IMPROVED RESOLUTION FROM SUBPIXEL SHIFTED PICTURES

In this paper we consider the problem of obtaining a picture with improved resolution from an ensemble of K spatially shifted low-resolution pictures. We show that it can be done and provide a procedure for its implementations. The framework and scheme utilized in the generalized multichannel sampling theorem of Papoulis and Brown was found to be an appropriate representation for our problem. The linear channels of this scheme correspond to the processes that produce the low-resolution pictures (by blur and shifts) and to the process of reconstruction. The total process is shown to consist of two distinct degradation-restoration pairs: blurring-deblurring and undersampling-merging. This paper discusses merging. Deblurring (deconvolution) can be done by any desired scheme independent of merging. It is shown that the procedure achieves the claims of super resolution schemes but, unlike many of them, is well posed and robust. This procedure holds for a selected ensemble so that K is not too large and no two of the pictures practially coincide. The number of multiplications per pixel of the improved picture can be reduced to O(K2). © 1992.