Facial Expression Segmentation Labeling and Modeling Using Movement Analysis
Chayanon Sub-r-pa, Ming-Zhong Fan, Rung-Ching Chen · 2023
Deep learning-based facial synthesis can generate realistic and detailed facial images. Nonetheless, altering the expression of a pre-existing facial image can be challenging. The primary challenge is preserving the original Image’s characteristics, such as skin tone, eye color, moles, freckles, or birthmarks. Limiting the editing area is an efficient method for preserving the identity of a facial image. And according to physiological research on facial expression, the most important areas are the eye and mouth. Concentrating on the eye and mouth can effectively reduce the modifying area. However, modifications beyond the eye and mouth are still being considered for a realistic facial image. Analysis of movement patterns is a potential method for locating the expression region or editing area. This study evaluated facial expression movement from the MMI Facial Expression Database to identify the expression area and presented a segmentation strategy that uses deep learning to extract the expression area.