2D-LPCCA and 2D-SPCCA: Two new canonical correlation methods for feature extraction, fusion and recognition

Two-dimensional canonical correlation analysis (2D-CCA) is an effective and efficient method for two-view feature extraction and fusion. Since it is a global linear method, it fails to find the nonlinear correlation between different features. In contrast, in this paper we propose a novel two-view method named as two-dimensional locality preserving canonical correlation analysis (2D-LPCCA), which uses the neighborhood information to discover the intrinsic structure of data. In other words, it uses many local linear problems to approximate the global nonlinear case. In addition, inspired by sparsity preserving projections (SPP), the two-dimensional sparsity preserving canonical correlation analysis (2D-SPCCA) framework is also developed, which consists of three models. Experimental results on real world databases demonstrate the viability of the formulation, they also show that the classification results of our methods are higher than the other's.