基于三维重建的人脸识别

发布时间：2018-04-14 12:25

本文选题：三维人脸重建 + 三维人脸识别　；参考：《南京航空航天大学》2015年硕士论文

【摘要】：三维人脸识别技术是当前计算机视觉和模式识别领域的一个热门课题,受到国内外专家和学者的广泛关注。三维人脸识别技术相对于传统二维人脸识别方法有诸多优点,可以解决二维人脸识别中姿态、光照变化鲁棒性差的问题,但三维人脸样本不易获取成为了其广泛应用的一个阻碍。针对上述问题,本文研究了基于三维重建的人脸识别技术,利用三维重建算法将单幅人脸图像重建三维人脸模型,并以此进行识别。本文主要内容如下:首先,对三维人脸样本库做数据标准化操作。首先对样本库中模型进行姿态矫正,然后根据形变模型的原理,采用三维人脸的自动对齐方法对三维人脸样本库进行数据标准化。其次,对经典形变模型进行改进,提出基于边缘线的三维人脸重建算法。通过提取二维人脸图像特征区域中的边缘线,采用改进型遗传算法进行寻优求解,此算法改善了形变模型重建效率。再次,对三维人脸识别算法进行研究,提出了基于测地线采样的人脸识别算法。根据三维人脸上的相邻点在模型发生较小形变时,其测地线距离相对不变的原理,在模型上提取测地线邻域特征表示三维人脸。经实验验证,此算法运算效率较高,并且对表情变化具有良好的鲁棒性。最后,对基于三维重建的人脸识别方法进行了验证。对人脸三维重建和三维人脸识别综合起来加以研究,介绍一种基于三维重建的3D+2D多模态人脸识别方法。经过实验验证,本文提出的基于三维重建的3D+2D人脸识别算法对单幅图像姿态变化具有较好的鲁棒性,较经典算法对姿态人脸识别率有所提升。
[Abstract]:The technology of 3D face recognition is a hot topic in the area of computer vision and pattern recognition, attention of experts and scholars at home and abroad. The technology of 3D face recognition compared with traditional 2D face recognition methods have many advantages, can solve the attitude of 2D face recognition, illumination robustness problem, but 3D face samples easy access to become a obstacle to its wide application. In view of the above problems, this paper studies the technology of face recognition based on 3D reconstruction, 3D reconstruction algorithm to reconstruct 3D face model from a single face image, and then recognition. The main contents of this paper are as follows: firstly, the data standardization operation of 3D face samples first. The model sample attitude correction, and then according to the principle of deformation model, automatic alignment method using 3D face of 3D face like The library data standardization. Secondly, the classical deformation model is improved, the 3D face reconstruction algorithm based on edge extraction. The 2D face image feature edge line, using the improved genetic algorithm to solve optimization, this algorithm improves the efficiency of model reconstruction. The deformation of the 3D face again. The recognition algorithm, and proposed a face recognition algorithm based on geodesic sampling. According to the 3D face of the adjacent points in the model had a small deformation, the principle of measuring geodesic relatively constant in the model, extracting geodesic neighborhood feature representation of 3D face. Experiments show that this algorithm is computationally efficient, and robust good for facial expressions. Finally, the method of face recognition based on 3D reconstruction is verified. The 3D face reconstruction and 3D face recognition are combined to study, A 3D+2D multimodal face recognition method based on 3D reconstruction is introduced. After experimental verification, the proposed 3D+2D face recognition algorithm based on 3D reconstruction has better robustness to pose change of single image, and improves the recognition rate of face recognition compared with classical algorithm.

【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【相似文献】