Parameterized discriminant analysis for image classification

In recent years, linear and nonlinear (i.e., kernel) discriminant analysis has been proposed to address the difficulties of small sample problem, curse of dimensionality, and multi-modality of image data distribution in content-based image retrieval (CBIR). The existing discriminant analysis is implemented either in a regular way such as MDA (C=2 for FDA) or in a biased way such as biased discriminant analysis (BDA). In this paper, a rich set of parameterized discriminant analysis is proposed as an alternative of the regular MDA and BDA when taking the regularization into account to avoid the singularity of the scatter matrices. Extensive experiments are carried out for performance evaluation and the results show the superior performance of the parameterized discriminant analysis over regular MDA and BDA for both linear and nonlinear settings.