Generalized Sparse Variation Regularization for Large Fluorescence Image Deconvolution

Abstract : In this work, we generalize the sparse variation (SV) combining the total-variation (TV) and the L 1 regularization and introduce a novel family of convex and non-quadratic regularizers for fast deconvolution of large 2D fluorescence images. These regularizers are defined as mixed Lp-L 2 norms (p ≥ 1) which group image intensity and spatial differentials, computed at each pixel of the image. By coupling a regularization term of this family with a quadratic data fidelity term, we propose a fast and efficient deconvolution method by using the primal-dual (proximal) algorithms to minimize the corresponding energy functional. Experiment results on both 2D simulated and real fluorescence scanner images demonstrate the performance of our method in terms of restoration quality as well as computational time.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

Cited literature [55 references]  Display  Hide  Download

https://hal.inria.fr/hal-01609810
Contributor : Charles Kervrann <>
Submitted on : Wednesday, October 4, 2017 - 9:35:29 AM
Last modification on : Tuesday, April 17, 2018 - 9:05:27 AM

File

HAL-SVDeconv.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01609810, version 1

Collections

Citation

Hoaï-Nam Nguyen, Vincent Paveau, Cyril Cauchois, Charles Kervrann. Generalized Sparse Variation Regularization for Large Fluorescence Image Deconvolution. 2017. ⟨hal-01609810⟩

Share

Metrics

Record views

206

Files downloads

355