A neural model of luminance-gated recurrent motion diffusion for 2D motion integration and segmentation

We propose a model of motion integration modulated by luminance information, which is able to explain the percept on a large class of motion stimuli facing the aperture problem. This model is related to other multi-layer architectures incorporating both feedforward, feedback and inhibitive lateral connections and is inspired by the motion processing cortical areas in the primate (V1, V2, MT). Our main contribution is to propose a new anisotropic integration model where motion diffusion through recurrent connectivity between layers working at different spatial scales is gated by the luminance distribution in the image. This simple model offers a competitive alternative to models based on a large set of cortical layers implementing specific form or motion features detectors. We demonstrate that the proposed approach produces results compatible with several psychophysical experiments concerning not only the resulting global motion percept but also the motion integration dynamics. It can also explain several properties of MT neurons regarding the dynamics of selective motion integration. As a whole, the paper affords an improved motion integration model which is numerically tractable and reproduces key aspect of cortical motion integration.