Self-correctional 3D Shape Reconstruction from a Single Freehand Line Drawing

The goal of sketch reconstruction is to take an inaccurate, 2D edge-vertex graph (i.e., sketch drawing) as input and reconstruct a 3D shape as output. However, traditional reconstruction methods based on image regularities tend to produce a distorted 3D shape. In part, this distortion is due to the inherent inaccuracies in the sketch, but it also relates to the failure to accurately distinguish between important and less important regularities. We propose a new self-correctional reconstruction algorithm that can progressively produce refined versions of sketch reconstructions. The algorithm corrects the shape and the drawing simultaneously using geometric error metrics. The proposed algorithm can minimize the distortion of the shape by adding 3D regularities to the image regularities. The self-correctional algorithm for minimizing the distortion of sketch reconstruction is discussed, and the experimental results show that the proposed method efficiently reconstructs more accurate 3D objects than previous ones.