FFT and Kernel filters perform a mathematical function known as Convolution. Blurring caused by the atmosphere and optics is also a form of convolution. It is possible in theory to "undo" a convolution filter using a process called deconvolution. In practice, actually performing deconvolution is rather tricky and complex.
Maximum Entropy and Lucy-Richardson deconvolution are advanced image restoration algorithms that can remove the effects of blurring in an image. These algorithms, first pioneered for radio astronomy, were popularized for visible-light astronomy when the problems with the Hubble Space Telescope optics were discovered.
Outside of astronomical applications, the same image technique can be applied to just about any image, from microscope pictures to security camera video frames. The only essential requirement is that the image be blurry. Note that images taken with very short focal length cameras may have limited resolution, yet they may have too few pixels to properly sample the blur that is in the image. If the pixels in the camera do not resolve the blur, then no image processing algorithm can improve the resolution; deconvolution will not improve the image.
Two things are required for deconvolution to work. The first is a model of the blur, known as the Point-Spread Function (PSF). The PSF tells the deconvolution algorithm how the image was blurred; it is essentially an image of a perfect point source taken with the same camera. For astronomical images, it is often easy to determine the point-spread functions since every single star image represents the PSF. For other types of images, it may be necessary to guess at a model; MaxIm DL includes features that help you choose the best model.
The second piece of information required is some information on the noise level and the average background level in the image. MaxIm DL has to know how hard to work at deconvolving each pixel; it uses a noise model and knowledge of the background level to do this.