Estimation of Facial Expression Intensity for Lifelog Videos Retrieval

Abstract

Facial expression intensity has been proposed to estimate the intensity of facial expression for the purpose of retrieving impressive scenes from lifelog videos. However, estimation of facial expression intensity involves manual work, and can only be relatively evaluated. We propose a new estimation method of facial expression intensity for reducing manual work, and for absolute evaluation. We estimate the proposed expression intensity for the lifelog videos used in previous research and MMI datasets, and compare it with the previous research, and evaluate the proposed method. As a result, it is shown it is possible to reduce the manual work while maintaining the estimation accuracy.