2D-3D图像视频转换的深度图提取方法综述.doc

2D-3D图像/视频转换的深度图提取方法综述 李可宏，姜灵敏，龚永义 广东外语外贸大学信息学院，广州，510006 摘 要 ：目的：深度图提取是计算机视觉领域的研究热点。随着3D显示设备的普及，2D-3D图像/视频转换的深度图提取研究受到越来越多国内外学者的关注。本文回顾了深度图提取研究历程，并对已有成果进行分类、概括和评述。方法：由于深度图提取方法的实现主要依赖于深度线索，不同方法存在人机交互程度上的差异，本文采用基于深度线索和基于人机交互程度的两种分类方法。结果：根据深度线索的不同，将深度图提取方法分为基于单目线索的方法、基于双目线索的方法和基于混合线索的3类方法。然后从人机交互的角度，将深度图提取方法分为人工法、半自动法和全自动法。介绍了这些方法的基本思想，比较归纳不同方法的优点与不足。最后,阐述了近年来热门的机器学习方法在深度图提取的应用。结论：文章结尾对深度图提取研究进行简要的总结和展望。2D-3D；深度图提取；立体视觉；深度线索；机器学习 Survey of depth map extraction methods in 2D-3D image/video conversion Li Kehong, Jiang Lingmin, Gong Yongyi Cisco School of Informatics, Guangdong University of Foreign Studies, Guangzhou 510006 Abstract: Objective: Depth map is the hotspot in computer vision research field. With the development and promotion of 3D display equipments, Depth map extraction methods are being paid more and more attention. This paper reviews the development of the depth map extraction and summarizes the existing methods. Method: This paper adopts the classifications based on depth cues and human-computer interaction degree. Result: The existing methods are grouped into three categories: monocular depth cue based, binocular depth cue based and multiple depth cues based according to different cues. Then, these methods are classified into three schemes based on different human-computer interaction, namely: manual, semi-automatic and automatic methods. This paper focuses on the basic principles of these methods, and points out their advantages and limitations. In addition, a detailed analysis is taken of the application and development of machine learning methods in depth extraction. Conclusion: Future development of depth extraction are discussed，including adoption of new methods from the current research hotspots and introduction of new depth cues. Key words: 2D-3D; depth map extraction; stereo vision; depth cue; machine learning 0 引言 随着数字图像处理技术与显示技术的发展，显示设备经历从黑白、彩色标清显示到彩色高清显示的发展过程。人们对视觉效果的更高要求促进显示设备实现3D显示效果，给用户带来真实生动的观赏体验。目前能够用于显示的3D视频资源种类和数目十分匮乏，限