一种用于动态图像序列的人脸识别算法

Face Recognition Algorithm For Dynamic Image Sequence

研究基于动态图像序列的人脸准确识别。人体在运动的过程中,脸部会发生较大幅度的震动,造成人脸姿态特征不可避免的发现形变,造成对图像的干扰,识别误差较大。传统的识别方法多是以人脸姿态特征作为识别的基础的静态识别,一旦人体运动幅度加大,人脸姿态特征发生变化,必然造成识别的准确度下降。为了避免上述问题,提出了一种利用Curvelet布尔核转换的动态图像序列人脸识别的算法。利用Curvelet转换算法对全部动态图像序列相关参数进行降维处理,利用布尔核转换将形变较大的人脸关键细节特点进行约束分类,建立多约束的完整动态人脸特征模型,保证姿态形变可控,最终实现了运动中的人脸识别。实验证明,利用上述算法进行运动中的人脸识别,提高了人脸识别的准确率。

Research face recognition method based on dynamic image sequence to enhance the human recognition accuracy. The paper put forward a face recognition algorithm using Curvelet Boolean nuclear conversion of dynamic image sequences. Curvelet conversion algorithm was used to reduce the dimension of concerned parameters of all the dynamic image sequences and remove the random noise. Boolean nuclear conversion was used to classify the key de- tails of dynamic image sequences to set up the complete dynamic constraint face feature model. Finally, the dynamic image sequence face recognition was realized. Experiments show that the algorithm of dynamic image sequence face recognition can improve the accuracy of face recognition.