This document summarizes a research paper that proposes a new methodology for face recognition in the presence of non-uniform (space-varying) motion blur. The key points are:
1) Existing face recognition methods cannot handle non-uniform blur that occurs in real-world camera shake situations.
2) The paper models blurred faces as a weighted combination of geometrically transformed versions of the original sharp image, using a "transformation spread function" to model non-uniform blur.
3) Algorithms are proposed to recognize faces under non-uniform blur alone, and also under combinations of blur, illumination changes, and pose variations. This is achieved by alternately estimating camera motion and illumination parameters.