Research Of Age Estimation Method Based On Static Facial Images

Age, as an important human identity information, has applied in many potential respects, such as security monitor, human-interactive, video retrieval. With the development of biometric recognition techniques in recent years, age estimation based on static facial images has become a crucial topic in computer vision. A growing number of scientists already put them into the field.Age estimation problem can be formulated as either a multi-class classification problem or regression problem, so this thesis briefly introduces some traditional age estimation techniques available. Since the classification approaches merely regard the age labels independent to each other but overlook the inter-relationship among the age values, they are more suitable for age group estimation. In this paper, we focus on regression methods to solve the age estimation problem.This paper summarizes several commonly used methods in age face feature extraction and regression training, and the different age estimation schemes are obtained by combining different features and regression methods, and then compared with the FG-NET face database. The result shows that the accuracy of the traditional regression method is lower because of sparse and imbalanced image samples.To solve the above problem, the paper introduces the concept of “attribute”, and proposes a method of face age estimation based on the weighted cumulative attribute. By analyzing the growth law of human beings, we find that the growth rate of the human being is different in different periods. The concept of weighted cumulative attribute is defined, which greatly reduces the influence of the sparse and imbalanced training samples. Then, the weighted cumulative attribute is used as the intermediate feature to construct the dual regression model between the AAM feature and the age label, thus the age of the human is estimated. Experimental result shows that the proposed method has a better accuracy than the traditional method, moreover, it can further reduce the age estimation error by using the weighted cumulative attribute.