摘要
针对人脸具有多样性和复杂性,同一人脸在不同条件下可能呈现出不同的形状和结构,导致了人脸特征点标定困难的问题,为准确提取出人脸特征点,提出基于遗传算法的三维图像人脸特征点标定。利用双目立体视觉系统实现人脸信息的多角度采集,从采集信息中提取主要特征信息,结合“三庭五眼”位置关系共性,在一般人脸模型上补偿面部轮廓以及五官局部变换坐标点信息,获得更接近真实人脸的三维坐标点位置信息集合。引入多种群遗传算法,以子种群内的基本染色体构成,表示三维人脸模型中的形状、姿态参数,通过采用不同的交叉和变异操作筛选最优染色体,实现三维人脸特征点的精准标定。实验结果证明,该方法建立的三维人脸模型真实度较高,特征点标定精准,标定的均方误差法(MSE:Mean Squared Error)最高为7.9%。
Faces have diversity and complexity,and the same face may exhibit different shapes and structures under different conditions,which makes it difficult to calibrate facial feature points.In order to accurately extract facial feature points,a 3D image facial feature point calibration based on genetic algorithm is proposed.A binocular stereo vision system is used to achieve multi angle collection of facial information,extracting main feature information from the collected information,and combining the commonality of the"three courtyards and five eyes"position relationship,compensating for facial contours and local transformation coordinate point information of facial features on a general facial model,obtaining a set of 3D coordinate point position information that is closer to the real face.A multi-population genetic algorithm is introduced to represent the shape and pose parameters of a 3D facial model based on the basic chromosomes within the subpopulation,and select the optimal chromosomes through different crossover and mutation operations to achieve accurate calibration of 3D facial feature points.The experimental results demonstrate that the proposed method establishes a three-dimensional facial model with high realism,accurate feature point calibration,and a highest MSE(Mean Squared Error)of 7.9%.
作者
李京燕
LI Jingyan(Film and Animation Department,Xi'an Academy of Fine Arts,Xi'an 710065,China)
出处
《吉林大学学报(信息科学版)》
2024年第6期1111-1116,共6页
Journal of Jilin University(Information Science Edition)
基金
国家民委民族研究基金资助项目(2021-CMD-098)。
关键词
双目立体视觉系统
三维人脸模型
多种群遗传算法
最优染色体的筛选
binocular stereo vision system
3D facial model
multi population genetic algorithm
screening of optimal chromosomes
