CCEDA: building bridge between subspace projection learning and sparse representation-based classification

Representation-based classification (SRC) is a face recognition breakthrough of recent years, but the dimensionality reduction for SRC has not been well addressed. The reason why existing dimensionality reduction methods are not effective for SRC is revealed for the first time. Based on analysis of the classification mechanism of SRC, the novel dimensionality reduction method for SRC is proposed, i.e. class coding error discriminant analysis (CCEDA), which simultaneously maximises the inner-class coding error and minimises the intra-class coding error. Extensive experiments show that the CCEDA feature achieves a better performance than the other features when using SRC as the classifier.