| Summary: | 碩士 === 國立清華大學 === 電機工程學系 === 96 === 3D Face Reconstruction is an important research topic nowadays. Because of the bottleneck on 2D Face Recognition and Face Synthesis technique, 3D Face Model has better accuracy and performance than 2D Face Image on Recognition and Synthesis. But 3D scan equipment is really expensive and hard to acquire. 3D model Reconstruction from 2D images is a cheaper way to get 3D data.
There are 2 main approaches for 3D face model reconstruction from 2D images. The first approach is Morphable Model proposed by Volker Blantz. Another approach is a factorization method based on Structure from Motion (SfM) proposed by Kanade. Morphable model has high accuracy, but takes much time for model fitting. SfM has lower accuracy than morphable model, but SfM is fast.
In this thesis, an automatic 3D model reconstruction system in image sequence is proposed. This system is based on SfM method. It contains modifying AAM feature extraction, model reconstruction and a refinement process we proposed.
In the experiment, our method can reduce the time spent in the feature extraction process. Our method also can reconstruct a 3D model which has good quality. The speed of this system is also acceptable.
|
|---|
