Time-Varying Mesh Generation Based on Iterative Feedback between Silhouette Extraction and Geometry Modeling
Toshihiko Yamasaki, Kentaro Yamada, Kiyoharu Aizawa · Hokkaido University Collection of Scholarly and Academic Papers (Hokkaido University) · 2009
This paper proposes a shape-from-silhouette-based 3D geometry modeling scheme that is robust to defective silhouette extraction. The silhouette extraction is a significantly important task because it directly affects the quality of the generated 3D models. Our proposed approach improves the silhouette extraction and as a result the generated 3D models by the iterative feedback between the 3D modeling using the improved silhouette images and the silhouette updating using the rendered 3D images. Namely, the generated 3D models are utilized to obtain better foreground/background seeds for graph-cuts-based silhouette extraction and vice versa. In this paper, we propose two modeling approaches aiming at more accurate seeds generation. As a result, 3D modeling featuring much less loss/surplus of voxels has been made possible. Experimental results demonstrated that the loss of voxels can be reduced from 2.1% to 0.90% and the surplus of voxels can be reduced from 9.4% to 1.2% as compared to the initial 3D model. In addition, the error rate is the minimum among the conventional approaches.