Sparsity Enhanced Correntropy Based Robust Principal Component Analysis

Although Principal Component Analysis (PCA) find widespread use in many applications, it is sensitive to non-Gaussian noise. Many Robust PCA models have been developed to deal with this issue. However, these methods only can handle one type of noise, e.g. the impulse noise in the feature domain or the outliers in the sample domain. This paper developed a new robust PCA model based on sparsity enhanced correntropy called SCPCA, which is robust against impulse noise and outlier simultaneously. Furthermore, a novel algorithm is proposed to solve our SPCA model based on Fenchel conjugate and accelerated block coordinate update (BCU) strategy to solve SCPCA model. To assess the robustness of our SCPCA method, extensive experiments on background reconstruction and face modelling have been conducted. The findings show that in most cases, our SPCA method outperforms the compared state-of-the-art methods.