Shift-invariant imaging using the bispectrum

Summary form only given. It is shown that shot noise limited image reconstructions can be obtained from an ensemble of randomly moving noisy images via bispectrum domain processing. In particular, the authors demonstrate that this method is more robust than second order techniques such as adaptive tracking with a matched filter, or centroiding. The bispectrum is inherently shift-invariant and therefore is not limited by the raw frame (or pixel) SNR. The image reconstructions are found to give comparable performance to perfect registration, and furthermore to retain the correct signal levels on a pixel and noise statistics (for uniformly distributed noise).<>