Statistical autofocus of synthetic aperture sonar images using image contrast optimisation

We present results of a technique for autofocus of stripmap synthetic aperture sonar (SAS) imagery. The technique of contrast optimisation is posed in a general Bayesian framework. The algorithm uses standard optimisation techniques on towfish motion parameters to minimise a cost function. The cost function consists of an image contrast measure, approximating the image likelihood, and a path likelihood measure, obtained from the towfish motion statistics. A number of simple image contrast measures are compared, and the optimal method for parameterising the towfish path is presented. Results showing the cost-function surface for simulated and real data are given. These show the technique of contrast optimisation is viable, and demonstrate the need for some form of regularisation in the cost function, justifying the Bayesian approach. Results on simulated data, with known towfish motion, show a high degree of accuracy is possible when minimising this cost function, but can depend on the image quality measure used. Local gradient optimisation algorithms are shown to be inadequate for real data, due to the high level of noise on the contrast surface. Preliminary results from a multiple-frequency interferometric SAS system, built by the Acoustics Research Group at the University of Canterbury are presented and ways to combine this extra information for autofocus purposes are discussed. When combined with a bulk-error removal method, or INS measurements, this technique shows promise for high-quality autofocus of regions of interest