Reading Texure paper
Reconstructing textured meshes from multiple range+rgb maps
ISTI-CNR
summary
- Minimizing redundancy and optimizing the color attribute represendtation
- Eliminate most of the color difference or discontinuity which exist in input images
- cross-correlation
- interpolation
Work flow
- visibility calculation
- get a set of valid cameras/images for the face
- ray-tracing calculation[耗时]
- hardware-accelerated OpenGL
- 每个面一个颜色,用OpenGL直接渲染,选择相机能看到的颜色作为可观测颜色
- 但如果多个面都投影到同一个像素的话,太小的面就会被判定为不可见
- 优化
- 通过迭代,每次迭代哪些尚未观测到的面
- 随机抖动相机,或微调景深z-index,使得小面被观测到
- get a set of valid cameras/images for the face
- patches generation
- reduce the number of patch
- sub-texture packing
- locally-optimum greedy process
- continuous texture but generally presents insufficient color continuity
- imporving color matching and continuity
- difference of color[iterate on vertices]
- 面$f$点$v$,$image_1\to patch_1$ has color $c_1$
- $image_2\to patch_2$ has color $c_2$
- $\bar image_{1,2}-c_1$意味着,$patch_1$到$patch_2$需要改变的颜色
- pull-push interpolation method
- difference of color[iterate on vertices]
state of the art
- texture mapping
- image registration
- point pair[点对]
- integrate geometric- and image-based alignment[几何对齐]
- silhouette-based registration[轮廓匹配]
- Tsai algorithm啥玩意
- texture image is reconstructed from input image
- map to mesh, sample color with non-redundant manner and optimizing
- weight blend
- image registration