Research On Face Age Estimation Algorithm Based On Video

Face age estimation is a popular research direction in the field of computer vision and pattern recognition.It has a wide range of applications in the fields of intelligent security,surveillance and image semantic understanding.Because face images are easily affected by lighting,occlusion,etc.,face age estimation is also a challenging subject.In recent years,with the popularity of deep learning,the age estimation algorithms based on deep learning surpass the traditional methods of manually extracting features in performance,and have become the mainstream methods of current age estimation.In recent years,research on age estimation is mainly for static images.Therefore,the data sets available in the age estimation field are all image data sets.However,more and more practical applications need to directly analyze and process facial related features for video data.Therefore,this article will study the key techniques for face age estimation for video.In view of the face age estimation in the video,this paper mainly completed the following aspects.Due to the lack of large-scale video face age dataset,this paper first constructs a large-scale video face age estimation database,which provides the necessary research foundation for expanding the scope of age estimation research and constructing video-oriented age estimation models.Specifically,the video is manually collected on the video website and downloaded by the crawler.The data set contains 37,184 face videos of 3,309 people.The total number of frames exceeds 10 million frames.The Videos are evenly distributed over 18 to 65 age range.Each video is manually screened and labeled with an age label,which has a high quality and can be used for the study of age estimation deep learning algorithms.Based on the constructed data set,the paper further analyzes the key issues of age estimation in video.According to the sequence characteristics of video data and the accuracy and stability of age estimation in video,this paper proposes two age estimation algorithms for video.The first algorithm is based on the idea of feature combination.Firstly,the convolutional neural network is used to extract the features of the input image sequence to obtain an age feature sequence.The attention mechanism is used to weight the feature sequences to obtain a new feature.Finally calculate the age using the fully connected layer.After using the convolutional neural network to obtain the feature sequences,the second algorithm uses the LSTM network to calculate the age on the age feature sequences.In order to optimize the model on both accuracy and stability,the paper proposes a new loss function for model training on video data.The new loss function adds a variance regular term to the MSE to constraint model stability in video.Experiments show that the proposed algorithm outperforms other image-based deep learning models in terms of accuracy and stability when predicting age in video.