Micro-expression recognition based on CBP-TOP feature with ELM

Abstract

Micro-expressions are short, involuntary facial expressions with revealing suppressed effect that people try to conceal. Detecting and recognizing micro-expressions have potential applications in clinical and national security fields. Centralized Binary Patterns from Three Orthogonal Panels (CBP-TOP) is a novel approach that can efficiently extract the micro-expression information in spatial and temporal domains. In this method, the micro-expression image sequences are initially preprocessed that includes face detection, interception, size normalization, and micro-expression detection. Then, the features of the human face are extracted from the blocks of images, using CBP-TOP operator. Finally, extreme learning machine (ELM) is used to recognize micro-expressions. The experimental results show that this method can extract motion features and dynamic texture information in micro-expression sequences more efficiently and has greater improvement in micro-expressions recognition rate than the traditional recognition methods.