A multi-resolution filling-in model for brightness perception

We present a multiscale neural filling-in model for brightness reconstruction of initial DoG filtered images. In contrast to the classical single-scale filling-in models it no longer requires an additional (luminance) signal to restore arbitrary images. Moreover, it substantially reduces the computational cost of the reconstruction process. We present a multilayered hierarchical neural network comparable to a Laplacian pyramid in which contrast measures are filled-in in dedicated frequency domains. We show in simulations how this model operates on synthetic as well as on real-world images.