基于单目视频的高质量人脸动画生成

发布时间：2018-04-04 01:28

本文选题：人脸检测　切入点：人脸特征点追踪　出处：《浙江大学》2017年硕士论文

【摘要】：人脸动画在工业界有着广泛的应用,但传统人脸动画制作方法费时、费力、高成本,因此找到一个简单、方便、低成本的人脸动画制作方法具有重大意义。本文提出了一个基于单目视频的高质董人脸动画自动生成算法。该算法分为三部分:人脸检测、人脸特征点追踪和高质量人脸动画生成。首先,基于哈尔(Haar-Like)特征和自适应增强(Adaboost)算法,用已有的人脸图像数据库,训练出级联人脸检测器。通过人脸检测器可以检测出视频帧中的人脸区域。在首次获取人脸区域后,我们通过人脸特征点定位算法定位人脸上一些重要的特征点(嘴角、眼角、鼻尖等)。人脸特征点定位算法是基于形状索引(Shape-Indexed)特征的双层增强回归(Two-Level Boosted Regression)算法。之后,基于人脸特征点和多线性人脸模型(Multilinear Facial Model)重建出人脸三维形状与姿态。然后,通过明暗成形(Shapefrom Shading,SfS)算法重建出人脸精细局部形状(皱纹、褶皱等)。最后,将视频每帧的人脸精细局部形状信息应用到人脸三维形状上,完成高质量人脸动画捕捉。
[Abstract]:Face animation is widely used in industry, but the traditional method of face animation is time-consuming, laborious and expensive. Therefore, it is of great significance to find a simple, convenient and low-cost method for making face animation.In this paper, a high quality face animation automatic generation algorithm based on monocular video is proposed.The algorithm is divided into three parts: face detection, face feature point tracking and high quality face animation generation.Firstly, based on Haar-Like feature and adaptive enhanced Adaboostalgorithm, a cascade face detector is trained by using the existing face image database.The face region in a video frame can be detected by a face detector.After obtaining the face region for the first time, we use the face feature point localization algorithm to locate some important feature points (mouth corner, eye angle, nose tip, etc.).Face feature point location algorithm is based on shape index and Shape-Indexeded-based two-layer enhanced regression level Boosted Regression algorithm.Then, 3D face shape and pose are reconstructed based on face feature points and Multilinear Facial Model.Then, the fine local shape (wrinkles, folds, etc.) of the face is reconstructed by shape-shape-shading from shading SfSs algorithm.Finally, the fine local shape information of each frame of the video is applied to the 3D face shape to achieve high quality face animation capture.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】