Abstract | ||
---|---|---|
3D face reconstruction is a long-term and challenging problem, which has wide application scenarios, such as animation, recognition. Inspired by recent works in face landmark marching, we develop a method for 3D face reconstruction using a novel landmark updating optimization strategy. Our method is also based on minimization of the landmark projection error for calculation speed. To achieve more accuracy, contour landmarks and self-occluded landmarks need to be updated. For contour landmarks, the detected landmarks on 2D image are to updated by ?nding a nearest position on contour curve. For self-occluded landmarks, we render the model onto image plane, extract the edge of projected area, and generate new correspondence landmarks according to the edge pixels. We test our method quantitatively with ground truth data, and provide qualitative example reconstructions, show that our method performs better with experiments on MICC dataset. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/MIPR.2019.00082 | 2019 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) |
Keywords | Field | DocType |
Face reconstruction,3D Morphable Model,Landmark update,Self-occluded,Facial feature point | Computer vision,Projected area,Computer science,Image plane,Minification,Ground truth,Pixel,Animation,Artificial intelligence,Landmark | Conference |
ISBN | Citations | PageRank |
978-1-7281-1198-8 | 0 | 0.34 |
References | Authors | |
0 | 4 |