计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (6): 207-212.DOI: 10.3778/j.issn.1002-8331.1812-0346

• 图形图像处理 • 上一篇    下一篇

改进的SIFT结合余弦相似度的人脸匹配算法

魏玮,张芯月,朱叶   

  1. 河北工业大学 人工智能与数据科学学院,天津 300401
  • 出版日期:2020-03-15 发布日期:2020-03-13

Improved SIFT Algorithm Combined with Cosine Similarity for Face Matching

WEI Wei, ZHANG Xinyue, ZHU Ye   

  1. School of Artificial Intelligence, Hebei University of Technology, Tianjin 300401, China
  • Online:2020-03-15 Published:2020-03-13

摘要:

针对人脸图像匹配在光照、姿态、表情等复杂背景下匹配耗时较长且正确率较低的问题,提出一种改进的SIFT(Scale Invariant Feature Transform,尺度不变特征变换)结合余弦相似度(Cosine Similarity,CS)的人脸匹配算法,通过构建圆形分区的特征描述符,降低特征向量维数,利用正反双向匹配以及匹配点对集中各匹配点对之间近似满足余弦相似的原则,采用余弦相似度来进行误匹配点对的剔除。在FEI人脸数据库上与目前流行的人脸匹配算法进行对比实验,实验结果证明了该算法在保证人脸匹配正确率和匹配点对数量的前提下,匹配速度平均提高2~2.5倍。

关键词: 尺度不变特征变换, 特征描述子, 余弦相似度, 人脸图像匹配

Abstract:

To solve the problem that face image matching takes a long time and has a low correct rate in complex background such as illumination, posture and expression, a face matching algorithm based on improved SIFT(Scale-Invariant Feature Transform) and Cosine Similarity(CS) is proposed, which constructs the features of circular partition. Descriptor reduces the dimension of eigenvector, uses forward and backward bidirectional matching and the principle of cosine similarity between pairs of matching points in the set to approximately satisfy the cosine similarity, and uses cosine similarity to eliminate mismatching point pairs. Experiments on FEI face database are compared with the popular face matching methods. The experimental results show that the algorithm improves the matching speed by an average of 2~2.5 times on the premise of guaranteeing the correct rate of face matching and the number of matching points.

Key words: Scale-Invariant Feature Transform(SIFT), feature descriptor, Cosine Similarity(CS), face image matching