Abstract
The functional organisation of neuronal receptive fields can give insight into how the visual system operates. In this regard, the center-surround organisation of the receptive field plays a crucial role in extracting spatial information from visual scenes, and requires constant adaptation as the visual scene is traversed. This adaptation is achieved through surround modulation which causes the classical receptive field to be influenced by stimulation of its surround. Surround modulation is considered an elementary computation of the visual system and its function is hypothesized to assist in object boundary segmentation, optimisation of neural coding as well as normalisation of visual scene contrast. A Difference of Gaussians with varying center-surround size- and/or gain ratio is often chosen to model receptive field surround modulation. By introducing the principle of DC-balanced filtering, we show that for any given size ratio there is a gain ratio that optimises the filters capabilities in terms of feature segmentation, neural coding and contrast normalisation. Examining these ratios by mapping population receptive fields in healthy human subjects using functional magnetic resonance imaging (fMRI), we find evidence that the human primary visual cortex adheres to the principle of DC-balanced filtering across all examined parts of the visual field and across all receptive field sizes. Since surround modulation is pervasive in all sensory modalities, we expect this result to be a corner stone in understanding how biological systems achieve their high information processing fidelity. Furthermore, research areas that finds inspiration in biological vision research, such as digital image processing with deep neural networks, could benefit from the insights gained here.