Reconstruction from Range Data: Model Coloring

Color is applied to the polygonal model that was obtained through iso- surface extraction using Marching Cubes [Lore 87]. We use an inverse ray- casting technique that assigns a color to each triangle of the polygonal model by reprojecting the triangles back to the original input images used in generating the space carved data set. Each triangle in the polygonal model is subdivided until its projected footprint in the images is subpixel size, so that it can simply take on the color of the pixel to which it projects. In most cases, a triangle projects to several of the original images. We combine the different colors from the images by weighted averaging. The weight of each color is calculated by taking the dot product between the triangle normal vector and the view direction vector of the camera that captured the color. Cameras with viewing directions that are nearly perpendicular to the triangle normal contribute less than those with viewing directions that are nearly parallel to the triangle normal. To simplify the calculation, the dot product is performed once the triangle has been transformed into the camera coordinate system so that the view direction vector is aligned with the z axis. We use z-buffering to ensure that only cameras with an unobscured view of the triangle can contribute to the triangle color. The figure below is a comparison of four of the original input images with rendered images of the reconstructed implicit surface from the same camera viewpoints. The images on the left are the original input images, while those on the right are the rendered images.