The Art of Vision:
Neuromathematics for Vision and Cognition

Interdipartimental Research Group University of Bologna

Presentation of the Group

 



Home


General presentation

People

Conferences

Publications

Links




 

The research of the group main focus on different aspects of cognitive science:

Problems of neuroscience, comunication sciences, and cognitive psicology.

Problems of visual perception from the point of view of the brain functionality, and of the psicology of the Gesthalt

Problems of Lie analysis for the description of objects and their functionality.



- Instruments of neurosciences: - the instruments of neuroimaging and integrate neuroscience allow to mathematically model neural mechanisms of Human cognition and to build up computational algorithms of image processing inspired to brain functionality. An essential aspect of this problem concerns the relationship between the analysis of the Sensorial Signal and the Geometric Structures of perceptive representations. In particular this is the problem of of optimal compact representation of signals by means of time-frequency or space-frequency packets. It goes from classical Fourier Analysis methods to more innovative multi-scale methods based on Wavelets and particularly to Gabor Transforms.


- algorithms of visual perception based on the psicology of the Gesthalt: It has been proved that there is a clear distinction between the physical images we look at and the images we perceive. Gaetano Kanizsa in [grammatica del vedere ] provided a taxonomy of perceptual completion phenomena and outlined that they are interesting test to understand how the visual system interpolates existing information and builds the perceived units. A tipical example is the famous triangle, where we clearly perceive lines which are not really present in the image. These boundaries are called subjective boudaries and they have been classically modeled (Mumford Nithzberg and Shiota) as minima of the elastica functional, which depeds on curvature. These models have been widely extended with methods of calculus of variations or evolution differential equations, in order to simultaneously segment the given image and reconstruct the spatial ordering (depending on depth) of the objects contained in the image.


- algorithms of visual perception based on the functionality of the brain: From the neurophysiological point of view, there is a considerable evidence that perceptual phenomena are accomplished by the first layer of the visual cortex by actively integrating local information. The principal level of organization of the visual cortex V1 is the columnar structure. The integration mechanism propagates columnar information in an orientation specific modality by means of long-range horizontal connections. In this setting the formation of contours is explained as the meeting of two neural activation flows shooted by local boundary inducers and integrating missing information. The specificity of this information propagation is described by the "association fields" that indicate boundary collinear directions as privileged diffusion directions to the detriment of orthogonal ones. The columnar organization has been modeled as a structure of contact on the Lie group of rototraslations. The notion of derivative is replaced in this setting by the Lie derivative, which is performed along suitable horizontal vector fields, similar to the association field recently proposed in vision neuroscience by Field, Hayes and Hess to explain modal boundary completion.


- Lie group analysis for the description of objects and their functionality: -The functionality of an object is usually described through its derivatives of first and hygher order. Alternately it is possible to lift original object to a hyger dimensional Lie group, whose elements are the original points together with their derivatives. The differential relations between the derivatives induce in the lifted Lie group differential costraints: not all the displacements are allowed, but only in the direction of integral curves of suitable vector fields. These constraints completely describe the geometrical and functional structure of the space: the distance, differential calculus and functional spaces. Hence properties of distances, curves and surfaces (in particular of geodesics and minimal surfaces) are object of studies. This can be done with real analysis methods, calculus of variation instruments and non linear partial differential equations of curvature type in Lie groups. In particular the structure of the visual cortex can be modeled with these instruments as the Lie group of rototraslations. The subjective contours are modeled as geodesics, and the objects as surfaces of the group.