PRACTICAL LIMITS OF THE DECONVOLUTION OF IMAGES BY KRIGING

In image analysis, the available data do not usually have a point support, but are convoluted by a weighting function p(x), which is determined by the physical process of the sampling mode. Deconvolution of images is generally treated by Fourier transform. This approach is known to be inoperable when considering convoluted data with noise or when a regular grid of data is not available and therefore missing data have to be interpolated. It is for this reason that it is better to use another system based on the process of estimating point data by deconvoluting kriging. The practical limitations of this method can be expressed in terms of variance of estimation (or of signal to noise ratio), accessible for each experimental situation after modeling. They are presented by practical cases of the weighting function p(x) and variograms gamma(h), from calculations and simulations.