What are the disadvantages of face recognition?

Disadvantages of face recognition: 1. Errors may occur, affecting people’s judgment results; 2. The reliability and stability of information are weak; 3. The amount of information contained in faces is relatively small. The complexity of the changes is not enough, and the recognition is not very high; 4. The internal changes of the person themselves and the changes of the external environment will affect the stability of the face information during collection.

The operating environment of this tutorial: Windows 7 system, Dell G3 computer.

From a technical perspective, the face is the only biometric information that can be collected without the user’s active cooperation. The collection process of other biometric features, such as fingerprints, palmprints, iris, veins, and retinas, requires the user's active cooperation. That is, if the user refuses to collect, high-quality feature information cannot be obtained. From a social psychological perspective, identifying identities through faces is consistent with people’s visual recognition experience and is easily accepted by users. For example, when people collect fingerprints and irises, they worry about privacy leaks, but they do not feel violated when being captured by hundreds of surveillance cameras on the streets every day, because the human face is naturally exposed and is considered a natural feature for identifying identity. So let’s talk about the disadvantages of facial recognition technology.

Technical disadvantages of face recognition

Face recognition technology will also have errors, which will affect people's judgment results.

One disadvantage of face recognition is that the reliability and stability of the information are weak.

The amount of information contained in the human face is relatively small compared with fingerprints, iris and other biological features, and its changes are not complex enough. For example, if two people's fingerprints or irises are basically the same, it would take dozens or even hundreds of bits to completely overlap. But if it is a human face, it is enough for a dozen bits to overlap. Many faces with similarities can be found all over the world. Therefore, the recognition of human faces is not very high, and it is not that unique.

In addition, internal changes in the person themselves and changes in the external environment will affect the stability of face information during collection. Compared with previous face recognition technology, the current face recognition technology has improved, but the specific application is still not perfect. It is conservatively estimated that the accuracy of face recognition technology can reach 99%.

Technical Difficulties of Face Recognition

1. Illumination Problem

Illumination changes are the most important factor affecting the performance of face recognition. The key factor, the degree of solving this problem is related to the success or failure of the practical process of face recognition. Due to the 3D structure of the human face, the shadows cast by light will enhance or weaken the original facial features. Especially at night, facial shadows caused by insufficient light will cause a sharp drop in recognition rate, making it difficult for the system to meet practical requirements. At the same time, theory and experiment also prove that the differences caused by different illumination of the same individual are greater than the differences between different individuals under the same illumination. The lighting problem is an old problem in machine vision, especially in face recognition. Solutions to solve the lighting problem include three-dimensional image face recognition and thermal imaging face recognition. However, these two technologies are far from mature, and the recognition results are unsatisfactory.

2. Posture problem

Face recognition is mainly based on the facial representation characteristics of people. How to identify facial changes caused by posture has become one of the difficulties of this technology. The pose problem involves facial changes caused by rotation of the head around three axes in a three-dimensional vertical coordinate system, where depth rotation in two directions perpendicular to the image plane will cause partial loss of facial information. This makes the posture problem a technical problem in face recognition. There are relatively few studies on posture. At present, most face recognition algorithms mainly focus on frontal and quasi-frontal face images. When pitch or left-right face images are severe, the recognition rate of the face recognition algorithm will also be reduced. A sharp decline.

3. Expression issues

Large facial facial expression changes such as crying, laughing, and angry also affect the accuracy of facial recognition. The existing technology handles these aspects quite well. Whether it is opening the mouth or making some exaggerated expressions, the computer can correct it through three-dimensional modeling and posture and expression correction methods.

4. Occlusion problem

For face image collection under non-cooperative conditions, the occlusion problem is a very serious problem. Especially in a surveillance environment, the subjects to be monitored often wear glasses, hats and other accessories, which makes the collected face images likely to be incomplete, which affects subsequent feature extraction and recognition, and even affects the face detection algorithm. of failure.

5. Age changes

With the change of age, a person changes from a teenager to a young man to an old man. His appearance may change greatly, which will lead to a decrease in the recognition rate. decline. For different age groups, the recognition rates of face recognition algorithms are also different. The most direct example of this problem is the identification of ID card photos. In our country, the validity period of ID cards is generally 20 years. During these 20 years, everyone's appearance will inevitably change greatly, so there are also great challenges in identification. question.

