[2K1-GS-7a] Vision, speech media processing

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[2K1-GS-7a-03] Comparison of Facial Expression Estimation Performance Using a Race-Specific Point Distribution Model

〇Rikuu Irie¹, Kazuya Mera¹, Yoshiaki Kurosawa¹, Toshiyuki Takezawa¹ (1. Hiroshima City University)

Keywords:

Point Distribution Model,Facial Expression Analysis,Machine Learning

The Point Distribution Model (PDM) is an indicator representing the overall trend of facial feature changes. However, the PDM used in OpenFace was created based on Caucasian facial images, potentially making it difficult to capture expression features of non-Caucasian individuals.
Therefore, this study creates a new PDM from Japanese facial images and analyzes the differences in key features between the Caucasian PDM and the Japanese PDM. Furthermore, using a machine learning model constructed based on features calculated from each PDM, we compare the performance of facial expression estimation for Caucasians and Japanese individuals.
Comparing the constructed PDMs revealed that the top-ranking features of the Caucasian PDM and the Japanese PDM represent significantly different trends in facial expression changes. Furthermore, in facial expression estimation experiments, using different PDMs resulted in partially different correct answer tendencies for each race.