TY  - THES
T1  - Primal-dual methods in total variation regularized L1 and L2 functionals arising in image denoising
A1  - Neri, Marrick C.
LA  - English
UL  - https://tuklas.up.edu.ph/Record/UP-99796217608549277
AB  - Forthedenoising of block images using the nonsmooth L2 TV-model, a new method based on the algorithm introduced by Chambolle [8] and on the primal-dual active set strategy due to Hinterm¨uller and Stadler [28] is introduced. The active set technique aims at decomposing the image into edges (active sets) and flat zones (inactive sets, which we call islands). The data reconstruction on the islands is based on an averaging technique reflecting the statistics of the noise. Utilizing the primal and a corresponding dual variable in the island detection stabilizes the algorithm. The dual update is based onthe strategiesin[8] and[28]. Numericaltests showthatthe methodishighly efficient in removing the noise, in restoring edges, and in the reconstruction of flat image features. This paper also includes a method of VU decomposition constructed to solve the L2 TV-model. This method partitions the solution into a smooth part ? the U-subspace ? and nondifferentiable part ? the V-subspace. We also construct a primal-dual active set method for the nonsmooth regularized L1 TV-model. The proposed method is effective in removing impulse noise, random noise, and outliers. Finally, we perform introductory explorations of modified versions of the L1 and L2 TV-models with the Laplacian in the regularization term.
NO  - Computer print out.
CN  - LG996 2008 M38 N47
KW  - Digital images.
ER  -