LBDAG-DNE: Locality balanced subspace learning for image recognition

Abstract

The cloud-computing environment makes it possible to select the best features when tuning parameters. Various dimensionality reduction algorithms can achieve the best features with the tuning of parameters. Double adjacency graphs-based discriminant neighborhood embedding (DAG-DNE) is a typical graph-based dimensionality reduction method, and has been successfully applied to image recognition. It involves the construction of two adjacency graphs, with the goal of learning the intrinsic structure of the data. However, it may impair the different degrees of importance of the intra-class information and inter-class information of the given data. In this paper, we develop an extension of DAG-DNE, called locality balanced double adjacency graphs-based discriminant neighborhood embedding (LBDAG-DNE) by considering the intra-class information and inter-class information of the given data differently. LBDAG-DNE can find a good projection matrix, which allows neighbors belonging to the same class to be compact while neighbors belonging to different classes become separable in the subspace. Experiments on two image databases illustrate the effectiveness of the proposed approach.