三维人脸表情图像预处理研究
|
摘要
在三维原始数据的基础上,提出三维人脸表情图像预处理方法.该方法主要进行尖点去除、点云三角化、点云平滑和孔洞修补等有效的预处理工作,并手动提取15个特征点,采用K-NN算法分类实验大大提高三维人脸表情识别率和鲁棒性.
Though the research of 3D facial expression and image is now more and more extensive, the collection of three-dimensional raw data usually contains a lot of redundant information and interference information, which makes computation much larger. These problems will have a negative impact on the extraction of subsequent features, resulting in the extraction of feature points and low accuracy. Therefore,the 3D facial expression image preprocessing methods are put forward based on the raw 3D data. It mainly processes the effective pretreatment work, such as cusp removal, point cloud triangulation, point cloud smoothing and repair work of holes, etc. It manually extracts 15 feature points and uses the K-NN algorithm classification experiments to greatly improve the 3 d facial expression recognition rate and robustness.
Though the research of 3D facial expression and image is now more and more extensive, the collection of three-dimensional raw data usually contains a lot of redundant information and interference information, which makes computation much larger. These problems will have a negative impact on the extraction of subsequent features, resulting in the extraction of feature points and low accuracy. Therefore,the 3D facial expression image preprocessing methods are put forward based on the raw 3D data. It mainly processes the effective pretreatment work, such as cusp removal, point cloud triangulation, point cloud smoothing and repair work of holes, etc. It manually extracts 15 feature points and uses the K-NN algorithm classification experiments to greatly improve the 3 d facial expression recognition rate and robustness.
引文
[1]BENTLEY J L.Multi-dimensional divide and conquer[J].Communications of the ACM,1980,23(4):214-229.
[2]罗月童,石放放.结合K-D树和Shell的快速动态等值面光线跟踪法[J].中国图象图形学报,2015,20(2):193-201.
[3]GOP M,KRISHNAN I,SILVA C T.Surface reconstruction based on lower dimensional localized delaunay triangulation[J].Computer Graphics Forum,2000,19(3):467-478.
[4]陈伟,刘肖琳.一种快速三维散乱点云的三角剖分算法[J].计算机仿真,2009,26(9):338-341.
[5]李凤霞,刘咏梅.一种基于映射法的散乱点云Delaunay三角剖分算法[J].计算机应用研究,2015(3):950-953.
[6]成宏璟,王庆.基于空间无序点云的层次构造三角化方法[J].科学技术与工程,2010(9):2114-2117.
[7]黄淼,张海朝.散乱点云的三角划分算法研究[J].微计算机应用,2007,28(10):1039-1042.
[8]ZHAO W,GAO S M,LIN H W.A robust hole-filling algorithm for triangular mesh[J].Visual Computer,2007,23(12):987-997.
[9]陈相,童小华.基于三角格网的点云空洞修补算法及精度研究[J].测绘通报,2013(4):1-3.
[10]孟庆顺,宁芊,雷印杰.一种新的基于参数估计的自适应双边滤波算法[J].计算机应用研究,2017,34(7):2221-2223.
[11]刘冲,颜世玉.多类运动想象任务脑电信号的KNN分类研究[J].仪器仪表学报,2012,33(8):1714-1720.
[2]罗月童,石放放.结合K-D树和Shell的快速动态等值面光线跟踪法[J].中国图象图形学报,2015,20(2):193-201.
[3]GOP M,KRISHNAN I,SILVA C T.Surface reconstruction based on lower dimensional localized delaunay triangulation[J].Computer Graphics Forum,2000,19(3):467-478.
[4]陈伟,刘肖琳.一种快速三维散乱点云的三角剖分算法[J].计算机仿真,2009,26(9):338-341.
[5]李凤霞,刘咏梅.一种基于映射法的散乱点云Delaunay三角剖分算法[J].计算机应用研究,2015(3):950-953.
[6]成宏璟,王庆.基于空间无序点云的层次构造三角化方法[J].科学技术与工程,2010(9):2114-2117.
[7]黄淼,张海朝.散乱点云的三角划分算法研究[J].微计算机应用,2007,28(10):1039-1042.
[8]ZHAO W,GAO S M,LIN H W.A robust hole-filling algorithm for triangular mesh[J].Visual Computer,2007,23(12):987-997.
[9]陈相,童小华.基于三角格网的点云空洞修补算法及精度研究[J].测绘通报,2013(4):1-3.
[10]孟庆顺,宁芊,雷印杰.一种新的基于参数估计的自适应双边滤波算法[J].计算机应用研究,2017,34(7):2221-2223.
[11]刘冲,颜世玉.多类运动想象任务脑电信号的KNN分类研究[J].仪器仪表学报,2012,33(8):1714-1720.