Passive millimeter-wave image resolution improvement by linear and non-linear algorithms

Criteria are specified for deciding on whether or not elements of a digital image are resolved, and metrics are defined for the extent of the improvement, if any, in that resolution brought about by image processing. These criteria and definitions enabled a quantitative comparison of the resolution improvement in simulated images of a two-squares pattern that were processed by the linear Two-Mu algorithm and the non-linear Maximum Likelihood (ML) and Maximum a-Priori (MAP) algorithms Both non-linear algorithms had identical performances, and yielded a maximum resolution improvement of x2.4 for up to 1000 iterations, compared to an improvement of x1.9 for the Two-Mu linear algorithm. The performance of these algorithms was also compared qualitatively for the more complex, measured, passive millimeter-wave (PMMW) images of a simple metal/radar absorbing material (RAM) pattern and an extremely complex PMMW image of a military vehicle. Just as for the two-squares images, NIL and MAP performed identically to each other for the complex images. In the metal/RAM images processed with ML and MAP the three RAM patches were imaged much more clearly than in those of Two-Mu at a range of 384 in. However, none of the algorithms could produce clear images of all three RAMs at a range of 512 in, nor did any of them yield satisfactory images of the military vehicle. The processing time for Two-Mu can be extremely short if the appropriate-size inverse filter matrix for the chosen processing parameters and the point spread function are available from a previous computation. The processing time for ML and MAP, however, is governed only by the number of iterations required, which is 1000 for the two-squares pattern image, x2.4 improvement.