AUTOREGRESSIVE SHAPE CLASSIFIERS IN TIME-VARYING NOISE

Good results have been reported in a number of recent studies where auto-regressive shape classifiers have been used to allocate objects to a pre-determined number of classes. It is shown that this approach may give unsatisfactory results when applied to imaging systems whose noise characteristics vary substantially over time. Such variation may be due to changes in atmospheric conditions or the relative effect of discretization errors. Observations are based on the reconstruction of a number of noisy synthetic templates.