Seventh Framework Programme

We propose to develop a formal model of information representation and processing in the part of the neocortex that is mostly concerned with visual information. This model will open new horizons in a well-principled way in the fields of artificial and biological vision as well as in computational neuroscience. Specifically the goal is to develop a universally accepted formal framework for describing complex, distributed and hierarchical processes capable of processing seamlessly a continuous flow of images. This framework features notably computational units operating at several spatiotemporal scales on stochastic data arising from natural images. Meanfield theory and stochastic calculus are used to harness the fundamental stochastic nature of the data, functional analysis and bifurcation theory to map the complexity of the behaviours of these assemblies of units. In the absence of such foundations the development of an understanding of visual information processing in man and machines could be greatly hindered. Although the proposal addresses fundamental problems its goal is to serve as the basis for ground-breaking future computational development for managing visual data and as a theoretical framework for a scientific understanding of biological vision. One often recognizes vision as the main sensory procedure by which we perceive our environment. Despite its apparent simplicity when considered from the naive introspection viewpoint its understanding remains a challenge for scientific investigation. With the advent of more powerful computers in the 70s and the 80s the field of digital image processing and analysis was born in the US while there had been for several centuries a tradition in Europe in particular of visual psychophysics [1]. But it was David Marr [3] at MIT who set up the stage for a joint study of artificial and biological vision by making the informed statement that vision was an information processing task which was relatively independent of the organism, natural or artificial, that was performing it. This raised a lot of enthusiam worldwide and started a long line of research which is continuing today. Nonetheless, after the death of David Marr the two communities of psychophysicists and neurophysiologists of vision on one hand, of artificial vision scientists on the other hand, which for a while had been partially united split again. The former went back to explore the maze of visual phenomena while the latter went back to developing “fast and robust” algorithms that work, in the spirit of good engineering. One of the main reasons for this division was the lack of a common framework for thinking about visual perception. Despite of this, the intellectual interactions between the two communities have continued over the years on such problems as the structure of the processing, bottom up, top down, or both, the statistical nature of natural images, the acquisition of knowledge and its later use as a prior. Artificial vision researchers have defined a number of organizational concepts such as the rich geometric structure underlying image formation or fundamental variational principles that are the basis of many state of the art computer vision algorithms. There remains the fact that many of these algorithms have to be hand tuned for a particular application and often fail in unconstrained