【作者】 梁丽；

【导师】 褚晶辉；

【作者基本信息】 天津大学 ， 信号与信息处理， 2012， 硕士

【摘要】 随着科技的进步和大众对生活品质的不断追求，记录真实三维世界的立体电影备受瞩目。三维电影给观众带来了前所未有的视觉冲击和丰盛的视觉享受，但同时也引发了新的问题——视觉疲劳。3D字幕对立体图像内容的干扰是一个主要原因。本文的研究目标是实现一个简单、便捷、观看舒适的3D字幕叠加系统。作者在认真分析和总结了国内外的3D字幕技术的研究成果基础上，以双目立体视觉技术为理论依据，全面分析人眼的立体视觉特性，重点剖析双目视觉和焦点调节与会聚两个影响因素，实现了一个基于双目视觉的3D字幕叠加的系统。该系统主要依据立体图像对的双目视差信息，采用OpenGL中文汉字将具有视差的字幕叠加在立体图像对上。根据人类的心理立体视觉因素将字幕区域选定在图像底部，同时字幕的属性也随图像内容自适应调整。针对3D字幕视差的计算，本文采用改进的基于线增长的立体匹配算法获取稠密视差图。改进措施有采用Harris法选取种子点，松弛迭代方法剔除误匹配，以保证种子点选取的高精度；用中值滤波消除匹配后的噪声，使视差平滑过渡，获得了较为理想的视差效果图。最终本文实现了基于立体图像内容的视差叠加3D字幕的系统，将叠加了3D字幕的立体图像对立体显示。实验结果表明，该3D字幕叠加系统可作为质量评价的测试平台，根据立体图像视差简单便捷的调整3D字幕的视差、属性等信息，以便达到同步舒适地观看立体图像和3D字幕的目的。更多还原

【Abstract】 With the technology development and increasing demand of high quality life, thestereoscopic film, which can capture the real information of the world has attractedheavy attention from the community. The stereoscopic content provides the illusion ofdepth perception and visually stimulating to the viewer, but a new problem imposedby stereoscopic film is eye strain. One key factor is the display of 3D subtitle onstereoscopic image content. The purpose of this paper is to realize a simple,convenient system for overlaying stereoscopic subtitles, thus viewers are able towatch the stereo images and read subtitles mutually with ease and comfort.The research of 3D subtitle technology home and abroad has been carefullystudied and summarized. After analysis of the binocular stereo vision and humanvisual feature, especially, the key influence factors------binocular disparity andconvergence & focus, the paper proposes that the main criteria for overlayingstereoscopic subtitles is the depth of stereoscopic image content. The system is basedon the principle of binocular disparity to overlay three-dimension subtitles by usingthe corresponding relation of 3D subtitles disparity and depth of field. The subtitlesare superimposed at the bottom of the image in terms of the human psychologicalstereo perceive. At the same time, the font attributes of subtitles changes adaptivelybased on the depth placement of image content. For the binocular disparitycomputation of stereoscopic subtitles, an improved dense stereo matching algorithmbased on line-growth is adopted. In order to guarantee the accuracy of the seeds, weadopt Harris corners detection to extract the seeds, and using relaxative iterationmethod to reduce the false match. Finally, we use median filter to eliminate imagenoise and unreliable disparity, thus obtaining clear disparity graph.In conclusion, a system for overlaying stereoscopic subtitles has beenimplemented based on the depth of stereoscopic image content. The experimentsshow that the system for overlaying stereoscopic subtitles, as a quality estimation testplatform, can adjust the disparity and attributes of subtitles conveniently, and enableviewers to read the stereo images and subtitles with ease and comfort.更多还原