Abstract

The problem of blind deconvolution, extracting both the original image and point spread function (PSF) from only the measurement of their convolution, may at first seem a futile task. However Lane and Bates [1] have shown via separation of zero sheets that blind deconvolution is theoretically possible for dimensions of two or greater when very little noise is present. In practice noise is a significant problem that limits the ability to perform successful deconvolution. Many researchers have also found that the only way of achieving blind deconvolution is to impose spatial and spectral constraints on the solution, derived from a priori information about the imaging system, and to eliminate the trivial solution of the image or PSF being a delta function. This becomes less useful in situations where very little else is known except the measured data.

© 1998 Optical Society of America