Analysis of range images used in 3D facial expression recognition
Analysis of range images used in 3D facial expression recognition
复制标题
DOI：
10.1109/tencon.2013.6718813
复制
发表时间：
2013-10
期刊：
影响因子：
--
通讯作者：
Xiaoli Li;Q. Ruan;Gaoyun An;Yi Jin
中科院分区：
文献类型：
--
作者：
Xiaoli Li;Q. Ruan;Gaoyun An;Yi Jin
BU-3DFE（Binghamton University 3D Facial Expression）数据库的建立极大地促进了3D人脸表情识别（3D FER）的研究，但由于3D方法的局限性，3D FER的研究目前遇到了一些瓶颈。解决这一问题的一种新方法是将人脸表情从3D模型转换为2D数据，从而产生了深度图像，在当前的研究中起着中介作用。到目前为止，有各种各样的距离图像，但要确定哪一个对于3D FER是最佳的相对坚韧。本文旨在分析不同的距离图像生成的三维FER的性能。因此，不同的特征和分类器的表情识别框架进行了研究。实验的识别率均在88%以上，基本在同一范围内。结果表明，深度图像仍然保留了主要的判别信息，这对三维FER具有重要意义。同时，各种深度图像的性能也存在一定的差异。更重要的是，深度图像可以使三维FER自动实现。上述研究结果对三维人脸表情识别的进一步研究具有重要意义。
Researches on 3D facial expression recognition (3D FER) have been greatly fostered by the creation of BU-3DFE (Binghamton University 3D Facial Expression) database, but given limited 3D methods, they encounter some bottlenecks now. A new approach to solve this problem is to transform facial expression from 3D models into 2D data, thus range image emerges, playing an intermediary role in current researches. There are various range images up to date, but it's relatively tough to identify which one is optimal for 3D FER. This paper aims to analyse the performances of different range images which are generated for 3D FER. Therefore, different features and classifiers are investigated in expression recognition framework. The recognition ratios of proposed experiments are higher than 88%, almost falling within the same scope. It reveals that range images still retain main discriminative information, which is significant for 3D FER. Also, there are a few differences among various range images' performances. More importantly, the range images can make 3D FER implement automatically. The findings above are so remarkable that they would benefit the future research of 3D facial expression recognition.