基于口罩佩戴情况下的人脸识别方法

Face Recognition Method under the Condition of Wearing a Mask

摘要: 随着新冠疫情的暴发，生活的方方面面都需要佩戴口罩。在正常条件下人脸识别的发展已经相当成熟，但佩戴口罩对人脸识别的精度和速度产生重大影响。SIFT算法对尺度、位置、旋转、光照等具有很好的鲁棒性。传统的SIFT识别算法存在特征维度高、匹配计算量大的问题，造成特征点匹配困难。针对于此，本文对SIFT算法进行了优化，加入了特征初筛阶段，改进了特征相似性度量指标。实验结果表明，改进后的SIFT算法相比于传统的SIFT算法平均准确率提高了9.8%，验证了算法的有效性。

Abstract: With the COVID-19 outbreak, masks are needed in every aspect of life. The development of face recognition under normal conditions has been quite mature, but wearing masks has a significant impact on the accuracy and speed of face recognition. SIFT algorithm has good robustness to scale, position, rotation and illumination. The traditional SIFT recognition algorithm has the problems of high feature dimension and a large amount of matching calculation, which makes it difficult to match feature points. In view of this, SIFT algorithm is optimized in this paper, the feature screening stage is added, and the feature similarity measurement index and corner screening constraint criterion are improved. Experimental results show that the average accuracy of the improved SIFT algorithm is increased by 9.8% compared with the traditional SIFT algorithm, which verifies the effectiveness of the algorithm.