Fully automatic 3D facial cosmetic surgery simulation

Abstract

Before cosmetic surgery procedures, the patients are commonly involved in the consultation process using interviews and reference images. The reference images typically consist of pre-post surgery images of other patients, leading to misunderstandings between patients and the surgeon. This thesis presents a fully automatic pipeline to simulate the whole face of post-surgery results. We first establish a 3D face registration and alignment based on the face surgery procedures of the current day and then generate the delta image. We proposed the delta image to solve the lack of dataset dilemma of the pre-post surgery face images. We also propose a convolutional autoencoder model to select the most similar face. Furthermore, we simulate the post-surgery results using surgery procedures retargeting. The last section validates the results using a subjective survey and shows the implementation notes.