We propose a Joint Statistical and Spatial Sparse representation models the image or image-set data for classification, by reconciling both their local patch structures and global Gaussian distribution mapped into Riemannian manifold.
J3S model can be considered as the joint sparse coding problem and solved by coupling the local and global image representations using joint sparsity. The learned J3S models are used for robust image and image-set classification.
Here we take UIUC-material database as an example.
Usage: run J3S.m
Method | ETH-80 | FMD | UIUC | YTC |
---|---|---|---|---|
AHISD | 72.50 | 46.72 | 55.37 | 64.65 |
CHISD | 79.75 | 47.52 | 65.09 | 67.24 |
MMD | 85.75 | 60.60 | 62.78 | 69.60 |
MDA | 87.75 | 62.50 | 67.13 | 64.72 |
SPDML-AIRM | 90.75 | 63.42 | 74.72 | 67.50 |
SPDML-SETIN | 90.75 | 63.80 | 68.24 | 68.10 |
LEML | 93.50 | 66.60 | 69.17 | 69.85 |
RMML-SPD | 95.00 | 68.88 | 70.09 | 78.05 |
RMML-GM | 93.00 | 69.62 | 76.48 | 69.15 |
CDL-LDA | 94.00 | 76.92 | 78.89 | 70.21 |
CDL-PLS | 94.00 | 75.36 | 76.39 | 69.94 |
RSR | 91.50 | 74.92 | 72.59 | 72.77 |
KGDL | 93.00 | 77.40 | 76.32 | 73.91 |
Liu et al. | 92.5 | N/A | N/A | 78.8 |
DRM | 98.12 | N/A | N/A | 72.55 |
DMK | 96.8 | N/A | N/A | 80.3 |
MMDML | 94.50 | N/A | N/A | 78.5 |
J3S w/o Sp Dict. | 97.75 | 82.12 | 84.03 | 83.34 |
J3S | 98.25 | 82.36 | 85.65 | 83.40 |
Paper is available here.
In case of use, please cite our paper:
H. Cheng, and B. Wen. "Joint Statistical and Spatial Sparse Representation for Robust Image and Image-Set Classification." 2020 IEEE International Conference on Image Processing (ICIP). IEEE, 2020.
Bibtex is here:
@inproceedings{cheng2020joint,
title={Joint Statistical and Spatial Sparse Representation for Robust Image and Image-Set Classification},
author={Cheng, Hao and Wen, Bihan},
booktitle={2020 IEEE International Conference on Image Processing (ICIP)},
pages={2411--2415},
year={2020},
organization={IEEE}
}
If you have any questions or suggestions, please contact me
HAO006@ntu.edu.sg