Abstract :

In the Wertheimer programme there are two kinds of organizing laws. The first kind are grouping laws which, starting from atomic and local data, recursively construct larger groups in the perceived image. Each grouping law focuses on a single quality (color, shape, direction,...). The second kind are principles governing the collaboration and conflicts of gestalt laws. We explain why and how the Wertheimer programme can be translated into a Computer Vision programme. This translation is not straightforward, since Gestalt theory did not address two fundamental matters : image sampling and image information measurements. Using them, we show that a quantitative version of Kanizsa's ``masking in texture'' can lead to numerical simulations of several grouping laws. As illustration, we display numerical experiments on several digital images. Analyzing successes and failures of these experiments, we conclude that Computer Vision is currently facing exactly the same challenges as were discussed in Gestalt theory, namely the collaboration and conflicts between grouping laws, and the masking issue.