Aggregate Regularization Technique for Imaging System Fusion With Multistate Maximum Entropy Neural Network

摘要

A new approach to the problem of improvement of the quality of images obtained with several imaging systems as required in remote sensing imagery is addressed. We propose to exploit the idea of combining the aggregate regularization and neural network (NN) based implementation of the system fusion method. The problem of image restoration with system fusion is stated and treated as a specific ill-conditioned inverse problem. The degraded data are provided by several systems, which employed the same or different image formation methods. The maximum entropy requirement as prior knowledge for restoration is posed and the regularization methodology is applied to perform the system fusion. Computationally, the restoration and fusion are accomplished by minimizing the energy function of the proposed modified multistate Hopfield-type neural network, which integrates the model parameters of all systems incorporating a priori information, aggregate measurements and calibration data. The developed maximum entropy neural network (MENN) is able to solve the system fusion tasks without complication of its structure independent on the number of systems to be fused. Simulation examples are presented to illustrate the good overall performance of the fused restoration achieved with the proposed MENN algorithm.