Image Deblurring | Iterative Least Squares । Constrained LS । Algorithms | python
(Non-blind) Deblurring of an image with Iterative and Constrained Least Squares Algorithms with python
All computations for the Inverse Imaging problem are performed in the frequency (Fourier) domain (H* is complex conjugate of the frequency response H of the PSF)
For ILS there is no explicit regularization,
although number of iterations can be thought of as a regularize
For CLS there is a smoothness penalty term along with the data fidelity term in the optimization objective
The matrix C is chosen to be 2D Laplacian for CLS
Both the methods are iterative, CLS contains an additional hyperparameter α that corresponds to smoothness penalty
Iterative updates are performed for 50 iterations for each of the algorithms, for couple of Kernels (Gaussian and motion blur)
There are some artifacts in the output created by the algorithms
#imageprocessing #imageprocessingpython #python #algorithm #machinelearning #optimization