Shulou(Shulou.com)06/02 Report--

[the article was first posted on the official account of Fengkou know, and the content comes from Fengkou Mini Program. Welcome to follow and understand ~]

Hello, hi. I'm long Peng. Share! go on!

This time, I will continue to share face image-related applications with you before the holiday. This sharing is about the age of the human face.

This sharing will include three aspects.

First of all, make a general introduction to the age of human face. In fact, the age of human face is not only a problem of numerical estimation, but also the estimation of its apparent age and true age, as well as the age of human face.

Secondly, the research methods of face age are introduced in detail. It mainly includes how to model the age of a face, and then the research methods of face age, including traditional methods and deep learning methods, are based on what kind of ideas to solve this problem.

Finally, the application of face age and its difficulties are summarized.

This is a sharing idea for us as a whole. Our sharing will be carried out according to this.

First, the age of the face. The so-called estimation of the age of the human face actually contains many meanings:

The main contents are as follows: 1) the estimation of face age, which can be divided into apparent age estimation and real age estimation.

The real age is the image of the human face we see. What is its true age? And the apparent age, that is, when we see an image of a human face, how old is it? There is a big difference between the two.

Because the age of the face is related to many factors, and it is related to his living environment and other maintenance conditions. Some female stars may be very old, in their fifties and sixties, but she is as young as the twenties and thirties that some ordinary people do not maintain, and the features of the facial images are exactly the same. Because of these factors, in fact, what we usually call the estimation of face age actually refers to the estimation of apparent age. Because of the true age, in fact, sometimes we human experts can not really judge.

2) the estimation of face age, it can also be divided into a specific age estimation and age estimation.

Because of the estimation of age, it is not a very good quantitative problem. It's actually hard to tell the difference between 29 and 30 in terms of image. Therefore, some studies are a problem of estimation of specific age groups.

For example, 11 to 15 years old, 20 years old to 30 years old; some also need to estimate a specific age problem, then the error of estimating a specific age problem must be larger than that of the estimated age group.

We can see that Zhao Liying's estimated age is 23, but in fact she is close to 30 years old. What is shown is a specific age estimate, as well as an apparent age estimate.

Let's take a look at the following picture:

It shows that it is an age estimation problem, that is to say, for such an input picture, we do not estimate its specific age, but estimate the age in which it falls.

This is the problem of estimating the age of human faces. What are the research methods for estimating the age of the human face?

First of all, let's talk about the age-related models of faces. The age of the face consists of three models:

A) Classification model. In other words, we can regard the estimation of the age of the human face as a matter of classification. Then the problem of classification can also be divided into two categories: the classification of a single age and the classification of an age group.

First of all, the classification of individual ages. Because the labeling of our ordinary people's age data will only be an integer one year old, two years old, three years old, three years old and ten years old, we can regard it as a classification problem of a continuous sequence. For example, from zero to one hundred years old, then this is a classification problem of 101 categories.

Secondly, according to the age group we marked, we can also regard it as a classification problem of an age group, for example, 0 to 5 years old is one category, and 6 to 10 years old is one category.

The classification problem is the simplest model.

B) regression model. Because the age of the face is an ordered sequence of continuous growth. Although the data we usually label do not show the age of 1.5 or 2.5 years old, in fact, people have such ages as 2.5 and 1.5 years old. So people's age is naturally a regression model.

Compared with the classification model, the regression model is more consistent with the definition of the problem, and it will be more difficult, because there is no age problem in the face regression model.

C) sorting model. The ranking model is that when we estimate the age of a face, we do not directly use a model to regression a number, or divide it into a certain category, but compare the image with a benchmark age group. By comparing with many age groups, we estimate a more appropriate age.

Why do you need to do this? Because it is inherently difficult for us to judge the true age of a person, but when we give a benchmark age, it will be more convenient and easier for us to judge whether the person's age is older or smaller than this benchmark age.

Therefore, based on the ranking model, it is actually more in line with a human mind to judge age.

These are the three models of the age of the face. These three models have been studied in detail in academic circles.

Second, based on these models, we give you an introduction from two aspects: traditional methods and deep learning methods.

The traditional features mainly include these types of features: anthropometric model, flexible model and apparent model.

The anthropometric model is mainly characterized by the outline of the human body.

The flexible model has something in common with the anthropometric model, and it can be regarded as an upgraded version of the anthropometric model. The so-called flexible model is based on the contour of the human face, it can be constantly deformed, and the flexible model is represented by ASM model. Because the shape of a person's face changes with age, the flexible model can be applied.

The traditional method is to represent the human face through these three kinds of traditional features. However, because the age of human face is a complex problem, the traditional method can not solve the estimation of face age.

With the development of deep learning and the development of big data, at present, almost all the methods for studying the age of human faces have adopted deep learning.

The idea of deep learning, as shown in the following figure:

This is a typical classification problem of in-depth learning ideas. It first inputs a picture, and then uses the face detection algorithm to detect a more appropriate angle. After correcting it, we send it into the deep learning model to learn the features. Finally, through the full connection layer, we get a representation vector.

What is shown here is the use of classification algorithms to study the age of human faces. Because the label is 0 to 100 years old, so this is a 101 category classification problem.

Based on the method of deep learning, generally speaking, big data is used to express the features implicitly.

Third, what applications can be used to estimate the age of human faces?

First, it can be used for security control. Because in many cases, such as a scene such as sharing bikes, if we can use face scanning to identify age, we can control how many people under the age of 10 cannot ride bikes.

Secondly, it can be used for face retrieval and recognition. With face age estimation, we can use age for rapid face retrieval and recognition.

There are also some difficulties in the study of face age, which mainly include two aspects:

The first aspect is the more complex pattern, because the age of human face varies greatly with the genetic living environment and human maintenance. Even for people of the same age, the image features they show may be very different.

The second aspect is that there is too much interference. At present, because beauty and other algorithms and people usually, especially women prefer to wear makeup, it will cause significant interference to the estimation of the age of the face.

Generally speaking, the age we usually estimate refers to the apparent age, that is to say, the image shows.

This is the sharing content of face age.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.