Face Age Estimation Algorithm Based On Deep Learning

Age information is of high application value and has a wide range of applications in human-computer interaction.With the development of science and technology,the technology related to face-based age estimation has been further developed,but there are still certain shortcomings.Face images are very typical biological features in terms of scientific research;however,due to differences in lifestyle,genetics,and other aspects of an individual's life,each individual at the same age often exhibits a great variety of distinguished characteristics,thus posing a great challenge to the study of age estimation techniquesCurrent age estimation techniques consist of two main parts:feature extraction and age estimation.In terms of feature extraction,extracting facial age features by manual design was the main way of age feature extraction in the early time,which is quite difficult and normally can extract superficial features,thus resulting in that the accuracy of age estimation cannot be further improved.In recent years,with the tremendous development of deep learning in various fields,especially in the field of image recognition,there are more advanced solutions in addressing challenges of feature extraction.In terms of extracting face age features,the image,which has undergone simple pre-processing,can be directly input into the neural network by using a convolutional neural network,and after multi-layer operation,the deep features of the image can be effectively extracted.In terms of age estimation,traditional algorithms are mainly accomplished by constructing classification models and regression functions,and the correlation among different ages is difficult to use effectively and estimation accuracy is difficult to improve.In order to solve this problem,this paper introduces recurrent neural network into the age estimation.Due to its superior memory capacity,recurrent neural network can not only accept the content from other neurons,but also acquire its own information,thus forming a ring-like network structure.This enables recurrent neural networks to better handle problems regarding time-dependent inference,and it has been widely applied in many fields such as speech recognition,automatic driving,and natural language generation.This paper uses the special network structure of recurrent neural networks to learn face aging patterns in age estimation,which greatly improves the accuracy of age estimation.The research completed in this paper mainly includes the following(1)This paper conducts a deep research in literature and methods related to age estimation,and provides a summary and overview of two key components of current age estimation techniques:feature extraction and age estimation.It also thoroughly investigates the methods and principles related to feature extraction and age estimation.According to the survey,the existing problems in the current age estimation technology are found,with corresponding solutions proposed.(2)In terms of feature extraction,this paper uses the Inception network from the Google's deep learning library for the extraction of age features.When the network structure is deep,the features extracted by traditional CNN models are mainly abstract and are lack of multi-scale information on faces,which in turn affects the accuracy of age estimation.The Inception network is capable of achieving multi-scale facial feature extraction by introducing convolutional kernels at each layer to fully preserve facial age feature information.(3)In terms of age estimation,this paper uses bidirectional LSTM to fuse the local features extracted from the feature extraction network,and mine the correlation information among regional facial features,thereby performing the final age estimation.The experimental results show that the performance of the age estimation model can be effectively enhanced by exploiting the correlation information among different regional facial features.(4)To further improve the accuracy of face age estimation,this paper designs a deep neural network model for age estimation of face images based on the theoretical knowledge related to deep learning,as well as the results of existing studies.The deep neural network model in this paper is a result of the Inception module from the Google's deep learning library and a bidirectional LSTM in combination with a label distribution.The model first extracts regional multi-scale features of faces using the Inception network,and then concatenates the features extracted into the bidirectional LSTM in combination with label distributions to learn face aging patterns and complete age estimation.Finally,in this paper,the age estimation model was experimentally verified through the FG-NET and MORPH II databases.The experimental results show that the method in this paper is effective in improving the accuracy and robustness of age estimation.