Human age prediction is useful for many applications. The age information could be used as a kind of semantic knowledge for analysis and understanding of various application domains. The facial image analysis for classifying human age has a vital role in image processing, pattern recognition, computer vision, cognitive science, and Forensic science. Age Specific Human Computer Interaction (ASHCI) has enormous prospects in daily life applications. However, more research has to be done on age estimation techniques. One of the main reasons is that, aging effects on human faces present several unique characteristics which make age estimation a challenging task. In this paper, we give a thorough analysis on the problem of facial aging in context to various approaches and summary of contributions in age estimation. We offer a comparative analysis of various approaches that have been proposed for facial feature extraction and facial aging.
In the recent years, there are large number of contributions in the field of computer vision. Human face is a unique identity that has its acceptance worldwide. Human face is having a level of abstraction that needs to be unveiled by applying a set of procedures. The hidden information that we require to extract is age factor. Human face has different attributes common to all, but values change.
These attributes can vary from gender to ethnicity, which need to be addressed. Human face can be categorized in terms of texture, physical appearance, emotional state and attractiveness. With rapid advances in Computer Vision and Pattern Recognition, computer-based age estimation via face has become a particularly interesting topic recently because of the emergent real-world applications, such as Electronic Customer Relationship Management (ECRM), security control and surveillance monitoring, biometrics, and entertainment.
Age estimation by machine has been revealed as a difficult and challenging problem. Different people have different rates of aging, which is determined not only by the gene, but also by many factors, such as health condition, living style, working environment, and sociality. Aging shows different forms for different ages. From early years to teens, the greatest change is the craniofacial growth (shape change). Overall, the face size gets larger gradually during the craniofacial growth. From adulthood to old age, the most perceptible change becomes the skin aging (texture change). The shape change still continues, but less dramatically. Thus, face aging is uncontrollable and personalized. Furthermore, males and females may have different face aging patterns displayed in images due to different extent in using makeups and accessories. Many female face images may potentially show younger appearances.
The human face is one of the easiest characteristics, which can be used in biometric security system to identify a user. Face recognition technology is very popular and is used more widely because, it does not require any kind of physical contact between the users and device. Cameras scan the user’s face and match it to a database for verification. Furthermore, it is easy to install and does not require any expensive hardware. Facial recognition technology is used widely in a variety of security systems such as physical access control or computer user accounts. Time is the most negative affective factor with face recognition technology since, as the user ages, it will change over time.
The present study reveals that, at every moment, the count of missing individuals is increasing. This is a matter of concern, basically the reason for missing are abduction, crime and societal condition in which he lives. Applications that can depict age progression can be helpful for finding the missing individuals.
Indexing of face images can be done on the basis of age estimation algorithm from the given database. The most common application of this methodology can be used to retrieve the photographs by specifying the age.
Alireza[1] proposed an averaging technique, which
is
further able to improve several individual age estimation
algorithms. The averaging operation takes a linearly
weighted summation of the individual estimator outputs.
The output of the combination is 
where L is the number of estimators,
fi is the output of the ith
estimator that is referred to as the individual estimator, wi
is
the corresponding non – negative real -valued
combination weight, xp is a testing datapoint, and
fens is a
convex combination of the individual estimators that is
referred to as the ensemble estimator.
FG-NET Aging Database was used for the automatic age estimation system and Mean Absolute Error (MAE) was evaluated as the age estimation performance parameter. The ensemble age estimator improves the age estimation MAE compared to the individual age estimators up to 4.85 years.
Andreas Lanitis et. al [10] generated a statistical model of facial appearance, that was further used as the basis for generating a set of parametric description of face images. Based on the model, classifiers were generated that accept the model based representation of the given image and produce an estimation of the age for the given face image. With the given training set, based on different clusters of images, classifiers for each age group was used for age estimation. Thus, the given requirement in terms of age ranges is the most appropriate classifier which was selected to compute accurate age estimation.
X. Geng et. al [5] proposed an automatic age estimation method named AGES (AGing pattErn Subspace). The basic idea was to model the aging pattern, which is defined as the sequence of a particular individual's face images which is sorted in time order, by constructing a representative subspace. The proper aging pattern for a previously unseen face image is determined by the projection in the subspace that can reconstruct the face image with minimum reconstruction error, while the position of the face image in that aging pattern will then indicate its age.
The test mode even had lower MAE (5.81) on the FG-NET Aging Database, where the whole data set was randomly divided into 800 training images and 202 test images. In order to verify, it was tested for AGES in such mode and a lower MAE of 5.27 was achieved.
Yixiong Liang et. al [2] proposed a novel method based on Gradient Location and Orientation Histogram (GLOH) representation[3] and Multi-Task Learning (MTL) feature selection[4] along with ridge regression for global age estimation. The basic idea is to use the state-of-the-art GLOH descriptor to represent the age related local and spatial information in the face image and utilize a sparsity-enforced MTL to select the most informative GLOH bins. The selected GLOH bins can be seen as a discriminant and a compact face representation which are then fed into ridge regressors to estimate the age.
