基于深度特征聚类的海量人脸图像检索

针对海量人脸图像数据库检索时长的问题,提出了一种基于深度特征聚类的海量人脸图像检索算法.该算法首先使用人脸图像训练集对深度卷积神经网络模型进行人脸图像分类训练,在此基础上采用三元组损失方法对已训练好的人脸图像分类网络模型进行微调,使得网络能够更加有效地提取人脸图像的高层语义特征,构建更具有表征性的人脸图像深度特征.其次采用K-means聚类算法对提取的人脸图像深度特征进行聚类,使得同一个人的人脸图像能够划分到同一簇中,然后在相应的簇中进行人脸图像的深度特征相似度匹配执行人脸图像检索任务.为了进一步提高系统的检索性能,提出人脸图像深度特征融合的查询扩展方法,对待检索的人脸图像深度特征进行融合再次执行检索任务得到最终的检索结果.通过在两个人脸检索数据集(Celebrities Face Set和Labeled Faces in the Wild dataset)上进行详尽实验验证,结果表明,该算法能极大地缩小海量人脸图像数据库的检索范围,在保证一定准确率的前提下有效地提高了人脸图像检索的速度.

Massive face image retrieval based on depth feature clustering

A massive face image retrieval method based on depth feature clustering is proposed to overcome the long time retrieving in a huge scale face image database. Firstly, the convolutional neural network model is trained for face classification by face image training set. Based on it, the triplet loss method is used to fine-tune the model so that the network can be more efficient to extract high-level semantic features and construct a more representative depth features of face image. Secondly, the K-means clustering algorithm is used to cluster the extracted depth features, so that face images of the same person can be divided into the same cluster, and then similarity matching of face images is performed in the corresponding clusters to perform the retrieval task. In order to further improve the performance of system retrieval, the face image feature fusion query expansion method is proposed to fuse the depth features of the face image to be retrieved. Through exhaustive experimental verification on two face retrieval datasets (Celebrities Face Set and Labeled Faces in the Wild dataset), the results show that the proposed method can significantly reduce the retrieval range of massive face image database, thus effectively increasing the face image retrieval speed while ensuring similar retrieval accuracy.