Abstract
This article describes how in order to facilitate the retrieval of impressive scenes from lifelog videos, a method to estimate the intensity of a facial expression of a person in a lifelog video is proposed. The previous work made it possible to estimate the facial expression intensity, but the previous method requires some training samples which should be manually and carefully selected. This makes the previous method quite inconvenient. This article attempts to solve this problem by introducing an unsupervised learning method. The proposed method estimates the facial expression intensity via a clustering on the basis of several facial features computed from the positional relationship of a number of facial feature points. For the evaluation of the proposed method, an experiment to estimate the facial expression intensity is performed using a lifelog video data set. The estimation performance of the proposed method is compared with that of the previous method.Article Preview
Top2. Previous Research
In this section, the estimation method of Sakaue’s facial expression intensity is explained as the proposed method of the previous research (Sakaue, Nomiya, & Hochin, 2017). The facial expression intensity of the previous research is computed based on the positional relationship of several facial feature points.