Synthesizing Neutral Facial Expressions on 3D Faces

Agianpuye, Agianpuye Samuel (2015) Synthesizing Neutral Facial Expressions on 3D Faces. Masters thesis, Universiti Malaysia Sarawak, (UNIMAS).

PDF (Please get the password by email to repository@unimas.my, or call ext: 3914/ 3942/ 3933) Agianpuye Samuel Agianpuye ft.pdf Restricted to Registered users only Download (4MB)

Abstract

Facial expression synthesis is a process of generating new face shapes from a given face and still retaining the distinct facial characteristics of the initial face. The generated facial expressions can be used to improve the performance of existing face recognition systems. Earlier work on synthesizing face shapes used 2D face images. As 3D scanners become more improved and widely available, the work has moved from 2D to 3D faces. The advantage of 3D faces over 2D image data is that 3D face holds more geometric shape data and is invariant to poses and illumination. This project presents a new approach to synthesize neutral facial expression on realistic 3D faces called Expression Proportion Distribution (EPD). EPD uses statistical approach to derive a method to neutralise facial expressions. The main challenge is to neutralise facial expressions especially those with jaw dropped and opened mouth. Jaw dropped and opened mouth facial expressions may be generated during articulations, or expressing emotional facial expressions, such as laughing or surprise. Opening of mouth moves both the facial muscles and the mandible, which causes the geometric face shape to deform. Other facial expression with mouth closed is also looked into. The experiments were carried out on two realistic 3D face datasets from Imperial College London and from the Binghamton University - 3D Facial Expression Dataset (BU-3DFED). The proposed neutral expression synthesis approach is evaluated in a face recognition domain.

Actions (For repository members only: login required)

View Item