In this paper, we apply blind deconvolution to infrared images. We concentrate on the important case where the blurring operation is spatially invariant; the point spread function (PSF) is more like a speckle, an out-of-focus, or their combinations with a little motion blur. The information suggestion on PSF is only Gaussian based spectra or shape. The restoration procedure involves estimating the PSF and the true image alternatively. For simplicity, we use gradient-based iterative method to solve minimization problem. The model is not complicated and we make the uncertainty of blind deconvolution into the choice of only a few parameters. Finally, deconvolution results are presented for both synthetic and real data.