to slits of various blur and width. All stimuli were moved across the receptive fields at various speeds. The responses of most cells increased with increasing edge sharpness, but usually reached a maximum at a blur corresponding to a high frequency cutoff at 0.6–1.2 c/deg. The responses to the sharpest edges were usually smaller than those to a blurred edge (up to −50% in individual cells and −15% in the average). After normalization, the responses predicted from the spatial frequency tuning curves and the Fourier transform of the edge stimuli corresponded well to the measured blur functions up to the maximum of the edge response which varied considerably between cells, however. At edge sharpness beyond that maximum, the predicted curves rose up to edge sharpness with high frequency cutoff 1.6–1.8 times above that which produced the experimental neuronal response maximum. On the other hand, responses could increase with edge sharpening in spatial frequency regions, in which no or only small responses were seen with sinusoidal gratings (e.g. at lower spatial frequencies in “band pass neurons”). Geniculate X- and cortical simple cells as well as those geniculate Y-cells which showed phase locked grating responses behaved similarly in all respects. We concluded that edge sharpness is not represented by response amplitude of individual neurons but by the spatial distribution of excitatory peaks across the representation of the retinotopic cortical map. Our findings further indicate that spatial models of receptive fields assuming linear signal summation have only a limited value for predicting edge sharpness.