Blind Inversion of Integral Equations
People
 P. Favaro, S. Soatto.
References
 A Geometric Approach to Blind Deconvolution with Application to Shapefrom Defocus (to appear, IEEE CVPR 2000)
Synopsis
 Once addressed the Observability of the accommodation cue - which establishes to what extent shape can be reconstructed from different images of the same scene taken with different imaging geometries - it is necessary to devise algorithms to carry out such estimation. Our emphasis is in provably convergent algorithms that return, together with an estimate of shape, a measure of reliability. We are currently exploring different performance criteria that range from the 2-norm to Total variation, and Information-divergence. For 2-norms, we use the geometry of (infinite-dimensional) Hilbert spaces to arrive at a provably optimal and provably convergent algorithm.