AI and the possibilities with image interpretationPublished 29 Jun 2017
With artificial intelligence it is possible to analyze and interpret large amounts of images effectively. Through machine learning a computer can be taught to perform different tasks like finding specific people, detecting medical conditions, or helping self-driving cars stay on the road.
“AI can help enhance human capabilities. When a computer is taught to solve easier tasks automatically, humans can focus on more difficult things,” explains Truls Fretland, Data Analyst at Knowit.
Image interpretation can be used for generic tasks like recognizing emotions in a person’s face or reading text within an image. Like taking a photo of the password on a router and automatically having your phone connect to the network, or translating a Chinese menu to English with your camera.
“By providing a computer with a large amount of images both with and without faces, it learns to recognize what a face looks like without explicitly programmed rules. Instead it solves a huge mathematical equation. This way the computer learns by example instead of by rules. That is the process of machine learning – training a computer to find patterns in the provided samples, and then having it perform a task on new samples.”
There are pre-trained models available to use for generic tasks, like Google’s Cloud Vision API and Microsoft’s Computer Vision API. By using the same building blocks as these APIs, businesses can teach a computer to perform specific tasks related to their own industry.
“A radiologist at a hospital talked about the large amount of x-rays they look at every day, most of them standard cases. By using image interpretation, a computer can help determine more trivial cases, giving the doctor more time to focus on more difficult patient cases.”
A computer can be taught to interpret medical images, like determining if a bone is broken from an x-ray image, or recognizing if a birthmark is possible skin cancer.
IBM talks about how AI and hyperimaging will give us superhero vision, and calls it one of five innovations that will change our lives within five years.
“A self-driving car will be able to detect a deer on the road in heavy fog by combining visible light with different electromagnetic spectrums it can sense through sensors. The human eye would have missed the deer.”
Superhero vision will help analyze the nutrition and safety of food, and determine if pharmaceutical drugs are real or fraud. The technology is expected to be made available in our everyday lives through portable and affordable devices.
“Any type of industry that currently relies on humans interpreting what they see can already benefit from image interpretation technology. Anything from surveillance to agriculture, where AI is used together with drones to detect disease in crops and to predict harvests.”
Truls Fretland sees how image interpretation can help businesses become more effective and get more reliable results from image analysis.
“Computers don’t get tired. A person looking at 10 000 photos a day might become more inaccurate in the afternoon. The computer can analyze images around the clock. And then people can focus on more interesting tasks instead.”