Abstract

ISAR (interactive synthetic aperture radar) images are formed by Fourier transformation of complex frequency-space data. These generally suffer from noise and can also be degraded by under-sampling due to deliberate frequency skipping. Such a problem of incomplete information is typically well handled by the concept of maximum entropy. However, to apply the approach one has to force maximum entropy in a complex function. This was accomplished as follows (ATB). Each part (real or imaginary) is regarded as the difference between two associated, positive-only functions. The latter are forced to not overlap, through the action of a tuning parameter. Then the overall entropy is the sum of the ordinary entropies of the associated functions. This approach was adapted (BRF) into a MAP estimation algorithm where the entropy steps are absorbed into a prior probability law and the likelihood law allows for Gaussian noise. Because the exp(·) imaging kernel is separable in x and y,the 2-D processing problem can be implemented as a sequence of 1-D problems, first row-wise and then column-wise. An ordinary Newton-Raphson procedure solves for the unknown Lagrange multipliers defining each 1-D problem. ISAR image examples will be shown.

© 1992 Optical Society of America