Merve Kilinc and Yusuf Sinan Akgul [15] proposed a new age estimation system that uses a number of overlapping age groups, and a classifier that combines geometric and textural facial features. The classifier scoring results are interpolated to produce the estimated age. Large number of tests were performed on geometric and textural facial features with age group classifiers. Eight different classifiers that use different facial feature vectors were tested. Some of these classifiers use textural features and some use geometric features and others use a fusion of textural and geometric features.
Xin Geng et. al [7], instead of considering each face image as an example with one label (age), their observation for each face image is as an example associated with a label distribution. The label distribution covers a number of class labels, representing the degree that each label describes the example. Through this way, in addition to the real age, one face image can also contribute to the learning of its adjacent ages. They proposed an algorithm named IIS-LLD for learning from the label distributions, which is an iterative optimization process, based on the maximum entropy model.
Xiaoyu Luo et. al [16] proposed an approach of applying Multi-Label Learning to the age features. In this approach, each facial image is treated as an example associated with the origin label, as well as its neighboring ages, which makes the data more reliable and sufficient. It comes from the observation that, when age changes slowly and smoothly, people would look quite like themselves before and after several years. The best MAE result obtained by IIS-LLD method used in [7] is 5.77. With the same feature and database, the proposed approach has MAE of 5.73.
Hewahi et. al [8] proposed a methodology based on neural networks in order to estimate human ages using face features. Due to the difficulty in estimating the exact age, they developed a system to estimate the age within certain ranges. In the first stage, the age is classified into four categories, which distinguishes the person’s oldness in terms of age. The four categories were child, young, youth, and old. In the second stage of the process, they classified each age category into two or more specific ranges.
To train and test the system, ready datasets organized in FG-NET and MORPH were used. The FG-NET contains 68 features, where each is a pair of points. The collection of features represent the mouth, nose, eye and the face surroundings as shown in Figure 1. Table 1 illustrates each point’s location on the face.
Figure 1. Points as landmarks according to FG-NET
It provides an image database containing face images that show a number of subjects at different ages. The database has been developed in an attempt to assist researchers who investigate the effects of aging on facial appearance. The FG-NET aging database is publicly available. It contains 1,002 high-resolution color or grayscale face images of 82 multiple-race subjects with large variation of lighting, pose, and expression. The age range is from 0 to 69 years with chronological aging images available for each subject as shown in Figure 2 (on average, 12 images per subject).
MORPH is the largest publicly available longitudinal face database. It is actively used in more than 30 countries. The MORPH data corpus embraces thousands of facial images of individuals across time, collected in real-world conditions (not a controlled collection). Moreover, these images are available to the public for continued research. Album 1 contains digital scans of 515 photographs which is publicly available. A subset of Album 2 is available for acedemic researchers and contains 55,134 images of 13,000 individuals collected over four years.
Table 1. Each point's location on the Face
The FERET image corpus was assembled to support government monitored testing and the evaluation of face recognition algorithms was done using standardized tests and procedures. The final corpus, presented here, consists of 14051 eight-bit grayscale images of human heads with views ranging from frontal to left and right profiles.
This face database was created by Aleix Martinez and Robert Benavente in the Computer Vision Center (CVC) at the U.A.B. It contains over 4,000 color images corresponding to 126 people's faces (70 men and 56 women). Images feature frontal view faces with different facial expressions, illumination conditions, and occlusions (sun glasses and scarf). The pictures were taken at the CVC under strict controlled conditions.
The YGA database contains 8,000 high-resolution outdoor color images of 1,600 Asian subjects, 800 females and 800 males, with ages ranging from 0 (newborn) to 93 years. Each subject has about five near frontal images at the same age and a ground truth label of his or her approximate age as an integer.
The MAE is defined as the average of the absolute errors between the
estimated age labels and the ground truth
age labels, that is, MAE =
where Ik is the
ground truth age for the kth test image,
Ik is the estimated
age, and N is the total number of test images. FG-NET
aging database and MORPH database are available
publicly. Extensive usage of these databases is done in
recent times, as it is obvious from the study shown in Table
3. Many other aging databases are available that are
public or private.
Figure 2. Example images of a subject in FG-NET
Table 2. A summary of contributions in the topic of facial aging based approach.
Table 3. Databases that were used for the task of age estimation.
Table 4. Summary of contributions in facial aging.
In this paper, we have incorporated the recent works in the area of age estimation through facial aging. Many researchers contribute their efforts in this area. Many problems exists in the current system regarding the actual age estimation that needs to be evaluated and addressed. Various approaches used in age estimation are listed in Table 2. A summary of various contributions in computer vision regarding facial aging is shown in Table 4. In general, many other works need to be carried out in age estimation for the better performance of security applications, biometric systems and business applications. There is a future scope of improving the MAE with the known algorithms implemented with different set of parameters. The scope is not limited, more unknown areas can be identified and appropriate measures can be taken in order to improve its efficiency further.