Noise in superresolution reconstruction

Abstract

Recently there has been a significant interest in reconstructing a high-resolution (HR) image based on a set of low-resolution (LR) images with relative displacement. These images are typically undersampled with respect to the image spectrum of a HR image. I show that, although ideally a resolution increase of N times is possible with N LR images, in a practical system noise is a limiting factor that increases substantially as we approach this theoretical superresolution limit. For one dimension and a special case with two LR images, I present an analytical result of the noise amplification as a function of their displacement. This is defined as a condition number of the superresolution system, with the associated definitions of a well-conditioned and ill-conditioned superresolution reconstruction system. © 2003 Optical Society of America.