Abstract:

A novel image feature extraction method called 2dimensional nonparametric feature analysis based on difference criterion (2DDNFA) is proposed, which combines the ideas of 2dimensional linear discriminant analysis (2DLDA), maximum scatter difference (MSD), and nonparametric feature analysis (NFA). Firstly, the betweenclass and withinclass scatter matrices are constructed by using the neighbors of the samples in the 2dimensional image space. And then, the projection matrix is computed based on the difference criterion. Finally, the feature matrix of an image is obtained by projecting it on the projection matrix. Experiments conducted on the measuring synthetic aperture radar (SAR) data demonstrate that the proposed method is more efficient than the methods, such as 2DLDA, 2dimensional nonparametric feature analysis (2DNFA), and 2dimensional maximum scatter difference (2DMSD).