Age Estimation Algorithm Based on Feature Space and Adaptive Regulator

A novel age estimation algorithm is proposed to address the two aspects of category imbalance in the age dataset and the large influence of character feature information on the accuracy of age estimation. Firstly, the feature memory space module is constructed to map the image to the feature space to get the corresponding image feature information, and then the image feature information of the same age category is augmented, which makes the image feature information of different categories more balanced after the augmentation. Then by designing the adaptive regulator module, the knowledge of how the identity features affect the age estimation results is learned, so that the overall model of the algorithm can be better adapted to different character attribute features. In addition, considering the problem of focusing too much on the local and ignoring the global information caused by the deepening of neural network layers, we introduce the feature pyramid attention module to improve the ability of recognizing and characterizing the global information of facial images. Finally, experiments show that the proposed algorithm has excellent performance on both MORPH II and ChaLearn LAP 2015 datasets.