基于单张图像的三维重建研究-信息材料专业论文.docxVIP

I I 摘要 近年来，三维显示技术发展迅速，在日常生活和工业产品中得到广泛应用。而如何高效 地获取三维数据，已经成为目前制约其进一步发展的主要瓶颈。传统的三维重建技术，例如 立体视觉法，由运动恢复结构法等，需要从两张或多张图片中恢复出物体的三维结构。但这 些方法都忽略了人类立体视觉感知中重要的单目线索，例如透视、纹理、阴影等。针对此问 题，本文围绕基于单张图片的三维重建这一研究课题，充分挖掘单目立体感知线索在三维数 据获取中的作用，主要工作有三方面： 第一，提出了一种基于 K-Means 算法的快速图像分割方法，在保证分割质量的同时，提 高了分割速度，为后续的三维重建过程提供了良好基础； 第二，将单目三维重建技术应用于扫描电子显微镜（Scanning Electron Microscope，SEM） 所成图像，利用由阴影恢复结构技术，采用基于线性逼近的方法，恢复样本表面的三维结构， 获得更好的三维可视化效果； 第三，将基于单张图像的三维重建技术与人脸识别技术相结合，解决人脸识别技术在现 实应用中存在的难题，并提出一种应用于会议社交中的新颖应用。 通过实验验证，本文工作在相关领域均取得良好效果，为三维重建这一课题提供了新的 研究思路。 关键词: 三维重建，单目立体感知，图像分割，SEM 图像，人脸识别 II II Abstract Due to the rapid development of 3D display technology in recent years, it has been widely used in our daily life and industrial products. Nevertheless, challenges such as how to acquire 3D format data efficiently for displaying, have become a significant bottleneck restricting its further development. For most traditional 3D reconstruction methods, such as stereovision, structure from motion and so on, two or more images are needed to recover 3D structure of the scene. However, these algorithms often end up ignoring the numorous additional monocular 3D perception cues that can also be used to obtain rich 3D information, e.g. perspective, texture and shading. To address this problem, this work explores the technique of 3D reconstruction based on single images, with the goal to fully exploit the monocular cues which are helpful for acquairing 3D 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 in 3D reconstruction. Secondly, in order to get a better 3D visualization effect of Scanning Electron Microscope (SEM) images, a shape from shading algorithm based on linear approximation is employed to recover the surface