Dynamic image modelling by neural networks
Satyendra Bhama, Harpreet Singh, Devinder Kaur · International Journal of Systems Science · 1994
Sometimes an image can be characterized with the help of a dynamic system in terms of a first-order differential equation, e.g. when a scanner output contains additive noise and/or is degraded due to interaction between the sensing elements (the cameras) or by other phenomena. In this paper, an algorithm for image estimation from the noisy output of a scanner is hypothesized. Determination of the parameters A, B, and the C matrices of a dynamic system leads to the design of an estimator whose input is the output of a scanner with uniform speed. A portion of the overall algorithm using neural networks trained by a gradient descent learning algorithm is discussed in detail. In particular, the determination of A. B, C from states, time derivatives of states and inputs are highlighted. The details of implementation are included. The salient points regarding the development of the complete algorithm are discussed. It is hoped that the results achieved so far will give rise to new techniques for the application of neural networks to image enhancement in particular and to image processing in general.