Multi-Stereorekonstruktion nicht-lambert'scher Oberflächen mittels Graph-Cut Optimierung
Abstract
In the last years computer vision problems like dense depth map reconstruction from
given images became more and more important. There exists a lot of algorithms that are
trying to assign a depth value to each pixel of the input image. Most of them have a
strong assumption on the pixels of the object e.g. the color value constancy.
In this work we are trying to create a dense depth map for each input image without
setting any strong assumptions, like diffuse surface, on the objects. We postulate
that problem as energy minimization problem and try to solve it by using graph-cut approach.
We also utilize the GPU to compute some parts of the energy function that does
increase the whole computation speed immensely. At the end we show the stability of
our algorithm by computing the depth maps on "real world" images.
Read Thesis
- 3918 times clicked
Bibliography
A. Tevs, "Multi-Stereorekonstruktion nicht-lambert'scher Oberflächen mittels Graph-Cut Optimierung" , Naturwissenschaftlich-Technischen Fakultäten der Universität des Saarlandes, pages 1-47, 2006.