This dissertation presents a new method to reconstruct 3D objects from a single image. We explore the Integration of our SFS approach with traditional CAD approaches to accelerate the quality of reconstructed objects. With our approach, the 3D object construction process is straightforward and effective. The results represent clear shape details on the reconstructed 3D object. In addition, we use image segmentation technique to specify the object region as a preprocessing for the 3D construction....[ Read more ]

This dissertation presents a new method to reconstruct 3D objects from a single image. We explore the Integration of our SFS approach with traditional CAD approaches to accelerate the quality of reconstructed objects. With our approach, the 3D object construction process is straightforward and effective. The results represent clear shape details on the reconstructed 3D object. In addition, we use image segmentation technique to specify the object region as a preprocessing for the 3D construction.

The first technique, image segmentation, is used to specify the interesting objects in the image. We apply "Snake" (the Active Contour Model) to deal with the image extraction problem. Unlike other low-level image segmentation operators, such as the edge operator, "snake" is a high level processing method which considers the entire geometric structure in the image and removes the edge-linking problem of low-level processing. Our extraction approach is not driven by gradient term. Thus our segmentation can detect the boundary without a gradient. However, since the classical active contour models include gradient term, these models only can detect the features with large gradient. By using a multi-pyramid structure, we accelerate the speed of image extraction.

Shape from shading (SFS) is a well-known technique in computer vision. In our approach, we use SFS to construct the 3D object described in the extracted image region. Our method is a global minimization SFS approach, which is based on a new energy function that represents the shading difference between the reconstructed image (obtained from the constructed 3D object) and the input image. A faster numerical method called conjugate gradient descent is applied to our iteration SFS method, instead of the classical variational calculus. By integrating our SFS technique with CAD tools, such as B-spline, our reconstructed results have clear micro surface structure. This technology may have a profound impart on various fields related to geometry modeling and reverse engineering.

Keywords: CAD, image segmentation, active contour, shape from shading, 3D object construction