Have you ever watched a TV Show or movie where a Crime Scene Investigation (CSI) or international spy team pull up a blurry image on a monitor (like from a satellite photo), the Team Lead says something like “Enhance it!”, and magically the computer expert presses a few buttons and the image goes from blurry to perfectly readable?
Yes, it was pure Hollywood ridiculousness. You cannot retrieve information which was blurred out previously.
Well, that technology is about to become a reality.
A team led by Chitwan Saharia has developed a machine learning algorithm called SR3 which has been trained to take extremely low-resolution pictures, and refine them to add back in what it thinks the missing details should be.
Take for example this image of an older man, where left is the original image the system was given (only 64×64 pixels), and it produced a high-resolution output including his wrinkles.
May I remind you that the image on the right is not a photo, it was created by software.
It is especially impressive if you look at the eyes.
In the original image on the left, the entire eye is made up of only about 6-8 pixels.
But the generated image on the right has perfectly formed eyes with an iris, a pupil, shadows and even a white reflection.
Or look at this image of a young lady, where the system estimated what the braids in her hair would look like.
Now, like with any machine learning algorithm, the system is only as good as the training data it is fed originally.
And of course the researchers will have selected the images which best exemplify their ideal results for the publication of the paper.
Who knows what the “worst predictions” look like.
But it could be that sometime soon, a system like this really could be used to “enhance” blurred pictures.
Who knows what that will be used for.
Latest posts by Nick Skillicorn (see all)
- Podcast S7E161: Tiffani Bova – Employee Experience leading to Customer Experience - May 19, 2022
- Spontaneous movements may kickstart your creativity - May 18, 2022
- Novelty Bias - May 17, 2022
- Winning for the wrong reasons - May 16, 2022