Image recognition: New roles for police, doctors and more
By Steve Xu
Next to a high school in Harbin, China, a criminal who has robbed many times before was caught by police officers soon after part of his face showed up on a camera.
On that day, he hadn’t committed a crime. But police were able to recognize him through a sophisticated camera system that uses AI.
Using image recognition technology, police are able to maintain a high efficiency on watching population movement and locating anyone’s position, which made policing easier than ever before.
As one of the most common applications of AI, image recognition is emerging since the recent technological progress. Some of the early developments of image recognition started back in 90s, while a lot of the further steps, such as research about image similarity search, were taken only after 2011.
Reasons that those progress are so late are many: the approach of neural network, a system mimicking human neuron that can improve on itself, is just developed, the corresponding processing power were not reached before, and the efficiency to collect data can only be satisfied during the age of internet and mobile internet.
Facial recognition, as an easy one to think of, is a usual example of image recognition which has an unlimited potential on a gigantic scale: many countries use them to help regulating movement of population on a national level, and millions of smartphones are trying to use it as the way to unlock. Official supports help the system to overcome the problem of data.
“Twenty-five states allow law enforcement to use driver license and ID photo,” said Rashida Richardson, the director of policy research at AI Now institute.
Big organizations such as government and tech cooperation, are using their ability to improve their recognition technology, then apply it to larger scale, and then improve it by this increase in scale.
While attracting the most attention, facial recognition does not mean the whole field of image recognition. Functions like character recognition, object identification or disease classification play an expanding role in today’s society.
Private parking places have been using human guards for years. However, the use of character recognition leads to an end of such situation: automatic license plate recognition is now used to replace human power as a more efficient and cheaper substitute in many places.
Using edge detection, characters can be easily extracted from an image by its pattern on a representative matrix with each pixel as an element. A more ambitious function of character recognition under research would be to combine with natural language processing to create the effect of “reading”.
The neural networks today give AI the ability not only to recognize human face or characters, but also complex objects.
Today’s AI is able to recognize object’s pattern after trained using neural network. When studying the pattern in the pixel matrixes of images that contain a kind of object, AI system is able to identify this kind of object’s image in the matrix form. It is being implemented today on camera of many smartphones, or algorithm of websites to categorize images. The most influential one, however, is on the self driving car.
“The car can identify obstacle and pedestrian it sees in camera,” said Brody Huval, the chief executive of Drive.ai.
The image recognition technology is now helping self driving car to process as human, which can identify the objects around it and use neural network to respond correspondingly just as a human driver would.
When AI technology looks able to solve many problems in various fields, another industry with huge demand — healthcare — is also beginning to apply image recognition.
Doctors use to identify diseases from pictures taking by the pictures taken by X-ray on patients, but image recognition, may be able to take that job away from them. Since such photos are still a kind of image, using neural networks and enough samples, A.I system is able to look for diseases in photos on themselves.
While revolutionizing many fields, image recognition technology is not yet perfect during many situations. After a new kind of font is being used on automobile license plates in China, it was seen that the AI system often fail to recognize the character correctly.
The development of image recognition exemplifies the progress of the whole A.I industries. The research on A.I can be traced back to few decades ago, but only emerges after recent years.
Using neural network with large computing power, and deep learning by inputting a large amount of data base, become a common direction in the new developments of any kind of usage of AI.
On the other hand, while many people are ignoring it, the data collecting method with unimaginable efficiency during the age of internet are the actual fuel of the neural networks.
But large scale data collection is only possible to few gigantic cooperation or government, which made their dominance on AI research seems difficult to challenge, creating some fear about privacy and other ethical issue.
Image recognition technology, as well as the overall A.I technology that it represents, is still problematic. However, that is what attract so many talents to its research: after it has proven its potential, more problems means more progress can be made.
