Chapter 1334 Human Eyes and Mosaic

Superboy is a pure technical novice who doesn't understand anything.

Although Zhou Buqi and Pei Yao are also technical laymen, as corporate executives, they still need to know certain technical theories, and they are far better than ordinary people if they are influenced by what they hear and see.

I am particularly excited about the concept of doubling the size of this image.

This is awesome enough!

No wonder Professor Tang Xiaoge swears that in the past few years, the scientific research level of his multimedia laboratory has been comparable to that of MIT and Cambridge.

This "picture enlargement" project is the core of technology!

Zhou Buqi even made a bold guess at one point.

This multimedia laboratory...

Could it be the later famous artificial intelligence laboratory SenseTime?

Artificial intelligence is a high-end and cutting-edge scientific field. Even if the Hong Kong school is very good, it cannot be played casually. Moreover, the paper scale has reached the level of MIT and Cambridge, which cannot be achieved overnight.

The moment he saw the enlarged picture, Zhou Buqi was determined.

This laboratory must be funded, supported, and invested!

This laboratory is 99.99% the future SenseTime!

The so-called picture enlargement is actually very simple. Just like Superboy's understanding, simply stretch the sides.

But when zooming in, problems arise.

As the image is enlarged, the resolution decreases.

A 720P picture becomes 360P when enlarged twice, and then becomes 180P when enlarged again, and then becomes a mosaic piece by piece.

Professor Tang’s project is to maintain the original resolution while enlarging the image.

This makes so much sense.

One of Professor Tang’s PhD students, Li Jinzhong, came on stage to explain the project in front of the projected pictures on the big screen——

"Many old photos from the past will become blurry over time or due to poor shooting equipment. Through this algorithm program, the photos can be made clearer."

"With further development, color restoration can even be performed. Old black-and-white photos from the past can be turned into color photos. Even blurry text in a photo can be recognized more clearly."

"Mr. Pei just said something about Meitu Xiuxiu. The teacher is not sure about it, but I know it. Our algorithm can be of great help to Meitu Xiuxiu when applied. Users can take a photo first, and then choose to Which point is the focus? You can also use filters to turn the picture into various styles."

Zhou Buqi was very satisfied.

Pei Yao was also very satisfied, with a smile on her face, and asked briskly: "Dr. Li, these ideas you mentioned, color restoration, and different styles of picture filters, can these be realized now?"

Dr. Li Jinzhong was a little embarrassed, "It's not possible yet. We have not conducted research in this area. However, the technical principles used are generally consistent with the application of artificial intelligence. They are similar to picture enlargement and are all deep learning image processing. "

Professor Tang next to him was more restrained and said seriously: "To achieve these points, we need deeper learning of artificial intelligence. It is not enough at the moment and requires the joint efforts of colleagues around the world."

Zhou Buqi smiled and said: "Professor Tang is free. I hope you can take your team to the capital to communicate with our CTO Dr. Shen Xiangyang. Our Ziweixing Research Institute is focusing on artificial intelligence and even founded one in Silicon Valley. With the AI ​​lab, one of the most important directions is the optimization and design of the deep learning algorithm framework."

Professor Tang was very happy, "If there is a chance, this would be the best."

Superboy had just had a brief communication with the experts from Ziweixing in private, and now he understood that this algorithm was used to improve the resolution of the picture, so he asked curiously: "I remember it was done like this in many Hollywood movies, what? "Mission Impossible", "007", "The Bourne Supremacy" and the like, it is said that the faces of people on the street were captured through satellites, and the pictures were very blurry. Then through technical processing, details were added and turned into a picture A complete face, and then connected to some system, you can find out all the information about this person."

Professor Tang laughed, "Yes, a lot."

"is that true?"

"Fake."

Superboy was a little disappointed, "Oh."

"Can you tell me more specifically?"

Zhou Buqi was reminded by Superboy's words and felt that this was very important.

Ziweixing is working on a project called Microdot Map, which targets Google Maps.

Google Maps uses commercial satellites from the United States to take pictures of the earth and transmit them to computers for data processing and integration. It then becomes a Google Map that is so sophisticated that you can even see the flowers and plants in the community. This is mainly due to the strength of technology. The precision of American commercial satellites is higher than that of domestic military satellites.

This has endangered national defense security and will be photographed in many confidential places. Americans don’t know what these secret places are, so it’s useless to take pictures of them.

But after "Google Earth" and "Google Maps" went online, they provided the "mark" function.

In other words, anyone who knows what these places are can mark them. Many Chinese people don't know why, so they mark it. Or some bad guys will secretly investigate after receiving money, and then mark it up after the investigation is clear. This is suspected of leaking secrets.

