Unsupervised Emotional Scene Detection for Lifelog Video Retrieval based on Gaussian Mixture Model

For the purpose of an efficient retrieval of impressive scenes from lifelog videos, we propose an emotional scene detection method based on facial expression recognition. Most of conventional facial expression recognition methods focus on dis- criminating typical facial expressions such as happiness, sadness and surprise, while lifelog videos contain various facial expressions. In addition, many training examples, which are quite troublesome to prepare, are required to construct the facial expression recognition models. The proposed method tries to solve these problems by constructing a facial expression recogni- tion model using an unsupervised learning based on Gaussian mixture model. Since our model is unsupervised, there is no need for preparing learning examples and predefining the types of facial expressions. The detection performance of the proposed method is evaluated in terms of detection accuracy and efficiency through several emotional scene detection experiments.