|
Due to the rapid development of3D display technology in recent years, it has been widely used in our daily life and industrial products. Nevertheless, challenges such as how to acquire3D format data efficiently for displaying, have become a significant bottleneck restricting its further development. For most traditional3D reconstruction methods, such as stereovision, structure from motion and so on, two or more images are needed to recover3D structure of the scene. However, these algorithms often end up ignoring the numorous additional monocular3D perception cues that can also be used to obtain rich3D information, e.g. perspective, texture and shading. To address this problem, this work explores the technique of3D reconstruction based on single images, with the goal to fully exploit the monocular cues which are helpful for acquairing3D information. In summary, the contributions of this work are three folds:Firstly, a fast image segmentation scheme based on K-Means algorithm is implemented. While producing comparable or even better segmentation quality, it has significant advantage over traditional K-Means on processing speed. It lays a solid foundation for futher processing in3D reconstruction.Secondly, in order to get a better3D visualization effect of Scanning Electron Microscope (SEM) images, a shape from shading algorithm based on linear approximation is employed to recover the surface structure of the specimen.Thirdly, by combining the monocular3D reconstruction with related face recognition algorithms, a novel application is proposed for conference socializing, which will provide a better socializing experience for users.According to the experimental validation, all the work stated above has achieved satisfactory performance, and is highly promising for providing new thoughts in related research fields.
|