Cross-database micro-expression recognition based on transfer double sparse learning

doi:10.1007/s11042-022-12878-0

2022-05-25

PRE

Jiateng Liu

Yuan Zong

Wenming Zheng *

DOI：10.1007/s11042-022-12878-0

原文链接

原文求助

摘要

En 中文

In recent years, the cross-database micro-expression problem has become a research hotspot in the affective computing and multimedia areas due to its vital role in analyzing human behavior and potential valuable application such as criminal investigation, lie detection and education which are closely associated with multimedia. Unlike common micro-expression recognition problem, cross-database micro-expression conducts micro-expression recognition using a database as training set (source database) while another database as testing set (target database), which is more challenging than common micro-expression recognition issue since it has a serious inconsistency of feature distribution between source database and target database. To handle the crucial cross-database micro-expression issue, a novel transfer double sparse learning method is proposed in this paper. The advantage of the proposed transfer double sparse learning model is that it can select the features and facial regions which have contributions to the cross-database micro-expression problem efficiently while further refining their corresponding features according to the importance of these features in cross-database micro-expression. Extensive experiments on three widely used micro-expression databases show that the proposed transfer double sparse learning model gets the best performance than other state-of-the-art methods. Specially, transfer double sparse learning model achieves which proves that the proposed it can cope with the cross-database micro-expression problem efficiently since it successfully refines the facial features and bridged the emotion gaps between different domains.

Keyword:

Transfer learning

Cross-database micro-expression recognition

Domain adaptation