Image subsampling and multi-platform data integration: a stochastic relaxation approach

Remotely sensed data are recorded as aggregates over space of a continuous variable, a situation which can cause possible variation in the performance of statistical image analysis models for noise removal and classification. Increasingly, the need for image subsampling methods has then been considered. A solution to the problem is proposed in the paper by adopting a multivariate model-based approach in a Bayesian context. The analysis is centred on the representation of aggregate spatial processes transformed through linear operators. This framework is shown to be suitable for treating image subsampling and multi-platform data integration simultaneously