Supervised Two-Dimensional CCA for Multiview Data Representation
Since standard canonical correlation analysis (CCA) works with vectorized representations of data, an limitation is that it may suffer small sample size problems. Moreover, two-dimensional CCA (2D-CCA) extracts unsupervised features and thus ignores the useful prior class information. This makes the extracted features by 2D-CCA hard to discriminate the data from different classes. To solve this issue, we simultaneously take the prior class information of intra-view and inter-view samples into account and propose a new 2D-CCA method referred to as supervised two-dimensional CCA (S2CCA), which can be used for multi-view feature extraction and classification. The method we propose is available to face recognition. To verify the effectiveness of the proposed method, we perform a number of experiments on the AR, AT&T, and CMU PIE face databases. The results show that the proposed method has better recognition accuracy than other existing multi-view feature extraction methods.
KeywordsMulti-view learning Canonical correlation analysis Two-dimensional analysis Dimensionality reduction
This work is supported by the National Natural Science Foundation of China under Grant Nos. 61402203, 61472344, 61611540347, 61703362, Natural Science Fund of Jiangsu under Grant Nos. BK20161338, BK20170513, and Yangzhou Science Fund under Grant Nos. YZ2017292, YZ2016238. Moreover, it is also sponsored by the Excellent Young Backbone Teacher (Qing Lan) Fund and Scientific Innovation Research Fund of Yangzhou University under Grant No. 2017CXJ033.
- 2.Andrew, G., Arora, R., Bilmes, J.A., Livescu, K.: Deep canonical correlation analysis. In: ICML, pp. 1247–1255 (2013)Google Scholar
- 8.Sharma, A., Kumar, A., Daume, H., Jacobs, D.W.: Generalized multiview analysis: a discriminative latent space. In: CVPR, pp. 2160–2167 (2012)Google Scholar
- 11.Sun, T., Chen, S., Yang, J., Shi, P.: A novel method of combined feature extraction for recognition. In: ICDM, pp. 1043–1048 (2008)Google Scholar