Sparsity preserving score for joint feature selection

Abstract

Based on recent advances in sparse representation technique, we propose in this paper Sparsity Preserving Score (SPS) to jointly select features. SPS evaluates the importance of a feature by its power of sparse reconstructive relationship preserving, which is achieved by minimizing an objective function with l 1-norm regularization and binary constrain. Our searching strategy, which is an essentially discrete optimization, jointly selects features by projecting the original high-dimensional data to a low-dimensional space through a special binary projection matrix. Theoretical analysis guarantees our objective function can get a closed form solution, which is as simple as scoring each feature by Frobenius norm of sparse linear reconstruction residual for each feature. Comparing experiments on two face datasets are carried out. The experimental results demonstrate the effectiveness and efficiency of our algorithm. © 2013 Springer-Verlag Berlin Heidelberg.