| Presentation | 2010-05-27 Subspace learning for cone-constrained translation-invariant sparse image representations Makoto NAKASHIZUKA, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In this paper, we propose a sparse representation of images with cone-constrained subspaces. The sparse representation provides a linear image generative model under a sparsity penalty. Due to the sparsity, the image is represented with a small number of atoms that are included in a dictionary. The dictionary learning have been proposed to improve the sparsity of the representation. In the dictionary learning, the image local structures can be represented by the atoms of the dictionary. Especially, the dictionary, on which the translation-invariance is imposed, represents the set of the image structures that repeatedly appear in an image. However, such image structures appear with some fluctuations due to geometric-transformations, changes of lighting conditions and so on. In order to represent such fluctuations in the generative model, we extend the set of atoms to the set of cones in subspaces. To learn the set of subspaces, we propose an iterative algorithm based on an alternative update method for the vectors that span the subspaces and decomposition coefficients. In learning example, we demonstrate that the proposed method learns a set of the cones for distorted image patterns and separates an image into a set of components according to the learnt cones. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Sparse signal representation / blind signal separation / image analysis / image feature analysis / unsupervised learning |
| Paper # | EA2010-18,SIP2010-18,SP2010-18 |
| Date of Issue |
| Conference Information | |
| Committee | SIP |
|---|---|
| Conference Date | 2010/5/19(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Signal Processing (SIP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Subspace learning for cone-constrained translation-invariant sparse image representations |
| Sub Title (in English) | |
| Keyword(1) | Sparse signal representation |
| Keyword(2) | blind signal separation |
| Keyword(3) | image analysis |
| Keyword(4) | image feature analysis |
| Keyword(5) | unsupervised learning |
| 1st Author's Name | Makoto NAKASHIZUKA |
| 1st Author's Affiliation | Graduate School of Engineering Science, Osaka University() |
| Date | 2010-05-27 |
| Paper # | EA2010-18,SIP2010-18,SP2010-18 |
| Volume (vol) | vol.110 |
| Number (no) | 55 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |