The project concerns automatic fusion of digital images acquired by real non-ideal sensors. In such a case, image degradations are described by a rather complicated model comprising non-rigid (affine, projective, or elastic) geometric deformations and blurring by unknown space-variant filters. By image fusion we understand the process of integrating complementary multisensor, multitemporal and multiview information into a single output frame of higher quality than the quality of the inputs. In this project, we propose to carry out the fusion in two stages: (a) geometric registration of the input images and (b) multichannel blind restoration. Since the current registration techniques are not able to handle blurred images with non-rigid geometric deformations, the first goal of the project is to develop a new method especially for this purpose. In the second stage, we propose new methods for multichannel blind deconvolution. Unlike the existing deconvolution algorithms, the new methods are designed particularly for images blurred by unknown filters and are tolerant of small between-channel misalignment. The theoretical results are applied to the enhancement of visual quality of astronomical observations, particularly to the images of fast solar events. Numerous applications outside astronomy are also demonstrated.