Optical sensed image fusion with dynamic neural networks

The neural network-based technique for improving the quality of the image fusion is proposed as required for the remote sensing (RS) imagery. We propose to exploit information about the point spread functions of the corresponding RS imaging, systems combining it with prior realistic knowledge about the properties of the scene contained in the maximum entropy (ME) a priori image model. Applying the aggregate regularization method to solve the fusion tasks aimed to achieve the best resolution and noise suppression performances of the overall resulting image solves the problem. The proposed fusion method assumes the availability to control the design parameters, which influence the overall restoration performances. Computationally, the fusion method is implemented using the maximum entropy Hopfield-type neural network (MENN) with adjustable parameters. Simulations illustrate the improved performances of the developed MENN-based image fusion method.