Image Clustering Using Exponential Regularized Discriminant Analysis

When clustering images, the images are typically sampled as nonlinear manifolds. In this case, local learning-based image clustering models are used. Several proposed clustering models are based on linear discriminant analysis (LDA). In image clustering based on linear discriminant analysis (LDA), the problem of small-sample-size (SSS) is presented when the dimensionality of image data is larger than the number of samples. To solve this problem, various image clustering models based on local learning have been introduced. In the proposed clustering models, we added tuning parameters to deal with the small-sample-size (SSS) problem arising in linear discriminant analysis (LDA). In this paper, we propose an exponential regularized discriminant clustering model as an image clustering model based on local learning. In the proposed local exponentially regularized discriminant clustering (LERDC) model, the local scattering matrices of the regularized discriminant model are projected into the exponential domain to address the SSS problem of LDA. Compared with previous clustering methods based on local learning, k-nearest neighbors and regularization parameter λ in the local exponentially regularized discriminant clustering model are the tuning parameters for clustering. The experiments are concluded that the clustering performance of the proposed LERDC model is comparable to that of the clustering methods based on previous local learning.