Image restoration: A wavelet frame based model for piecewise smooth functions and beyond

Jian Feng Cai, Bin Dong, Zuowei Shen

Research output: Contribution to journalArticle

32 Citations (Scopus)

Abstract

In this paper, we propose a new wavelet frame based image restoration model that explicitly treats images as piecewise smooth functions. It estimates both the image to be restored and its singularity set. It can well protect singularities, which are important image features, and provide enough regularization in smooth regions at the same time. This model penalizes the ℓ2-norm of the wavelet frame coefficients away from the singularity set, while penalizes the ℓ1-norm of the coefficients on the singularity set. This model explicitly models images as piecewise smooth functions with a general smoothness regularization and characterizes rather general singularity set, which includes both jump discontinuities and jumps after certain orders of differentiations. As we know, all types of singularities are important image features and need to be recovered. Furthermore, the singularity set can be robustly estimated by wavelet frame transform during the image recovery procedure, which makes our model easy to solve numerically; hence, the model is insensitive to the estimation of the singularity set.The proposed model is in discrete setting and is a wavelet frame based approach. To further understand the piecewise smooth nature of the obtained solutions, we connect it to a variational model on the space of piecewise smooth functions and prove rigorously that the discrete model converges to the variational model as image resolution goes to infinity. Also, we show that the approximate solutions of the discrete model can be regarded as an approximation of those of the variational model. Through these theoretical analysis, we manage to connect the proposed discrete wavelet frame based model with the variational model. Such connection not only enables us to acquire deeper understandings of the discrete model, but also leads us to the discovery of a variational model new to the literature, which is more general and works better than the Mumford-Shah model [1] for image restoration problems. Although the focus of the paper is to propose the new model and provide theoretical studies, we still conduct numerical simulations to support our claims and theoretical findings. Our numerical studies show that the proposed model is the right one for image restorations, when the underlying solutions are piecewise smooth. Generally speaking, this model combines the merits of the PDE based approach [1-7] and the wavelet frame based approach [8-10].

Original languageEnglish (US)
JournalApplied and Computational Harmonic Analysis
DOIs
StateAccepted/In press - Nov 5 2014
Externally publishedYes

Fingerprint

Piecewise Smooth Functions
Wavelet Frames
Image Restoration
Image reconstruction
Variational Model
Singularity
Discrete Model
Model
Regularization
Jump
Image Recovery
Norm
Image Model
Coefficient
Numerical Study
Smoothness
Discontinuity
Theoretical Analysis
Approximate Solution

Keywords

  • (Tight) wavelet frames
  • Framelets
  • Image restoration
  • Mumford-Shah model
  • Piecewise smooth functions
  • Pointwise convergence
  • Split Bregman
  • Γ-convergence

ASJC Scopus subject areas

  • Applied Mathematics

Cite this

Image restoration : A wavelet frame based model for piecewise smooth functions and beyond. / Cai, Jian Feng; Dong, Bin; Shen, Zuowei.

In: Applied and Computational Harmonic Analysis, 05.11.2014.

Research output: Contribution to journalArticle

@article{3f592dd6130041d09a01020cf3243426,
title = "Image restoration: A wavelet frame based model for piecewise smooth functions and beyond",
abstract = "In this paper, we propose a new wavelet frame based image restoration model that explicitly treats images as piecewise smooth functions. It estimates both the image to be restored and its singularity set. It can well protect singularities, which are important image features, and provide enough regularization in smooth regions at the same time. This model penalizes the ℓ2-norm of the wavelet frame coefficients away from the singularity set, while penalizes the ℓ1-norm of the coefficients on the singularity set. This model explicitly models images as piecewise smooth functions with a general smoothness regularization and characterizes rather general singularity set, which includes both jump discontinuities and jumps after certain orders of differentiations. As we know, all types of singularities are important image features and need to be recovered. Furthermore, the singularity set can be robustly estimated by wavelet frame transform during the image recovery procedure, which makes our model easy to solve numerically; hence, the model is insensitive to the estimation of the singularity set.The proposed model is in discrete setting and is a wavelet frame based approach. To further understand the piecewise smooth nature of the obtained solutions, we connect it to a variational model on the space of piecewise smooth functions and prove rigorously that the discrete model converges to the variational model as image resolution goes to infinity. Also, we show that the approximate solutions of the discrete model can be regarded as an approximation of those of the variational model. Through these theoretical analysis, we manage to connect the proposed discrete wavelet frame based model with the variational model. Such connection not only enables us to acquire deeper understandings of the discrete model, but also leads us to the discovery of a variational model new to the literature, which is more general and works better than the Mumford-Shah model [1] for image restoration problems. Although the focus of the paper is to propose the new model and provide theoretical studies, we still conduct numerical simulations to support our claims and theoretical findings. Our numerical studies show that the proposed model is the right one for image restorations, when the underlying solutions are piecewise smooth. Generally speaking, this model combines the merits of the PDE based approach [1-7] and the wavelet frame based approach [8-10].",
keywords = "(Tight) wavelet frames, Framelets, Image restoration, Mumford-Shah model, Piecewise smooth functions, Pointwise convergence, Split Bregman, Γ-convergence",
author = "Cai, {Jian Feng} and Bin Dong and Zuowei Shen",
year = "2014",
month = "11",
day = "5",
doi = "10.1016/j.acha.2015.06.009",
language = "English (US)",
journal = "Applied and Computational Harmonic Analysis",
issn = "1063-5203",
publisher = "Academic Press Inc.",

