Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 49060 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Marvasti, Farrokh
- Abstract:
- The emerging field of compressed sensing has found wide-spread applications in signal processing. Exploiting the sparsity of natural image signals on basis of a set of atoms called dictionary, one can find numerous examples for applications of compressed sensing in the field of image processing. One of these interesting applications is to help recover missing samples of a damaged or lossy image signal which is also known as image inpainting. There are dozens of reasons why an image may get damaged, for instance, during data transmission, some blocks of an image (or frames of a video ) may get lost due to error in the telecommunication channel (this is known as block-loss). In this case image inpainting can be used to fill in the lost blocks of the image. Besides there are many other applications for image inpainting including automatic object/text removal and old art-work restoration (or retouching). There are a variety of algorithms for image inpainting. Most of these algorithms use iterative linear or non-linear methods to recover image from its remaining samples. As a means, sparse representation can be used to extract from the intact patches, the necessary information required for reconstructing the missing portions. There are two approaches for this purpose. The first one assumes for each patch a separate sparse representation within a transform domain and hence applies dictionary learning or sparse recovery algorithms to restore the missing samples. The second approach exploits non-local patch similarities in a natural image and applies low-rank matrix recovery to complete the approximately low-rank matrix resulting from concatenating these similar patches. In this thesis we investigate these two approaches. First we have a review on the techniques of sparse representation, sparse recovery and low-rank matrix completion. Then we propose efficient algorithms for reconstructing lossy image patches. Specifically we use a criterion based on structural similarity index for solving the optimization problem defined for sparse image recovery. We also propose algorithms for image inpainting which use one of the approaches introduced for exploiting sparse representation in image completion. These algorithms in fact try to improve the quality of existing methods and we compare our results with the methods from literature at the conclusion
- Keywords:
- Random Sampling ; Image Inpainting ; Sparse Recovery ; Low-Rank Matrix ; Structured Sparse Representation ; Structural Similarity Index
Digital Object List
محتواي کتاب
Bookmark