基于单幅图像的三维人脸建模及应用研究

发布时间：2018-03-09 06:36

本文选题：三维人脸建模　切入点：形变模型　出处：《上海交通大学》2009年硕士论文　论文类型：学位论文

【摘要】： 三维人脸建模是计算机图形学和计算机视觉领域最根本。最困难的问题之一,也是近年来国际上最活跃的研究热点。真实感人脸建模的研究在电影。广告。人物动画。计算机游戏。视频会议。人机交互。医学。公安安全等许多领域有广阔的应用前景,其中人脸识别是重要应用之一。目前的人脸识别方法主要集中在二维图像方面,但由于受到光照。姿势。表情变化等因素的影响,其识别的准确度受到很大的限制。利用三维人脸模型是解决目前问题的有效方法之一。另外,随着通讯业和数字娱乐产业的发展,由计算机自动生成三维卡通人像的技术将得到广泛应用。以动画。卡通。视频及网络游戏。手机游戏等为代表的动漫产业已经成为许多国家的经济增长点。作为数字化动漫的关键技术的三维卡通人像生成技术,也值得深入探索。传统的三维人脸建模方法大多需要提取多幅图像的人脸信息,再通过修改一般模型完成。形变模型τMorphable Modelυ方法是近几年出现的三维人脸建模新方法,可以针对单幅图像进行三维建模。虽然该模型同以往的参数模型。肌肉模型以及视觉模型相比,具有真实感好。自动化等特点,但目前该模型在模型建立。优化等方面还存在诸多问题,在模型应用方面也缺乏探索。针对上述问题,本文在形变模型方法的基础上,对单幅图像的快速人脸建模方法及应用方面展开了研究,主要的工作内容包括: 1)基于单幅图像的三维人脸建模方法及改进。在研究了三维人脸形变模型的方法的基础上,针对其现存的一些问题,提出了改进的基于稀疏形变模型的人脸建模方法,顺利地解决了传统形变模型方法中需要将三维人脸库进行像素级对齐的困难问题,实现了基于单幅照片进行快速三维人脸建模的系统。主要技术包括稀疏模型的建立,人脸图像的检测和特征点提取,参数的最优化,以及几何模型的弹性变形技术。另外,提出了基于肤色模型的皮肤纹理合成方法,通过引入肤色模型的插值技术,增强了三维人脸侧面皮肤的真实感和应用价值。 2)三维人脸表情的合成。研究了基于表情编码系统生成表情的方法,将其应用在三维人脸模型上,实现了灵活可控的三维人脸的表情合成。 3)三维人脸卡通化。研究了计算机自动生成卡通人脸的技术,并将其应用在三维人脸模型上。提出了基于全局结构和局部形状进行自动特征夸张的人脸卡通化方法,实现了由单幅人脸照片生成多种不同风格的卡通人脸技术。 4)人脸识别的应用研究。本文针对人脸识别中的光照和姿态问题,研究了基于三维人脸模型的人脸识别方法。利用三维人脸合成了多种姿态,光照和表情的虚拟人脸图像,用于补充训练样本,增强了多角度人脸识别的准确性和鲁棒性。通过对以上几个方面的深入研究,本文实现了基于单幅图像的三维人脸自动重建方法,解决了三维人脸建模及应用中的一些关键问题,获得了一定研究成果。
[Abstract]:The 3D face modeling is the most fundamental field of computer graphics and computer vision. One of the most difficult problems in recent years is also the most active research focus. Research itcanbewidelyusedinmanyareas in movies. Advertising. Character animation. Computer games. Video conferencing. Human-computer interaction. Medicine. Many areas of public security have application prospect broad, including face recognition is one of the important applications.
Current face recognition methods mainly focus on the two-dimensional image, but because of the influence of illumination. Influence factors such as changes in posture. The expression, the recognition accuracy is very limited. Using the 3D face model is one of the effective methods to solve the problem at present. In addition, with the rapid development of communication industry and digital entertainment industry, automatic generate 3D cartoon portrait by computer technology will be widely used. In the cartoon animation.. video and network games. Mobile phone game as the representative of the animation industry has become the economic growth in many countries. As a 3D cartoon portrait generation technology of key technology of digital animation, but also worthy of further exploration.
The traditional method of 3D face modeling mostly needs to face information extraction of multiple images, then by modifying the general model. The deformation model Morphable Model V is a new method for 3D face modeling in recent years, 3D modeling for a single image. Although the model with the parameters of the model. Compared with the muscle model and vision model is realistic. Automation, but the model established in the model. There are still many problems in the aspects of optimization, model application is also a lack of exploration. Aiming at the above problems, based on the deformation method, fast face modeling method for single image and application aspects of the research. The main contents include:
1) the method and improvement of 3D face modeling based on single image.
Based on the method of 3D morphable models, aiming at some problems of it, and proposed a face modeling method based on improved sparse morphable model, successfully solve the traditional deformable model need to be 3D face database of pixel level alignment problem, realize the system of three-dimensional face based on the modeling of a single photo. The establishment of the main technologies include sparse model, face image detection and feature extraction, parameter optimization, and geometric model of elastic deformation technology. In addition, the skin texture synthesis method based on skin color model, skin color model by introducing interpolation technique, the 3D real face skin sense and application value.
2) synthesis of 3D facial expression.
The method of expression based on expression coding system is studied, which is applied to 3D face model, and the expression synthesis of flexible and controllable 3D face is realized.
3) 3D face cartoon.
Study on the automatic generation of computer cartoon technology, and its application in the 3D face model is proposed. The method of automatic face cartoon features exaggerated global structure and local shape based on the cartoon faces from a single face image to generate a variety of different styles.
4) the application of face recognition.
Aiming at the face recognition of illumination and pose problems, research on face recognition method based on 3D face model. The gesture was synthesized by using 3D virtual face images, light illumination and expression, used to supplement the training samples, enhances the accuracy and robustness of the multi angle face recognition.
Through in-depth study of the above aspects, this paper implements a 3D face automatic reconstruction method based on single image, which solves some key problems in 3D face modeling and application, and achieves certain research results.

【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP391.41

【引证文献】