6. More advanced questions

6.1 Myopic inversion

In the problems we have studied so far, we have assumed that the direct model is perfectly known. In practice, it often happens that this model depends on parameters that are poorly known or not known at all. In such cases, the problem is said to be "myopic" or "blind". A typical example is deconvolution with an unknown kernel. Blind deconvolution involves an informational obstacle that is even more severe than that of known-kernel deconvolution. Indeed, in the absence of any a priori information about the objectx and the impulse response h, it is obviously impossible to discriminate the pair (x,h ) from any other pair (h ₁ ★ x, h ₂ ) such that h _1...