Facial Expression Hallucination Through Eigen-Associative Learning

摘要

As an important human characteristic, facial expression plays an important role in the applications of identity authentication, animation production, human-computer interaction, web-based education, etc. In this paper, we propose a facial expression hallucination approach through eigen-associative learning (EAL). The approach consists of two steps, in the first step, the global facial expression is estimated, and in the second step, we synthesize high-frequency image features to enhance the global face. The proposed EAL approach is adopted in both steps, which can synthesize the imaginary facial expressions of the input face with neutral expression. Compared with existing method, the EAL approach can be easily applied to new test data and retain high computational efficiency. Experiments show that the EAL approach generates reasonable imaginary facial expressions.