**1. Introduction**

Machine learning-based age estimation from face images is becoming more and more important because it is widely used for individual authentication [1], forensic research [2], security control [3], human–computer interaction [3] and social media [4]. Recently, there have been many studies using deep learning based on CNNs [3], such as AlexNet [5], VggNet [6], and Inception [7], with wide use for image classification and image detection. CNN-based learning, as one of the machine learning-based approaches, enables automatic and accurate feature extraction and classification for sample sets that are too large for humans to describe all cases of matching patterns. AlexNet, VggNet, and Inception have recently been used for multi-class classification, and they are widely used as the base models of CNN.

Deep expectation (DEX) [4] is an age estimation approach based on CNN models. It uses VggNet to resolve multi-class classification problems for age estimation and adopts a method to estimate the appropriate age through expectation value calculation, for which the trained results in the softmax layer are considered the probability in the corresponding class. Instead of considering the age estimation problem from the perspective of multi-class classification, this approach applies multi-task CNN by considering the age classification problem as a regression-based problem by estimating continuous variables [8].

As another approach, a binary classifier with shallow layers is applied for all classes of age instead of using a CNN model with deep layers. The final age estimation is deducted through the ranking-based comprehensive combination of all results by each binary classifier [9]. This ranking CNN is one of the existing machine learning methods using the cascaded-based combination of the results of binary classifiers.
