Fast facial smile detection using convolutional neural network for the students in intelligent learning environment

Abstract Infrared facial expression recognition is very important to understand the human emotion and behaviors. Smile detection plays a key role in facial expression recognition. In this paper, we have developed a fast facial smile detection algorithm with convolutional neural network and infrared imaging technology. Due to random noise and band overlap in the infrared imaging spectroscopy, we propose a maximum likelihood (ML) restoration algorithm with spectral-spatial L1-norm constraint for low-resolution infrared spectrum. In this model, we also employ the spectral-spatial L1 norm to contrain the smoothness of the infrared spectrum data. With the aid of spectral-spatial L1-norm minimization as compelling constraints, the proposed method can not only remove noise dramatically but also compute the instrument blurry function simultaneously. The results show that the developed method can recover a noise-free infrared spectrum, and the spectral structure is richer than the compared methods. The high-resolution infrared spectrum could be utilized for fast facial smile detection task in the intelligent learning environment.