Pain detection from facial images using unsupervised feature learning approach.

In this paper a new method for continuous pain detection is proposed. One approach to detect the presence of pain is by processing images taken from the face. It has been reported that expression of pain from the face can be detected utilizing Action Units (AUs). In this manner, each action units must be detected separately and then combined together through a linear expression. Also, pain detection can be directly done from a painful face. There are different methods to extract features of both shape and appearance. Shape and appearance features must be extracted separately, and then used to train a classifier. Here, a hierarchical unsupervised feature learning approach is proposed in order to extract the features needed for pain detection from facial images. In this work, features are extracted using convolutional deep belief network (CDBN). The extracted features include different properties of painful images such as head movements, shape and appearance information. The proposed model was tested on the publicly available UNBC MacMaster Shoulder Pain Archive Database and we achieved near 95% for the area under ROC curve metric that is prominent with respect to the other reported results.