Photorealistic Scene Reconstruction by Voxel Coloring
CVPR '97 Proceedings of the 1997 Conference on Computer Vision and Pattern Recognition (CVPR '97)
Novel volumetric scene reconstruction methods for new view synthesis
Novel volumetric scene reconstruction methods for new view synthesis
| Hi-index | 0.00 |
This paper implements the reconstruction of 3D object from real scene input images using voxel coloring. We proposed new color consistency checking formula in voxel coloring by using mean charts. Using this color consistency checking, the system do not require to be given parameters as constant threshold input and the system can reconstruct more detail than using original color consistency checking formula using threshold on standard deviation of image pixels value. Photorealistic 3D object reconstruction is an active area in computer graphics. Input of this system is multiple photographs which are taken from multiple view points of scene. And then the object in the photographs will be projected back to the 3D coordinate and applied voxel coloring method to reconstruct and view the object in 3D. Voxel coloring works by discretizing scene space into voxels, the smallest unit in 3D, then traversed and colored in specific order. The output of voxel coloring is a collection of voxels that represent the object. Those voxels can be rendered to get a new view of that object. To determine whether voxel has to be colored or be emptied, some color consistency checking formula must be used. The original color consistency checking formula is standard deviation calculation of all image pixels which voxel has been projected and compare it with constant threshold which is defined by user. This constant threshold must be chosen carefully because wrong value can cause defect reconstruction. We propose new color consistency checking using mean charts which do not require user to choose constant threshold input.