Face clustering method and device and computer readable storage medium

Abstract

The invention provides a face clustering method and device and a computer readable storage medium. The human face clustering method comprises the steps of obtaining human face features of at least onehuman face and space-time information corresponding to the human face features; determining the similarity of any two face features in the face features and the matching degree of space-time information corresponding to the two face features; based on the similarity of the two face features and the matching degree of the spatio-temporal information corresponding to the two face features, determining the joint probability that the two face features point to the same identity; and in response to the fact that the joint probability of the two face features reaches a first threshold, determiningthat the two face features point to the same identity and clustering the two face features. The face clustering method provided by the invention is higher in accuracy, and filing is more accurate by combining user behaviors, so that the method better meets the requirements of actual application scenarios, and has higher practicability.

Description

technical field [0001] The present invention relates to a method for sorting out face clustering, in particular to a conditional random field-based method, device and computer-readable storage medium for face clustering. Background technique [0002] Identity file sorting is an important application of face recognition technology in the surveillance and retail fields. The identity file sorting based on the face clustering algorithm can cluster face images according to identity information, transform the recognition of a single face image into face recognition for identity, and can effectively reduce the random noise caused by a single image ( For example, there are some particularly abnormal pictures, such as a picture whose features are not obvious because the person is wearing a hat or a mask), causing false passes and false alarms. In addition, face clustering can establish identities for collected face images without registration information, and combine them with behav...

AI Technical Summary

Problems solved by technology

In these algorithms, there are the following problems: the number of sample categories is a hyperparameter that needs to be set in advance, but it is difficult to pre-acquire the number of face image categories in actual scenes; the time complexity is high, and hyperparameters such as the number of neighbors Clustering performance has a direct impact and needs to be adjusted, so it is difficult to obtain ideal clustering speed and clustering accuracy on large-scale face datasets; the termination conditions of iterations vary from scene to scene, and setting appropriate termination conditions for the scene is often calculated large

Images