Maximum likelihood extension for non-circulant deconvolution

摘要: Directly applying circular de-convolution to real-world blurred images usually results in boundary artifacts. Classic extension techniques fail provide likely results, terms of a boundary-condition observation model. Boundary reflection gives raise non-smooth features, especially when oblique oriented features encounter the image boundaries. Tapering boundaries support, or similar strategies (like constrained diffusion), provides smoothness on toroidal support; however this does not guarantee consistency with spectral properties blur (in particular, its zeros). Here we propose simple, yet effective, model-derived method for extending images, so that they become Gaussian We achieve artifact-free even under highly unfavorable conditions, other methods fail.