Computational Layout Perception using Gestalt Laws

We present preliminary results on computational perception of interactive layouts. Our goal is to algorithmically estimate how users perceive a layout. Potential applications range from automated usability evaluation to computer-generated and adaptive interfaces. Layout perception is challenging, however, because of diverse features, combinatorial complexity, and absence of approaches. We have explored Gestalt laws as parsing heuristics. Our approach finds a parametrization that optimally resolves conflicts among competing interpretations of a layout. The output is a hierarchical grouping of main elements. The results are promising: an implementation of just four Gestalt laws enables hierarchical grouping that presents promising results in 90% of our (realistic) test cases.