Registration of blurred images

The images taken at nonideal imaging conditions are often blurred and noisy. The blur is typically introduced by such factors as camera defocus, imaging through turbulent media, motion of the scene or the sensor, etc., but can be also caused by preprocessing of the images. Blurred image can be modeled as a convolution of the the input (ideal) image and the point-spread function (PSF) describing the degradation effect of the imaging system. To register blurred images, it is necessary to have descriptors invariant to convolution. In this talk we introduce invariants to convolution with an arbitrary centrosymmetric PSF both in the space and Fourier domains. We demonstrate their performance when registering defocused satellite images and defocused images from a moving camera. Finally, we show some applications in superresolution imaging.