Abstract: Removing age-related features from face features to obtain pure identity features is an important means to achieve cross-age face recognition. However, the mainstream identity feature extraction methods ignore the processing of identity-age sharing features, resulting in incomplete extracted identity features. To this end, a new method of introducing identity-age sharing features was proposed, decoupling mixed face features into pure age-related features, pure identity-related features and identity-age sharing features, and then multi-dimensional coupling of pure identity-related features and identity-age sharing features to obtain complete identity features and effectively improving the accuracy of cross-age face recognition. The proposed method achieved a recognition accuracy of 97.07% on the face aging benchmark dataset Age-DB30 and 99.73% on the LFW dataset, demonstrating the effectiveness and advancedness of the proposed method.