一种基于关键点和体素联合的交互头部三维重建方法

发布时间：2018-04-10 00:24

本文选题：三维重建　切入点：关键点　出处：《天津大学》2016年硕士论文

【摘要】：三维重建技术一直是计算机视觉领域的研究热点。头部的三维重建更是在人脸识别、医学、影视、安全等领域有着广泛的应用。在有足够的比较准确的关键点的情况下应用基于关键点的方法去重建三维模型,已经有了比较成熟的方法和技术。对于头部的三维重建情况,由于能从面部提取大量的关键点,而头发部分不能提取出足够的关键点,基于关键点的三维重建算法能够很好的重建面部部分,却不能够重建头发部分。传统的基于体素的三维重建方法由于只利用轮廓信息,只需要判断体素是否能投影到轮廓中,而不需要提取关键点。同时头发部分通常是近似于凸面,多角度的轮廓能够很好的反映头发部分的形状,从而能够较好的重建头发部分。本论文在对以上两种方法进行深入研究后,针对头部三维重建问题,提出了一种基于关键点和体素联合的交互式三维重建方法。其过程为:首先,经过关键点提取和简单的用户交互的GrabCut的图像轮廓提取;其次,用基于关键点的方法对头部进行重建;再次,由用户指定头部重建模型中需要重新重建的区域并对该区域进行基于体素的三维重建;最后,将两种结果保存在一起,得到头部三维重建的最终结果。通过这种联合的三维重建方法重建的三维模型比只用一种方法的重建结果更好。
[Abstract]:Three dimensional reconstruction technology has been a research hotspot in the field of computer vision. The head of the 3D reconstruction is in face recognition, film and television, medical, security and other fields are widely used. Based on the key point of the method to reconstruct the 3D model is used in the key points of enough accurate circumstances, have methods and techniques mature. For the three-dimensional reconstruction of the head, because a large number of key points can be extracted from the facial hair, and some key points can be extracted enough, the key points of the 3D reconstruction algorithm can reconstruct facial part well based on, but not be able to rebuild the hair part. Based on the traditional method of 3D reconstruction of voxel due only use contour information, only need to determine whether the voxel can be projected to the contour, and does not need to extract the key points. At the same time the hair part is usually similar to the convex, multi angle contour can be Good hair reflects the shape of the part, so as to rebuild the hair part better. In this thesis, the above two methods after research for 3D head reconstruction problem, and propose an interactive 3D reconstruction method is the key point and the voxel based on joint. The process is as follows: first, after the image outline key point extraction simple user interaction and the extraction of GrabCut; secondly, using the method based on key points of head reconstruction; thirdly, specified by the user head reconstruction model needs to be reconstructed and 3D reconstruction based on the plain area; finally, the results of the two methods are stored together to get the final 3D head the reconstruction of 3D model reconstruction. The 3D reconstruction method of the combined ratio of only one method of reconstruction result is better.

【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】