A Coherent Computational Approach to Model Bottom-Up Visual Attention