Photometric shape recovery of microstructure objects

Recovering the 3D shape of an object from its 20 images has always been a popular topic of computer vision. Various areas of computer graphic have applications for shape recovery, such as the inspection of manufactured products, 3D object recognition, facial recognition and model construction from physical objects.;In this thesis, a shape recovery technique has been developed to reconstruct the 3D surface of a micro structure object from its optical microscope images. One of the challenges of this research is the capillary tube is positioned between two glass layers and a strong reflection from the bottom layer. To deal with this problem, object images with the same viewing direction, but different light directions, are initially captured using a common optical microscope. Next, masks are created manually to eliminate regions principally illuminated by the reflected light. A modified photometric method was developed and the method experimented with pure Lambertian and non pure Lambertian surfaces. Ultimately this method becomes weak if specular reflection is present. To overcome this problem, a photometric method, with global optimization, is studied further and adopted. By experimenting with synthetic images, results show a significant improvement in the recovered object shape from the image containing specular reflection.;Both methods are applied to a set of synthetic testing images. Error analysis is presented to evaluate the accuracy of the results. The methods are also applied to images of real capillary tubes. The results appear to be very reasonable. However, verification of their accuracy is not feasible at this stage.;To the best of our knowledge, this research is the first preliminary study to apply photometric methods to a glass-type micro structure 3D object using a conventional optical microscope. It is hoped the experience, obtained from this study, will lead to future, valuable exploration.;Recently, researchers at the Petroleum Technology Research Centre (PTRC) in Regina showed interest in constructing a 3D model of capillary tubes from optical microscope images in their research. However, there are relatively few shape-from-image techniques dealing with micro structure object images. In addition, the majority of their applications are either not designed to process typical microscope machine images or have limitations regarding object surface properties.