Differential evolution-based subspace clustering via thresholding ridge regression

Abstract

A robust subspace clustering assigns a label to each data point in a noisy and high dimensional dataset which has a collection of multiple linear subspaces of low dimension. To reduce the effect of noise in the dataset for subspace clustering, many methods have been proposed such as sparse representationbased, low rank representation-based, and thresholding ridge regression methods. These methods either reduce the noise in the input space (sparse representation and low rank representation) or in the projection space (thresholding ridge regression). However, reduction of noise in the projection space eliminates the constraints of spars errors and a prior knowledge of structure of errors. Further, thresholding ridge regression method uses k-means algorithm for clustering which is sensitive to initial centroids and may stuck into local optimum. Therefore, this paper introduces a modified thresholding ridge regression-based subspace clustering method which uses differential evolution and k-means algorithm. The proposed method has been compared with six different methods including thresholding ridge regression on facial image dataset. The experimental results show that the proposed method outperforms the existing algorithms in terms of accuracy and normalized mutual information.