Image identification and restoration: A survey

Blurred and noisy images can often be represented as nonstationary two-dimensional stochastic processes that can be modelled by a set of linear space-varying state equations or by an ARMA input-output equation. Effects of any distortion resulting from blur and noise might then be removed by Kalman-type filtering and/or by ARMA smoothing respectively, provided that the appropriate image parameter are available. However, because such parameters are a priori unknown for any arbitrary image and furthermore vary spatially as the image is scanned, adaptive identification-filtering procedures appear to be very useful for satisfactory restoration.