Therefore, the domestic approach is to restrict the use of related products on the one hand, and to launch similar domestically controlled products on the other. This important task related to national security is mainly undertaken by Baidu Maps and Microdot Maps.

However, what should we do if the precision of domestic commercial satellite photography is too poor?

The National Geographic Bureau and other relevant departments have organized some civilian or military helicopters to conduct aerial photography of large cities. The precision of this kind of aerial photography is much higher than that of satellites.

But the problem came again.

How could aerial photography be used in such a big country? How can there be so much funding, manpower, material resources and aircraft? (In fact, there is no solution now. We will just ban Google once and for all.)

The person who solved this problem was Zhou Buqi. He took the lead in contacting the National Geographic Bureau and Bajixing, ordered a batch of drones, and took pictures through drones.

This is a very meaningful strategic non-profit project that records the rapid development and changes of the country.

But there is another problem here.

The technology of drones is not mature enough and may not be stable enough when shooting, which will cause the pictures taken to be out of focus, blurred, and even have many detail problems due to reasons such as sunlight and exposure.

Aerial photography from a drone is definitely not as good as a person sitting in a helicopter using a professional camera with a telephoto lens.

The current research project of the Multimedia Laboratory seems to be able to solve this problem!

The multimedia laboratory, as the name suggests, is a laboratory that studies "projection", "imaging" and "picture processing". In this field, they are more professional than artificial intelligence expert Shen Xiangyang.

Through artificial intelligence algorithms, the photos taken by drones are processed and turned into clear photos, which can be almost as effective as manual aerial photography.

Because this is Ziweixing's project, Microdot Maps has a technology monopoly, and Microdot Maps can suppress Baidu Maps.

Achieve multiple things with one stone!

What's more, it can even be extended to the military field.

For example, in military reconnaissance, the blurry pictures taken can be processed to make them clearer; for example, if the foreign pictures taken by domestic satellites are too blurry, they can also be made clear.

However, this obviously still has a long way to go. Professor Tang said: "During the investigation process of the police and FBI, some blurry photos of criminal suspects are often obtained. If the photos can be processed with super-resolution, they can be If the photos become clearer and more beautiful, the crime detection rate will be greatly improved. But in fact, the United States has been doing this for decades and has not been able to do it so far."

Zhou Buqi asked: "Why?"

"The problem with algorithms," Professor Tang sighed, "we were very disappointed with past algorithms."

Zhou Buqi asked: “Can deep learning based on artificial intelligence solve it?”

Professor Tang said: "This matter is very complicated. Computers process two signals: 0 and 1. It will be very complicated to judge some human behaviors through these two signals. For example, there is a table here, which we can see at a glance. It can be judged without reaction time. But if a computer wants to identify a table from a picture, it may require millions of calculations. In fact, in the field of picture recognition in the past few decades, the processing efficiency of computers has been far higher Lower than the human eye. Even if the human eye cannot see clearly, it is even less possible for a computer to see clearly.”

Zhou Buqi seemed to understand, "Computer processing capabilities in the past were too poor, but now there are many supercomputers. Ziweixing can even perform clustered cloud computing, and the computing power has been greatly increased."

Professor Tang's eyes changed a bit when he looked at Boss Zhou, and he said happily: "Yes, the improvement of computing power is the basis for the development of artificial intelligence. However, the algorithms of the past are definitely not enough, and new ones are needed. Innovation and new academic discoveries. Let me give you an example. 1 magnified 2 times is 2; magnified 100 times is 100; magnified 1000 times is 1000. A certain number 1, how many times it is magnified, gets are all exact numbers. However, if an uncertain number ε is magnified 10 times, it is 10ε, and if it is magnified 100 times, it is 100ε. No matter how many times it is magnified, the result is an uncertain number."

Everyone present is highly educated, and even laymen can understand the logic.

Just like this portrait picture demonstrated on the big screen.

The original photo of this picture is clear, which is a certain number, so no matter how many times it is enlarged, it is based on the algorithm and is certain. After enlarging, it can give a clear image.

But if the original photo is too blurry, so blurry that even the eyes cannot see clearly, it is an uncertain number. No matter how many times it is magnified, it is still uncertain.

Zhou Buqi nodded, feeling more at ease with this.

Even if the photos taken by drones are blurred, they can still be seen clearly by the human eye. If the human eye can see clearly, it should be able to handle it.

Professor Tang went on to say: "However, with new technologies, uncertain numbers that cannot be distinguished by the human eye may not be transformed into definite numbers!"

"oh?"

Zhou Buqi's eyes lit up.

Professor Tang said: "Rely on big data."