几何约束下虚拟人脸重复视觉特征点匹配研究

Research on repetitive visual feature point matching of virtual face under geometric constraints

为最大程度地减少虚拟人脸特征点的误匹配,文中研究几何约束下虚拟人脸重复视觉特征点匹配方法。首先,利用高斯滤波构造虚拟人脸图像的多尺度空间,结合形状变更指数检测虚拟人脸在多尺度空间的重复视觉特征点;然后,采用基于动态特征矩阵求解(DFMS)的特征点匹配方法,完成重复视觉特征点初始匹配后,依据匹配点对的连接线距离、斜率一致的特点,构建最佳几何约束,有效删除错误匹配点对;最后,经RANSAC算法进行二次过滤后,实现了虚拟人脸的重复视觉特征点最佳匹配。实验结果显示,所提方法可在虚拟人脸的关键位置检测到重复视觉特征点,并最大程度地删除误匹配点对,实现重复视觉特征点的精准匹配。

In order to minimize the mismatching of virtual facial feature points,a method for matching repeated visual feature points of virtual faces under geometric constraints is studied.The Gaussian filtering is used to construct the multi-scale space for virtual face images,and combined with the shape change index to detect repeated visual feature points of virtual faces in the multi-scale space.The feature point matching method based on dynamic feature matrix solution(DFMS)is used to complete the initial matching of repeated visual feature points.Based on the consistent distance and slope of the connecting lines between the matching point pairs,the optimal geometric constraints are constructed to effectively remove mismatched point pairs.After secondary filtering by RANSAC algorithm,the best matching of repeated visual feature points of virtual faces is realized.The experimental results show that the proposed method can detect repeated visual feature points at key positions of virtual faces,and remove mismatched point pairs to the greatest extent,realizing the accurate matching of repeated visual feature points.