Three-dimensional Analysis of Faces and Application to Facial Animation with Expressions

本课题取得了以下研究成果：1.人脸的三维测量首先用激光测距仪测量了30多名被试的中性表情和六种主要表情（愤怒、厌恶、惊讶、恐惧、快乐、悲伤）。在该测量中获得了圆柱范围和表面颜色图像。为了消除受试者姿态和位置的影响，我们在人脸上定义了一个以物体为中心的坐标系。初步实验表明，所采用的特征点表示方法对人脸形状的变化具有近似不变性.使用距离图像的面部三维形状分析我们开发了一种通过计算表情期间距离图像中的特征点处的3D运动矢量来量化六种主要面部表情的方法。这样的运动矢量指示表情期间的面部变形。我们还开发了一种自动提取面部成分的方法， ...更多信息 nts从彩色图像，以及人脸识别的子空间方法的基础上，使用范围图像。已经观察到，图像的镶嵌提高了识别率。距离图像的图像变形已经建立了从中性图像生成3D面部表情图像的框架。该方法是基于应用图像变形技术的范围和表面的颜色图像。所采用的图像变形是基于径向基函数。现在，我们可以从一个中性表情的图像生成具有六个主要表情的面部图像。对生成图像的质量进行了主观评价.具有表情的面部动画面部动画是通过在生成具有上述六个主要表情的面部图像的过程中逐渐改变表情的幅度来计算的。此外，我们已经研究了基于动力学的运动动画，使用肌肉骨骼模型对未来的研究。少

We have obtained the following research results in this project.1. Three-dimendional Measurement of Faces Firstly faces of more than thirty testees were measured in neutral and six primary expressions (anger, disgust, surprise, fear, happiness, sadness) by using a laser rangefinder. Both cylindrical range and surface color images were obtained in this measurement. In order to remove the influence of pose and position of testees, we defined an object-centered coordinate system on a face. Preliminary experiments have shown that the employed feature point representation is approximately invariant to variations of facial shapes.2. Three-dimensional Shape Analysis of Faces Using Range Images We have developed a method of quantifying six primary facial expressions by computing 3D motion vectors at feature points in range images during expressions. Such motion vectors indicate the facial deformation during expressions. We have also developed a method of automatically extracting facial compone … More nts from color images as well as that of face recognition based on a subspace method using range images. It has been observed that the mosaicing of images increases the recognition rate.3. Image Warping for Range Images A framework of generating a 3D facial expression image from a neutral image has been established . The method is based upon applying an image warping technique to range and surface color images. The image warping employed is based upon radial basis functions. Now we can generate facial images with six primary expressions from an image with neutral expression. The subjective evaluation has been conducted for evaluating the quality of generated images.4. Facial Animation with Expressions The facial animation is computed by gradually changing the magnitude of expression in the process of generating facial images with six primary expressions, which is mentioned above. Moreover, we have investigated in the motion animation based on dynamics using a musculoskeletal model toward the future research. Less