L1/2 Regularization Based Low-Rank Image Segmentation Model

Abstract

In the spectral-type subspace segmentation models, the rank minimization problem was relaxed as Nuclear Norm Minimization(NNM) problem. However, to guarantee the success of NNM, one needs some strict conditions, and NNM may yield the matrix with much higher rank than the real one. In this paper, the L1/2 regularization is introduced into the low-rank spectral-type subspace segmentation model, combining Augmented Lagrange Multiplier(ALM) method and half-threshold operator, a discrete algorithm to solve the proposed model is given. A large number of experiments in section IV demonstrate the effectiveness of our model in data clustering and image segmentation.