helsinki1.jpg

Aapo Hyvärinen, Patrik Hoyer, Mika Inki, Urs Köster

We have introduced modifications of the basic ICA model that describe further statistical structure of natural images and explain further aspects of the primary visual cortex. The modifications use a linear decomposition as in the basic ICA case described above, but the components are not assumed to be all independent.

Patrik Hoyer and Aapo Hyvärinen

The above results, on the link between simple cell receptive fields and ICA applied to natural images, were first demonstrated for static, monochrome (gray-scale) images. Subsequently, it has been shown that the features learned from dynamic image data (video) also strongly resemble simple-cell receptive fields (see, e.g. van Hateren and Ruderman, 1998; Olshausen 2000). The goal of this project was to extend the analysis to consider the effect of chromatic and stereo information.

Natural images can be defined as photographic images taken in our natural surroundings. For example, of the images below, those on the left side would be considered part of the class of natural images, as opposed to the various other kinds of images on the right-hand side.

The statistical structure of natural images is important for two reasons:

Consortium

The research consortium was funded for 2003-2005 as part of the research programme on Proactive Computing (PROACT) of the Academy of Finland. The consortium consists of three research groups working on two subprojects.

For the subproject Algorithmics:

Reconfiguration of a mobile device based on context is our contribution to the ITEA project Space4U. To enable reconfiguration we use the frameworks and mechanisms developed in the preceding project ROBOCOP and build only the needed extensions. Project ROBOCOP defined and implemented a middleware architecture especially suitable e.g. for mobile devices and PDAs.