Atomic Decomposition by Basis Pursuit
SIAM Review
Convex Optimization
Robust Face Recognition via Sparse Representation
IEEE Transactions on Pattern Analysis and Machine Intelligence
Gabor feature based sparse representation for face recognition with gabor occlusion dictionary
ECCV'10 Proceedings of the 11th European conference on Computer vision: Part VI
Alternating Direction Algorithms for $\ell_1$-Problems in Compressive Sensing
SIAM Journal on Scientific Computing
Is face recognition really a Compressive Sensing problem?
CVPR '11 Proceedings of the 2011 IEEE Conference on Computer Vision and Pattern Recognition
Robust classification using structured sparse representation
CVPR '11 Proceedings of the 2011 IEEE Conference on Computer Vision and Pattern Recognition
Are sparse representations really relevant for image classification?
CVPR '11 Proceedings of the 2011 IEEE Conference on Computer Vision and Pattern Recognition
Sparse Representation for Color Image Restoration
IEEE Transactions on Image Processing
Close the loop: Joint blind image restoration and recognition with sparse representation prior
ICCV '11 Proceedings of the 2011 International Conference on Computer Vision
Sparse representation or collaborative representation: Which helps face recognition?
ICCV '11 Proceedings of the 2011 International Conference on Computer Vision
| Hi-index | 0.01 |
In this paper, a novel classification framework called two-dimensional relaxed representation (2DRR) is proposed for image classification. Different from recent popular vector-based representations with/without sparsity which encode a vector signal as a sparse/nonsparse linear combination of elementary vector signals, 2DRR is based on 2D image matrices, where each column of the input matrix signal is represented by a combination of the corresponding columns of the elementary matrices. In order to preserve the global linear coding relationship between the input matrix and these elementary matrices, the proposed 2DRR constrains the coding coefficients corresponding to each column of the input matrix to be locally close. Then two algorithms are derived from the 2DRR framework under the l"2 norm and the l"1 norm respectively. Extensive experimental results show the effectiveness of the proposed algorithms in comparison to three existing algorithms.