Abstract
A two-phase image restoration method based upon total variation regularization combined with an L1-data-fitting term for impulse noise removal and deblurring is proposed. In the first phase, suitable noise detectors are used for identifying image pixels contaminated by noise. Then, in the second phase, based upon the information on the location of noise-free pixels, images are deblurred and denoised simultaneously. For efficiency reasons, in the second phase a superlinearly convergent algorithm based upon Fenchel-duality and inexact semismooth Newton techniques is utilized for solving the associated variational problem. Numerical results prove the new method to be a significantly advance over several state-of-the-art techniques with respect to restoration capability and computational efficiency.
| Original language | English |
|---|---|
| Article number | 5428846 |
| Pages (from-to) | 1731-1739 |
| Number of pages | 9 |
| Journal | IEEE Transactions on Image Processing |
| Volume | 19 |
| Issue number | 7 |
| Early online date | 11 Mar 2010 |
| DOIs | |
| Publication status | Published - Jul 2010 |
| Externally published | Yes |
Funding
The work of R. H. Chan was supported by HKRGC under Grant CUHK 400508. The work of Y. Dong and M. Hintermüller was supported in part by the Austrian Science Fund FWF under SFB F32 “Mathematical Optimization and Applications in Biomedical Science” and in part by the Austrian Ministry of Science and Research under START-grant Y305 “Interfaces and Free Boundaries.” The associate editor coordinating the review of this manuscript and approving it for publication was Prof. Eric Kolaczyk.
Keywords
- Fenchel duality
- Image deblurring
- Impulse noise
- L data fitting
- Noise detector
- Semismooth Newton method
- Total variation regularization