Human age estimation based on facial images
Date of Issue2017-11-23
School of Electrical and Electronic Engineering
With the rapid development of computer vision, pattern recognition and biometrics, more and more attention has been paid to computer-based human facial age estimation, which is widely used when an individual's age needs to be obtained without specifically identifying other irrelevant personal information. Thus it has broad computer vision applications including security surveillance, forensics, biometrics, human-computer interaction (HCI), electronic customer information management, age-specific precision marketing (like age-based visual advertisement), entertainment and so on. There are quantities of realistic scenarios where facial age estimation can be put to good use: with a monitoring camera the juveniles can be warned or even stopped from purchasing cigarettes or banned drugs from vending machines; the underage can be prevented from entering wine bars and buying alcohol drinks; the elders can be cautioned when they want to try some high-risk rides in theme parks, like roller coasters and forest adventures; children's entry into harmful websites can also be restrained which contain violence, pornography or other restricted contents, etc. However, there is no doubt that this task is tough and challenging. The difficulties of computer-based facial age estimation are reflected in the following aspects. 1) Difference of aging process: different people have their own living environment, ethnic group, gender, lifestyle, social contact, health condition and even gene diversity, which all together determine the speed of aging. 2) Shape or texture: different forms of aging will emerge in different age levels. For example, from infancy to adolescence, the craniofacial growth (shape growth) is the main change. However, from adult period to old age, the craniofacial change decreases remarkably and skin transformation (texture change) would be the most prominent change. 3) Data insufficiency: nowadays only a very limited number of aging datasets are available, especially which can cover all the age range. 4) Disturbance: under normal conditions, almost all the people are eager for youth and tend to show their younger face, especially for females. So the final age estimation results will be largely interfered by using cosmetics, accessories and even plastic surgery, which can obtain the therapeutic or cosmetic reformation of appearance and tissue. Consequently, judging from the significance of both research and application, facial age estimation is important for the computer vision community. The goal of this thesis is to present various estimation algorithms in different perspectives to handle this challenging research topic. In the past several years, a lot of facial age estimation algorithms have been put forward, some of which are able to obtain rather satisfying performance. Among them, fundamentally, most of the traditional facial age estimation approaches are based on classification, regression, the hybrid of the two and ordinal ranking problem. However, most of the existing facial age estimation methods are originated from solving complicated mathematical optimization problems and thus consume huge quantities of time in the training process. To refrain from such algorithm complexity while maintaining a high estimation accuracy, a multi-feature extreme ordinal ranking machine (MFEORM) is proposed for facial age estimation, which combines multi-feature extraction, ordinal hyperplanes ranking and extreme learning machine's faster learning speed together for both high estimation accuracy and efficiency, in which better performance has been obtained. Experimental results clearly demonstrate that the proposed approach can sharply reduce the runtime (at most nearly one hundred times faster) while achieving comparable or better estimation performances than the state-of-the-art approaches. The inner properties of MFEORM are further explored with more superiorities. The second contribution of this thesis is driven by ordinal hyperplanes ranking, which is one of the state-of-the-art algorithms that achieve top performance in facial age estimation. By further experiments, it is found that this approach is overly dependent on the idealizations which underlie the ranking rule. This lack of robustness can result in unnecessary erroneous estimation deviations which degrade the performance. So two approaches are proposed with new ranking rules which combine the classifier accuracy and the obtained label in each binary classification substep for ranking criteria. As a result, these two approaches minimize the deviations (normal error variance) of binary classifiers. Also, the extreme learning machine is utilized, taking full advantages of its high learning speed and accuracy. Experimental results from public datasets are presented to demonstrate that the proposed algorithms reduce the mean absolute error and improve age estimation performance while reducing the runtime significantly. Thirdly, label distribution learning (LDL) is another state-of-the-art methodology in facial age estimation. LDL actually depicts the age label from a different perspective: it takes the age of each facial image instance as a label distribution with a series of age labels rather than the single chronological age label which is commonly used. Since adjacent age labels are not separated entities and they have a close relationship with each other, LDL can better take into account the inter-relationship among neighboring age labels. However, this methodology is deficient in its simple decision-making criterion: the final predicted age is only selected at the one with maximum description degree. In many cases, different age labels may have very similar description degrees. Consequently, blindly deciding the estimated age by virtue of the highest description degree would miss or neglect other valuable age labels which may contribute much to the final predicted age. In this thesis, a strategic decision-making label distribution learning algorithm (SDM-LDL) is proposed with a series of strategies specialized for different types of age label distribution. Experimental results show the superiority and validity of all the proposed strategic decision-making learning algorithms over the existing label distribution learning and other single-label learning algorithms for facial age estimation. The inner properties of SDM-LDL are further explored with more superiorities.