Image modeling and restoration: A new approach (Q1123657)

Many image restoration methods proposed in the literature are based on the assumption that the image spatial correlation can be approximated by an autocorrelation function which is to be obtained from the original image. If a noise-free image is not available, the autocorrelation function must be estimated. The main drawbacks of these methods are the increased complexity and the high computational cost. The authors propose here a restoration procedure which does not require a priori knowledge of the autocorrelation function or the use of identification-estimation algorithms. The basic hypothesis is that the image can be modeled as an ensemble of two-dimensional differentiable surfaces. This idea is embedded in a bidimensional Kalman filtering approach.