}

TY - JOUR

T1 - Image restoration

T2 - A wavelet frame based model for piecewise smooth functions and beyond

AU - Cai, Jian Feng

AU - Dong, Bin

AU - Shen, Zuowei

PY - 2014/11/5

Y1 - 2014/11/5

N2 - In this paper, we propose a new wavelet frame based image restoration model that explicitly treats images as piecewise smooth functions. It estimates both the image to be restored and its singularity set. It can well protect singularities, which are important image features, and provide enough regularization in smooth regions at the same time. This model penalizes the ℓ2-norm of the wavelet frame coefficients away from the singularity set, while penalizes the ℓ1-norm of the coefficients on the singularity set. This model explicitly models images as piecewise smooth functions with a general smoothness regularization and characterizes rather general singularity set, which includes both jump discontinuities and jumps after certain orders of differentiations. As we know, all types of singularities are important image features and need to be recovered. Furthermore, the singularity set can be robustly estimated by wavelet frame transform during the image recovery procedure, which makes our model easy to solve numerically; hence, the model is insensitive to the estimation of the singularity set.The proposed model is in discrete setting and is a wavelet frame based approach. To further understand the piecewise smooth nature of the obtained solutions, we connect it to a variational model on the space of piecewise smooth functions and prove rigorously that the discrete model converges to the variational model as image resolution goes to infinity. Also, we show that the approximate solutions of the discrete model can be regarded as an approximation of those of the variational model. Through these theoretical analysis, we manage to connect the proposed discrete wavelet frame based model with the variational model. Such connection not only enables us to acquire deeper understandings of the discrete model, but also leads us to the discovery of a variational model new to the literature, which is more general and works better than the Mumford-Shah model [1] for image restoration problems. Although the focus of the paper is to propose the new model and provide theoretical studies, we still conduct numerical simulations to support our claims and theoretical findings. Our numerical studies show that the proposed model is the right one for image restorations, when the underlying solutions are piecewise smooth. Generally speaking, this model combines the merits of the PDE based approach [1-7] and the wavelet frame based approach [8-10].

AB - In this paper, we propose a new wavelet frame based image restoration model that explicitly treats images as piecewise smooth functions. It estimates both the image to be restored and its singularity set. It can well protect singularities, which are important image features, and provide enough regularization in smooth regions at the same time. This model penalizes the ℓ2-norm of the wavelet frame coefficients away from the singularity set, while penalizes the ℓ1-norm of the coefficients on the singularity set. This model explicitly models images as piecewise smooth functions with a general smoothness regularization and characterizes rather general singularity set, which includes both jump discontinuities and jumps after certain orders of differentiations. As we know, all types of singularities are important image features and need to be recovered. Furthermore, the singularity set can be robustly estimated by wavelet frame transform during the image recovery procedure, which makes our model easy to solve numerically; hence, the model is insensitive to the estimation of the singularity set.The proposed model is in discrete setting and is a wavelet frame based approach. To further understand the piecewise smooth nature of the obtained solutions, we connect it to a variational model on the space of piecewise smooth functions and prove rigorously that the discrete model converges to the variational model as image resolution goes to infinity. Also, we show that the approximate solutions of the discrete model can be regarded as an approximation of those of the variational model. Through these theoretical analysis, we manage to connect the proposed discrete wavelet frame based model with the variational model. Such connection not only enables us to acquire deeper understandings of the discrete model, but also leads us to the discovery of a variational model new to the literature, which is more general and works better than the Mumford-Shah model [1] for image restoration problems. Although the focus of the paper is to propose the new model and provide theoretical studies, we still conduct numerical simulations to support our claims and theoretical findings. Our numerical studies show that the proposed model is the right one for image restorations, when the underlying solutions are piecewise smooth. Generally speaking, this model combines the merits of the PDE based approach [1-7] and the wavelet frame based approach [8-10].

KW - (Tight) wavelet frames

KW - Framelets

KW - Image restoration

KW - Mumford-Shah model

KW - Piecewise smooth functions

KW - Pointwise convergence

KW - Split Bregman

KW - Γ-convergence

UR - http://www.scopus.com/inward/record.url?scp=84936818253&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84936818253&partnerID=8YFLogxK

U2 - 10.1016/j.acha.2015.06.009

DO - 10.1016/j.acha.2015.06.009

M3 - Article

AN - SCOPUS:84936818253

JO - Applied and Computational Harmonic Analysis

JF - Applied and Computational Harmonic Analysis

SN - 1063-5203

ER -