Age Estimation And Implementation Based On Label Distribution

Face images contain many important biological features.The study of face images mainly includes face age estimation,gender judgment and emotion recognition.Such as face age estimation,it can be widely used in biometrics,personalized services and other fields.At present,the research on age estimation mainly uses classification algorithm or regression algorithm for age single label learning.They are unable to take full advantage of the correlation between age labels.So people add some surrounding age labels to describe the images.It can improve the accuracy of age estimation.In order to estimate the age images,this paper mainly studies from the following three aspects:(1)In recent years,a variety of label distribution learning algorithms have been proposed,the specialized algorithms constructed by the maximum entropy model to solve many problems with label ambiguity well,but they are extremely computational intensive.Based on this,the kernel extreme learning machine model with fast running speed and high stability is introduced,and a label distribution learning algorithm based on this model is proposed.It can reduce the time consumption while improving the prediction accuracy.(2)Today,many label distribution learning algorithms achieve high performance by selecting appropriate classifier.The existing algorithms have different classification results due to different strategies.In this paper,several classifiers are combined to focus on the method of heterogeneous ensemble algorithm to improve the accuracy of label distribution learning.A heterogeneous ensemble learning algorithm based on label distribution learning is proposed.The method of ensemble learning can improve the prediction accuracy of the algorithm.At the same time,the predicted result is a label distribution with confidence,which conforms to the label distribution paradigm.(3)Since the face images have a slight change in adjacent ages.It is difficult to generalize the age with a single label,but label distribution learning can solve this problem well.In this paper,the normal probability density function is used to distribute the age labels.The maximum probability age and the expected age are combined for joint prediction.The comparison between single label and label distribution learning shows that method of this paper can effectively reduce the age prediction error.