The Optimization Of Massive Face Image Storage And Index

With the construction of smart city,Face Recognition technology has been applied in more and more scenes,among which how to store face images and quickly search face images is an important research topic.At present,distributed file system is widely used in storage system.HDFS,GFS and other distributed file systems are the mainstream distributed file systems,but they are mainly used for large files.While image similarity retrieval is mainly based on content retrieval,extracting the feature vectors from images and calculating distance for measuring the similarity among images in the past,but the feature vectors of face images are not only complex but also multi-dimensional.Therefore,how to design efficient index for face image with high dimension is an important technical problem.Firstly,this thesis deeply analyzes the performance bottleneck of the mainstream distributed file system technology for small image files storage and the principle of Asymmetrical distance calculation and inverted indexing mechanism(IVFADC),then proposes targeted solutions.The research content of this paper are as follows:1.This thesis analyzes the advantages and disadvantages of the distributed file systems used nowadays.FastDFS distributed file storage system is used to store massive face images.Designing image files storage scheme to store the massive face images,not only effectively enhances the speed of massive human face images storage,but also designing storage scheme fully exploits FastDFS storage space.2.IVFADC is a good index mechanism for integrating product quantization,asymmetric distance calculation and inversion index for dealing with high dimensions.However,this algorithm which adopts the scheme of quantifying facial feature vectors and assigns them to a unique inversion linked list ignoring some boundary vectors,so it cannot obtain a high recall rate.Z-IVFADC improved IVFADC index mechanism.In the process of global feature vector quantization,IVFADC uses the "one to one" in which a feature vector can only be assigned to a clustering,but Z-IVFADC uses the "one to many" in which a feature vector can only be assigned to multiple clustering in accordance with the set distance range,the experiment shows that the Z-IVFADC algorithm make the recall ratio of massive face images and the retrieval time improved.