A Facial Fatigue Expression Recognition Method Based on Sparse Representation on the Low-Resolution Image

Abstract

AbstractIn order to effectively improve the performance of facial fatigue expression recognition on the low-resolution image, a method of fatigue facial expression recognition based on Sparse Representation is proposed. At present, study on facial fatigue expression recognition is almost based on high-resolution, high-quality images. In the network environment, especially in the Internet of Things environment, the images of facial fatigue expression mostly are low-resolution images, in which the performance of facial fatigue expression recognition will degrade with image quality. In order to improve the recognition rate on the low-resolution image, a facial fatigue expression recognition method based on sparse representation on the low-resolution image is proposed. At the same time, in order to improve the credibility of the results of the experiment,Kendall’s coefficient of concordance method was used to construct the low-resolution facial fatigue expression database—TIREDFACE, whose words mean human’s tired-face. The TIREDFACE database consists of about 240 Image sequences of 10 university students. Size of all images is normalized to 92 × 112 pixels with 8-bit precision for grayscale values. Firstly, TIREDFACE database is set up, and then we exploit the discriminative nature of sparse representation to perform fatigue expression detection. Instead of using the generic dictionaries, we represent the test sample in an overcomplete dictionary whose base elements are the training samples themselves. Compressed sensing theory was used to solve its sparsest representation. With sufficient training samples for each fatigue status, including awake, slight fatigue and severe fatigue, it is done to represent the test samples as a linear combination of just those training samples from the same classification, and the sparse representations of the low resolution facial fatigue expression image of the identified test sample in the database TIREDFACE are given. After that, according to the sparsest representation solution, the low-resolution facial fatigue expression status classification is performed. In order to further investigate the advantages of this method, we compare this method with the linear classifier,the nearest neighbor (NN), support vector machine (SVM) and the nearest subspace (NS) on our database TIREDFACE. Experimental results show that the recognizing accuracy on low resolution facial fatigue expression images by the method of this paper is much higher than the linear classifier,the nearest neighbor (NN), support vector machine (SVM) and the nearest subspace (NS), etc. Therefore, the proposed method in this paper on the low-resolution facial fatigue expression recognition tasks is feasible.KeywordsCompressed sensingSparse representationFatigue expression recognitionKendall’s coefficient of concordance