Structural sparse representation with class-specific dictionary for ECG biometric recognition

•We propose a structural sparse representation algorithm with class-specific dictionary for ECG biometric recognition.•A mixed regularization is employed to heartbeats which belong to same subject for obtaining a robust and stable feature.•A graph regularization is used to minimize the difference between heartbeats of intra-class.•We also learn a sub-dictionary for each class making the feature more discriminative.