Biologically inspired multi-scale image analysis

Medical images are nowadays produced in such qualities, and societal demands for quality are so high, that a strong demand for sophisticated image analysis techniques has emerged: 'computer aided diagnosis'. There is much to be learned from the brain. In this presentation we will study a number of examples where neurophysiological findings in the early stages of vision have inspired advanced mathematical algorithms for medical image analysis: - The models for simple cells in V1 have inspired a robust class of regularized spatio-temporal differential operators for images. - The strong feedback from V1 to LGN had lead to a wealth of adaptive ('geometry-driven') diffusion algorithms for edge preserving smoothing. - The model of the multi-scale Reichardt detector for motion triggered the design of multi-scale Lie derivatives for robust optic flow detection. they outperformed all classical methods. - The revolution due to the voltage sensitive dyes has shown short range cortical column orientation connections: this has inspired context filters, and tensor voting for Gestalt grouping processes; we developed from this a new theory for the invertible orientation wavelet transform. - Finally, the multi-scale receptive field sampling evident from the retina has lead to many projects studying the 'deep multi-scale structure' of images, to study and understand the difficult hierarchical structure of images. The lecture will address these issues in an intuitive, well-illustrated manner.