Blind filter identification and image superresolution using subspace methods Superrésolution aveugle d'images par la méthode des sous-espaces

Subspace methods are a powerful tool to recover unknown filters by looking at the second order statistics of various signals originating from the same source (also called a SIMO problem). An extension to the multiple source case is also possible and has been investigated in the literature. In this report we show how the blind superresolution problem can be solved by this tool. We first present the problem of superresolution as a multiple input multiple output (MIMO) one. We show that the subspace method can not be used, as is, to recover the filters affecting each image, and we present two possible solutions, based on the statistical characteristics of the images to solve this problem. Experiments are shown which validate these ideas.