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: 44641 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Ghaemmaghami, Shahrokh; Fatemizadeh, Emad
- Abstract:
- Sparse representation has attracted a lot of attention during the last few years. This adaptive representation has been used as an alternative of the classic transforms. In this kind of representation, signals are decomposed in terms of some basis functions. This basis functions are called “atoms” and their collection is called a “dictionary”. Dictionary learning should be such that signals have a sparse representation. Specified dictionary could apply some other properties except sparsity for the transform domain representation.In this thesis after study of methods convert the dictionary discriminative, we propose KLDA method for learning a discriminative dictionary. Also a new algorithm for subspace clustering will be proposed and use it to learn a dictionary efficient for unsupervised dictionary learning. Exploit of some training data is the success factor of sparse methods. Training data is the atoms of the dictionary. Regardless of the dimension of data that less or more than number of atoms, learning by some relevant atoms may be suitable. In some application like face recognition that number of training samples is less than dimension of data, learning of overcomplete dictionary is impossible. For these scenarios, sparse regression has been studied in this thesis. Effect of sparsity on the coefficients of regression and effect of sparsity on the error of regression, separately have been investigated. An algorithm for solving group sparsity problems is proposed; also a new robust regression is in theInnovations of the thesis
- Keywords:
- Sparse Representation ; Pattern Recognition ; Dictionary Learning ; Sparse Regression
Digital Object List
-
محتواي کتاب
- view
Bookmark