Multichannel Blind Deconvolution Using Low-rank and Sparse Decomposition
Multiple images of the same scene improve the result of multichannel blind deconvolution problem. However, existing techniques require alignment of channels. We define a multichannel deconvolution scheme for highly correlated images to estimated the blur kernel and the underlying images. The recovery algorithm is formulated by alternating minimization which is proved to have the convergence guarantee. The scheme does not require the image alignment and decompose the acquired images into a low-rank and sparse component in order to exploit different types of correlations. Our scheme has applications in video sequence, biomedical images, and light fields blind deconvolution.