Three dimensional interference from a single photograph using shading and shadows

Abstract

We can only see a two dimensional information from a photograph. People are then interested in building a three dimensional model from a single photograph. Shape from shading is the most well-known technique to reconstruct three dimensional images. In this thesis, I would like to apply the fast marching method to tackle the problem of shape from shading under the orthographic projection. Under orthographic projection, it assumes that there is only a single light source and it is far enough from the object. Shape from shading is the technique that uses the different intensity and shading on the object to recover the three dimensional image. In my final year project, I will make use of the singular points as a vision cue to recover object shape. Singular points are the essential information in solving shape from shading problem. They are the strongest intensity position on an object and appear at the local maximum and local minimum positions. Once determine how singular points are related and time to start propagation in fast marching method. The object shape is recovered. First, we input the intensity map of the object to find the singular points. There are not exactly one but several singular points with maximum intensity at the local maximum and local minimum position. So blob extraction to group the singular points is necessary. Secondly, each singular point is then put to the fast marching method and calculates the relative height difference with each other. When there are N singular points, NxN matrix is formed. A novel algorithm is then applied to the matrix for finding out the order and time of singular points start propagation. At last, according to the starting time of singular points, we load them into the fast marching method. An object shape is reconstructed. It will further extend to the non-frontal light source. Then the experimental results are reported.