动画人脸图像 - 使用变形? [英] Animate face image - using warping?

问题描述

我打算在 android 和 Iphone 中做一个类似下面的应用程序.

I am planning to do a similar kind of application like the below one in android and Iphone.

http://www.motionportrait.com/en/technology/

http://www.youtube.com/watch?v=P-mBdV3icMY

我尝试了什么?

1) 我尝试了图像变形技术来表达，我已经达到了不错的准确性.这是结果图像.

1) I tried image warping technique to make expressions, which I have achieved to a decent accuracy. And this is the result image.

2) 我也尝试过 opengl+Texture 映射，但我无法获得更好的准确性.

2) I have also tried opengl+Texture mapping where I couldnt get better accuracy.

我在哪里被困?

我在面部旋转等几件事上并不成功.

I am not success in few things like face rotation.

有人可以指导我了解此应用程序背后使用的技术，以及如何从这里继续进行吗?

Can some one guide me through the techniques used behind this application, and how to proceed further from here?

输入:我的应用程序的输入是一个 2D 正面图像.

Input: Input to my application is a 2D frontal image.

预期输出:旋转的脸和脸上的表情.

Expected Output: Rotated face and expressions on the face.

推荐答案

据我所知，要进行旋转，您应该拥有头部的参数化 (3D) 模型(20-100 个参数).获得这个模型是项目中最难的部分.通常你会得到大约 1000 次 3d 头部扫描，然后对它们应用 PCA，以获得参数.用户拍照后，您可以评估该模型的参数(变形).现在您拥有用户的 3D 头部模型和纹理模型的正面视图.您应该做的就是将照片投影到此模型并获取纹理.例如，未知区域填充了来自开放皮肤的纹理图案.

As I know, to do rotation, you should have parametric (3D) model of head (20-100 parameters). Getting this model is hardest part of project. Usually you get about 1000 3d-head scans, then apply PCA to them, to get parametrs. After user takes a photo you can evaluate parameters of this model (morph it). Now you have 3D head model of the user and frontal view of textured model. All you should to do is to project photo to this model an get your texture. Unknown areas filled with texture patterns from open skin, for example.

这篇关于动画人脸图像 - 使用变形?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！