Face clustering method based on graph convolutional neural network

Abstract

The invention belongs to the technical field of face clustering, and particularly relates to a face clustering method based on a graph convolutional neural network. According to the method, firstly, feature extraction is performed on face data, face features are regarded as nodes, and local density values of all the nodes are calculated; then, the data is divided into high-density nodes and low-density nodes based on the local density value, and the high-density nodes are connected with the nodes which are high in density in the nearest neighbor to form a plurality of clustering centers; constructing a self-adaptive sub-graph for the low-density nodes, taking the self-adaptive sub-graph as the input of a graph convolutional neural network, and predicting the connectivity between the nodes; and finally, after the two parts are combined, edges which do not meet requirements are cut off by using pseudo-label propagation, and a final clustering result is obtained. According to the method, the data is divided into two parts based on the density, only the sub-graph is constructed for the low-density part for reasoning, the clustering efficiency is improved, meanwhile, the adaptive sub-graph can extract richer context information, reasoning on the sub-graph is more accurate, and the clustering accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of face clustering, and in particular relates to a face clustering method based on a graph convolutional neural network. Background technique [0002] With the development of science and technology, face recognition technology has gradually become popular and widely used in aerospace, entertainment, education, security and other fields, and the performance of face recognition technology has gradually improved. However, the current high accuracy of face recognition technology The efficiency depends on large-scale face annotation data sets. The more annotated face data sets, the better the effect. Although unlabeled data can be easily obtained from the Internet, manual annotation is not only time-consuming, but also expensive. In order to take advantage of these unlabeled data, face clustering methods are a good choice. [0003] At present, face clustering and technology are widely used. Some common scenarios...

AI Technical Summary

Problems solved by technology

However, the link prediction method also has the problem of high time complexity, especially in the face of large-scale face data sets

The link prediction method regards each face image as a node after extracting features, constructs a subgraph for each node as input, and uses a graph convolutional neural network to infer the subgraph to judge the connectivity between nodes, thereby obtaining clustering , however, the constructed subgraphs are highly overlapping, which not only causes redundant computing resource consumption, but also affects the efficiency of clustering

Images