Integrating Inhomogeneous Processing and Proto-object Formation in a Computational Model of Visual Attention