Abstract
We describe the results of a new algorithm for processing a series of noise degraded and randomly shifting digital images using the bispectrum. The algorithm is shown to yield a reconstructed signal-to-noise ratio (SNR) improvement which is nearly equal to perfect frame stacking using no prior knowledge whatsoever. The technique is quite robust and has wide ranging applications for situations in which imaging platform stabilization and noise are problematic The algorithm is very efficient and can completely process 128 250 × 128 format image frames in less than ten minutes on a VAX workstation. This is significantly faster than other algorithms we are currently aware of. Comparisons of the bispectrum algorithm with perfect frame registration and an adaptive matched filter algorithm are given for a wide range of simulated data sets. The data sets include both shot noise limited simulations and background noise limited simulations. The results indicate that the technique is more robust than second order matched filter methods due to the inherent shift invariance of the bispectrum. A description of the algorithm is given. It is quite different from other image reconstruction algorithms we are aware of and holds out the possibility of real time digital image processing using only two or three transputer array boards.
© 1990 Optical Society of America
PDF ArticleMore Like This
V.A. Klückers, A.J. Longmore, and J.C. Dainty
RWB1 Signal Recovery and Synthesis (SRS) 1995
Andy Stein, Arnab Sinha, and George Barbastathis
ITuB4 Integrated Computational Imaging Systems (ICIS) 2001
Robert D. Brandt, Feng Lin, and Yang Zhao
MI5 OSA Annual Meeting (FIO) 1990