Irregular adaptative pyramid of agents for segmentation to interpretation of image

The paper presents the main concepts of a machine vision architecture based on a hybrid and a fine granularity multiagent system, that encourages incremental design via modular and hierarchical structuring of knowledge and pattern recognition mechanisms. The objective is not optimality of the image segmentation/interpretation but rather reliability versus unforeseen observation. We then present a first implementation of the architecture that tends to validate the approach and also that shows up a physical distribution of computation.