Multi-frame real image restoration based on double loops with alternative maximum likelihood estimation

A multi-frame blind restoration algorithm based on alternative maximum likelihood estimation using double loops is proposed to restore object images. An anisotropic regularization term is incorporated into the integral likelihood function to avoid over-smoothing of details of the restored images and improve the stability of image restoration. In order to make full use of the data relationship in the observed multi-frame images which contain the same objects with different point spread functions (PSF) in a short sequence, an iterative strategy based on double loops is employed in estimating the unknown different PSFs of the observed images. Experiments on synthetic and real turbulence-degraded images illustrate that the proposed algorithm is effective in restoring real turbulence-degraded images by using very few frames.