Multi-feature Shared and Specific Representation for Pattern Classification
Sparse representation has been widely applied to pattern classification,where the input is coded as a sparse linear combination of training samples and classified to a category with the minimum reconstruction error.In the recent years,multi-feature representation based classification has attracted widespread attention and most of these methods have showed the superiorities compared to the classification model with single feature.One key issue in multi-feature representation is how to effectively exploit the similarity and distinctiveness of different feature,which is still an open question.In this paper,we present a novel multifeature shared and specific representation(MFSSR)model,which not only keeps the distinctiveness of different features,but further exploits their similarity with a shared representation coefficient.In addition,different features are weighted differently to reflect their discriminative abilities.Several representative experiments have shown the effectiveness and simplicity of the proposed MFSSR.
Multi-feature representation Shared and specific representation Pattern classification
Kangyin Ke Meng Yang
School of Data and Computer Science,Sun Yat-Sen University,Guangzhou,China
国际会议
广州
英文
573-585
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）