The goal of the project is firstly to understand the state of the art with regards to establishing dense correspondence between two or more 3D objects. In particular, non-rigid facial models. This is done at present by an open source framework for modelling 3D morphable faces. It has been used in the past to determine age, gender, and cultural background from the 3D model scans. The face data is stored as a point cloud.
I will first work to understand the current code (Python), and then try to use similar statistical clustering to determine the expression of the 3D face, which has not been done before.
As a simplified example, subtracting a neutral face model from a smiling face model, will result in only the parts of the model related to smiling (for example, the mouth, corners of the eyes). This can apply to any expression. In theory, I will then be able to apply expressions on to neutral faces.
Finally, I will look into applications of this for other areas. While this project is only for 3D faces, it can apply to any non-rigid objects.
Edith Cowan University
Paimoe is a high-performing third-year student studying data science and mathematics at Edith Cowan University. Bringing a background of freelance programming, he hopes to continue into the fields of computer vision, machine learning and AI, as well as having an eye towards postgraduate studies in mathematics and statistics. To unwind, he enjoys quizzes, movies and everything rugby.