In solving sparse representation problems in image processing generally the L1-minimization techniques gives good result compared to other greedy algorithms. But sparse coding method that solves the minimization based on 'Feature-sign search algorithm' is slow. Does the methods like, iterative shrinkage thresholding algorithm (ISTA) or FISTA (fast ista) which are commonly used in image denoising problems, will be more efficient if it is used to replace the 'Feature-sign search algorithm' in image super-resolution ?
links: Article Image Super-Resolution Via Sparse Representation
The choice of an efficient algorithm mainly depends on your application needs (Dictionary Learning, Denoising, Inverse Problem Solving/Image Restoration, etc.).
Neverthless, without any more information, I suggest that you jointly follow these 2 references in order to help you in solving your problem with computational efficiency:
- Abo-Zahhad et al., " Compressive Sensing Algorithms for Signal Processing Applications: A Survey", 2015 - https://file.scirp.org/pdf/IJCNS_2015060910194817.pdf
- Bajwa et al., "Toeplitz-Structured Compressed Sensing Matrices", 2007 - http://www2.ece.umn.edu/users/jdhaupt/publications/ssp07_toep_cs_matrices.pdf
Regards
Thank you Sir for responding to my question. Actually what happens, if I try to use ISTA/FISTA for performing patch based reconstruction of Super-resolution image as done in the paper..Article Image Super-Resolution Via Sparse Representation
it requires much time as for each patch it solves the sparse representation L1-minimization problem. Since I am not working on the whole image at once, so setting a high iteration number in ISTA/FISTA algorithm is not preferred. Is there any other efficient method for such requirements, i.e. solves patch wise L1-minimization for an image with faster result.
Regards
Helal
DCT-based L1-regularization for robust data interpolation might be an efficient solution. Follow:
- Tepper et al., "L1 Splines for Robust, Simple, and Fast Smoothing of Grid Data", 2012 - https://arxiv.org/pdf/1208.2292
In search of efficiency, switching to L2-Regularization could be interesting. To help you choose the best compromise, check out:
- Van Den Doel et al., "The lost honour of L2-based regularization", 2012 - http://math.boisestate.edu/~calhoun/GEMRES/PNWNAS/Ascher_talk.pdf
Regards
- Badges
- Science topic
- Similar topics
- Computer Science
- Images