Robust, discriminative and comprehensive dictionary learning for face recognition
作者：
Highlights：
• A robust, discriminative and comprehensive dictionary learning (RDCDL) method is proposed.
• The robust dictionary is learned from sample diversities by extracting real face variations and generating virtual face images.
• RDCDL learns the dictionary including class-shared dictionary atoms, class-specific dictionary atoms and disturbance dictionary atoms to completely represent the practical data.
• Discriminative regularizations on the dictionary and the representation coefficients are used to exploit discrimination information.
• RDCDL outperforms those state-of-the-art methods.
摘要
•A robust, discriminative and comprehensive dictionary learning (RDCDL) method is proposed.•The robust dictionary is learned from sample diversities by extracting real face variations and generating virtual face images.•RDCDL learns the dictionary including class-shared dictionary atoms, class-specific dictionary atoms and disturbance dictionary atoms to completely represent the practical data.•Discriminative regularizations on the dictionary and the representation coefficients are used to exploit discrimination information.•RDCDL outperforms those state-of-the-art methods.
论文关键词：Dictionary learning,Face recognition,Sparse representation
论文评审过程：Received 31 October 2016, Revised 7 February 2018, Accepted 23 March 2018, Available online 30 March 2018, Version of Record 18 April 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2018.03.021
