Don’t like the pictures you snapped with your smartphone? Researchers have developed an AI program that can enhance them into photos with DSLR-like quality.
The program is available online and designed to add the colors, brightness, and textures that photos taken by DSLR cameras tend to possess over images taken with smartphones.
That may not sound like much. Photo enhancement software has been around for years, and usually comes bundled with a smartphone’s operating system. But even those tools can fail to improve image sharpness, texture details, or minor color variations in a picture, according to researchers at the Computer Vision Lab at ETH Zurich in Switzerland.
The researchers, however, came up with a novel solution. Why not train an AI to look at what makes DSLR-taken photos special?
The solution was a program based on an artificial neural network that studied the differences between smartphone-taken pictures and those snapped by DSLR cameras.
Researchers created a rig that took thousands of photos simultaneously using three different smartphones and a DSLR camera. By analyzing those photos, the computer program then trained itself to enhance smartphone-taken images into photos with DSLR-like quality.
“The network automatically learns what makes good pictures look good, and tries to improve the images so that they match these criteria,” the researchers said in an emailed statement.
The researchers have uploaded dozens of example photos, showing darker images that look brighter and more vibrant thanks to the software. However, it isn’t perfect. The program may sometimes add artificial colors to an object, such as turning asphalt into a greenish color, or introducing visual distortions into the picture, the researchers said in their paper.
The software is designed to work with photos taken by an iPhone 6, iPhone 7, or Android phones, but it can improve any digital images that might need a makeover. However, the researchers said the website for their photo enhancer is currently experiencing huge demand, causing slowdowns.
