Research On The Synthesis Of Face Aging Combined With Age Estimation

Face attributes are the most direct and convenient means of identity verification.Each person's appearance will change to a certain extent with age,which will affect the result of face recognition.Therefore,the large amount of age information contained in the face is important for identifying individuals.in accordance with.Age estimation and facial aging synthesis research related to face image analysis are current hot issues,and they can be applied in many important fields,such as criminal investigation,digital entertainment,and so on.With the development of society,various age-based human-computer interaction information systems continue to improve and upgrade,and there are still some unsolved problems in the continuous progress of age estimation and facial aging synthesis technology.Due to the lack of age label data of the same subject covering a long time range,and the need to ensure the aging effect and identity information of the generated aging face at the same time,there are many challenges in the synthesis of face aging.Face aging will cause changes in facial features.The age estimation of the face depends on the facial features.Therefore,the age estimation is closely related to the aging of the face.Due to the complexity of face aging and the incompleteness of the existing database,it is even more difficult to judge age by age estimation methods.In this paper,we have conducted an in-depth study of face aging synthesis based on the discussion of related issues of age estimation,and completed the following work:(1)In terms of age estimation:the manifestations of face aging vary from person to person,and it is difficult to achieve the complex and diversified tendency of face aging only by relying on a single age estimation model.In practical applications,the collected data sets have unbalanced category distribution.Categories with a large number of samples will have a critical impact on the overall loss value in the process of facial age estimation,and the cost of misclassification between different categories is Not equal.In view of the above two major problems of unbalanced category distribution and the unequal misclassification cost,the loss function is improved: the loss function of each age category is established to solve the imbalanced category distribution problem;the problem of unequal misclassification cost is fully considered,The cost matrix is used to indicate that the cost of each category misclassification has a significant difference.In terms of neural network design,an age estimation model based on cost-sensitive and lightweight convolutional neural network is constructed.This paper adopts an age estimation algorithm based on cost-sensitive lightweight convolutional neural network,which solves the problem of sample imbalance and misclassification cost while making the network structure lighter,thereby improving the accuracy and efficiency of age estimation.(2)In terms of face aging:The human face aging process has inherent complexity,that is,different aging conditions are presented based on individual differences,but the synthesis of face aging needs to consider both the authenticity of the aging effect and the consistency of identity information.The authenticity of the aging effect is based on the accurate expression of aging characteristics and the clarity of the depiction of aging details.The consistency of identity information means that during the aging process,the aging face needs to maintain a certain similarity with the original young face in terms of feature information.Based on the above two issues and the instability of GAN training,a face aging method based on the hybrid domain attention mechanism to generate an anti-network is proposed.A large number of experiments show that this method can synthesize aging people under the requirements of ensuring the aging effect and identity information.(3)Combination of age estimation and face aging: The age estimation algorithm is applied to the image preprocessing stage of face aging.The experimental results show that compared with the previous results that have not been processed by the age estimation algorithm,the aging accuracy is an indicator.With some improvements,it once again illustrates the effectiveness of the age estimation algorithm.