基于纹理特征融合的人脸图像质量评估算法-《集美大学学报（自然科学版）》

文章信息/Info

Author(s):: CHEN Zhenghao1; WU Yundong1; CAI Guorong1; CHEN Shuili2; (1.School of Science,Jimei University,Xiamen 361021,China;2.Chengyi University College,Jimei University,Xiamen 361021,China)

Keywords:: face quality assessment; texture feature fusion; support vector machine

摘要:: 从视频序列中选取同一人的高质量人脸图像是人脸识别技术的关键步骤。为了提升人脸图像评估的可靠性，提出了纹理特征融合的人脸质量评估算法。首先，针对人脸图像提取HOG、GIST、GABOR和LBP等纹理特征；接着，根据标注数据训练分类器，实现单特征得分评估；再将多特征得分值融合成特征向量；最后，通过多项式核函数升维得到新的特征向量，并根据该特征训练SVMs分类器，以回归人脸图像质量得分。实验结果表明，基于特征融合的方法能有效提升人脸图像质量评估的效果，特别是HOG-GIST特征组合的算法具有很好的效率，目标人脸在不同姿势及遮挡的情况都能得到可靠的评估结果。

Abstract:: In face recognition technology,selecting high-quality face images from video is the key step.In order to improve the reliability of face image evaluation,this paper presents a face quality assessment algorithm based on texture feature fusion.First,the texture features of HOG,GIST,GABOR and LBP are extracted for face image.Secondly,according to the annotated data training classifier,the single feature score evaluation is realized,and the multi-feature score is integrated into the feature vector.Finally,the new feature vector is obtained through the polynomial kernel function,the SVMs classifier is trained with this feature vector,and the obtained classifier returns the quality score of face image.The experimental results show that this method can effectively enhance the effect of face image quality evaluation,especially HOG-GIST features combination algorithm has good efficiency,the reliable evaluation results can also be obtained for the target face different position and shadowing.