Cross-age face recognition method based on attention mechanism and data imbalance

Abstract

The invention discloses a cross-age face recognition method based on an attention mechanism and data imbalance. The method comprises the following steps: firstly, extracting facial photos of different ages of the same person from a cross-age face recognition data set to form a face time sequence, and processing through a self-attention mechanism to obtain a processed feature sequence; performing linear fusion on the feature sequence processed by the self-attention mechanism; decomposing the fused facial features into identity features and age features; and supervising the age estimation task and the identity estimation task by adopting the reweighted age loss and identity loss, and constraining the similarity between the age feature and the identity feature by adopting the correlation loss. According to the method, facial photos of different ages of the same identity and a self-attention mechanism are fused, the influence of the ages on the face is learned, and more robust identity features are extracted; by re-weighting the age loss, the accuracy reduction caused by age data distribution imbalance is relieved.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a cross-age face recognition scheme based on an attention mechanism and data balance, a cross-age face recognition scheme. Background technique [0002] With the development of social economy and science and technology, face recognition has been widely used in various industries and has achieved impressive performance. However, in the face of face aging with age, general face recognition schemes have exposed their defects of low robustness, and cannot accurately and stably recognize faces with large age differences. [0003] As people age, sometimes the gaps in faces between different ages are even greater than the gaps between different people. Therefore, a key challenge of cross-age face recognition is to extract age-independent identity features from faces, so as to overcome the influence of age. In general, cross-age face recognition can be divided into generative methods and...

AI Technical Summary

Problems solved by technology

Many discriminative methods simply extract identity features by removing the age factor in face features, but this method will lose identity information to a certain extent

In addition, deep learning has a huge dependence on data sets, and cross-age face data sets usually show serious uneven distribution of age labels, that is, there are more face data of young and middle-aged people, and face data of young and old people. Less, this unbalanced data distribution greatly affects the effect of cross-age face recognition

Images