scientific article; zbMATH DE number 1335865

From MaRDI portal
Publication:4258554





This chapter is concerned with reconstructing images from noisy, spatially correlated data. We focus on the context of reflectance data obtained from a satellite by remote sensing. The satellite is assumed to collect data with a sensor that receives signals from the same pixel in \(p\) frequencies of the electromagnetic spectrum. So signals are \(p\)-vectors, but they are corrupted by a noisy sensor, clouds and moisture that produce noise, and other sources of noise. This chapter generalizes results of \textit{R. Klein} and \textit{S. J. Press} [Commun. Stat., Theory Methods 18, No. 9, 3177-3202 (1989; Zbl 0696.62152)] that adopted vague priors. Images on the ground are reconstructed by classifying all the pixels in a scene of interest into one of \(K\) unordered classes. For example, \(K\) classes of possible agricultural land use, such as rice, soybeans, or water. The pixels are classified contextually, using the data from pixels in the vicinity of a pixel to be classified. For example, if all the pixels in the vicinity of the main pixel of interest have been classified as ``rice, we are likely to classify the main pixel as rice. The statistical approach to inference will be Bayesian. Well find the predictive probability for classifying a pixel into a given class, for all classes, and we'll classify the pixel into that class that has the greatest predictive probability. For the entire collection see [Zbl 0913.00046].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4258554)