结合局部关键点集与测地线的三维人脸识别

3DFace Recognition Combining Local Keypoints with Isogeodesic Curves

为了克服面部表情变化导致的三维人脸识别精度不高的问题,提出了一种结合局部关键点集与测地线的三维人脸识别算法。首先,根据表情变化对人脸识别具有分区域影响的特性,将三维人脸划分出刚性区域和非刚性区域;然后将由鼻部和眼部组成的区域作为刚性区域,进行有效关键点检测,提取多种几何特征,构成局部描述子,进行相似度匹配;接着在非刚性区域提取测地线环带并进行相似度匹配;最后将两个区域的匹配程度进行加权融合,得到最终的匹配结果。该算法分别在Bosphorus和FRGC v2.0数据库上进行了实验验证,结果表明算法识别率分别达到了97.01%和98.63%,由此证明本文算法对三维人脸的表情变化有较强的稳健性。

To overcome the challenge of low accuracy of 3 Dface recognition caused by facial expression variations,a new algorithm that combines local keypoints with isogeodesic curves is proposed herein.First,a 3 Dhuman face is divided into rigid and non-rigid regions as facial expressions have different impacts on different regions.The rigid region comprising the eyes and a nose is proposed and effective keypoints are detected in this area.Then,a variety of geometric features are extracted to form a local descriptor for similarity matching.Furthermore,isogeodesic curves extracted from the non-rigid region are utilized for similarity matching.Finally,the matching degrees of the two regions are weighted and combined to obtain the final matching result.A number of experiments are conducted using the public databases Bosphorus and FRGCv2.0 and results show that the recognition rate of the proposed algorithm could reach 97.01% and 98.63%,respectively.The algorithm proposed herein is robust to variations in 3 Dfacial expressions.