6. Face similarity

There is not much difference between different individuals. The structure of all faces is similar, and even the structure and appearance of facial organs are very similar. Such characteristics are advantageous for using faces for positioning, but are disadvantageous for using faces to distinguish human individuals. Human factors such as makeup and plastic surgery aimed at imitating a certain celebrity make this problem more difficult. Especially for the issue of twins, whether the face recognition system can correctly identify them is actually a matter of debate in the academic world. Some experts believe that twins should not be distinguished by facial recognition technology at all. It cannot be accurately distinguished using facial recognition technology.

7. Dynamic recognition

In the case of non-cooperative face recognition, blurred facial images caused by movement or incorrect camera focus will seriously affect the success rate of facial recognition. This difficulty is obviously prominent in the use of security and monitoring identification such as subways, highway checkpoints, station checkpoints, supermarket anti-pickpockets, and border inspections.

8. Face anti-counterfeiting

The mainstream deception method for forging face images for recognition is to build a three-dimensional model or graft some expressions. With the improvement of face anti-counterfeiting technology, the introduction of 3D facial recognition technology, cameras and other intelligent computing vision technologies, the success rate of forged facial images for identification will be greatly reduced.

9. Lack of samples

The face recognition algorithm based on statistical learning is currently the mainstream algorithm in the field of face recognition, but the statistical learning method requires a lot of training. Since the distribution of face images in high-dimensional space is an irregular manifold distribution, the samples that can be obtained only sample a very small part of the face image space. How to solve the problem of statistical learning under small samples remains to be further studied. Research. In addition, the face image databases currently participating in training are basically images of foreigners, and there are very few face image databases of Chinese and Asians, which makes training face recognition models more difficult.

10. Image quality issues

The sources of face images may be diverse. Due to different collection equipment, the quality of the obtained face images is also different, especially for those with low resolution. How to effectively perform face recognition on face images with high noise and poor quality (such as face images taken by mobile phone cameras, images taken by remote monitoring, etc.) is a problem that requires attention. Similarly, further research is needed on the impact of high-resolution images on face recognition algorithms. Now, when we perform face recognition, we generally use face images of the same size and very close resolution, so the image quality problem can basically be solved. However, in the face of more complex problems in reality, we still need to continue to optimize the processing.

Safety risks of face recognition

In recent years, face recognition technology has become increasingly innovative and breakthrough-prone, and has been widely used in various industries. The project is obvious to all, but the current technology still cannot keep up with the rapidly changing social changes and market demands. For example, this year's new coronavirus attack caused a large number of facial products in my country to be unable to scan and identify people while wearing masks. Afterwards, major manufacturers immediately Update the algorithm, but this time also reminds us that in the face of future uncertainty, technology cannot remain static and requires continuous innovation and breakthroughs.

In addition, how to better recognize faces under different lights and angles? Issues such as how to clearly and accurately determine identity are still technical pain points that need to be solved.

A study conducted in 2012 showed that facial algorithms provided by vendor Cognitec were 5% to 10% worse at identifying African Americans than Caucasians; in 2011, Some researchers have found that facial recognition models developed in China, Japan and South Korea have difficulty distinguishing between Caucasians and East Asians. In February this year, researchers from the MIT Media Lab pointed out that facial recognition technology from Microsoft, IBM and Chinese manufacturer Megvii had an error rate of up to 7% in identifying light-skinned women, and an error rate of identifying dark-skinned men. 12%, and the false positive rate for dark-skinned women reached 35%.

There are far more examples of algorithm errors. Recent findings revealed that a system deployed by London's Metropolitan Police produced up to 49 false matches every time it was actually used. At a House Oversight Committee hearing on facial recognition technology last year, the FBI admitted that its algorithms used to identify criminal suspects had an error rate of up to 15%. In addition, an ongoing study by researchers at the University of Virginia found that two well-known research image collections - ImSitu and COCO (COCO was jointly built by Facebook, Microsoft and the startup MightyAI), have poor performance in sports, cooking and a variety of other activities. There is a clear gender bias in the descriptions (for example, shopping images are generally associated with women, while coaching images are often associated with men).

How to better recognize faces under different lights and angles? Issues such as how to clearly and accurately determine identity are still technical pain points that need to be solved.

However, even if bias is addressed and facial recognition systems can operate in a way that is fair and equitable to everyone, there is still a potential risk of failure. Like many other artificial intelligence technologies, even if bias factors are completely eliminated, facial recognition solutions usually have a certain degree of error. All tools can be used for good or evil, and the more powerful the tool itself, the more obvious the benefits or harm it may bring.

For more related knowledge, please visit theFAQcolumn!

The above is the detailed content of What are the disadvantages of face recognition?. For more information, please follow other related articles on the PHP Chinese website!