Chinese company Xiaomi’s AI Lab has published a new paper detailing an AI network called “DeepExposure” that improves low-quality images through machine learning. “Comparing with other methods,” the researchers explained in their paper, “our algorithm can restore most of the details and styles in original images while enhancing brightness and colors.”
DeepExposure utilizes generative adversarial networks and asynchronous adversarial learning to split low-quality images into segments called sub-images. The system computes both local and global exposures for these sub-images, then evaluates their quality before blending them with the original image.
The end result is an image with improved exposure and detail, opening the door for future solutions that may automatically enhance low-quality photos. It’s possible that Xiaomi may one day offer this technology on its smartphones, which already offer other AI capabilities.
The researchers point out in their paper, “Due to the requirement of expertise of photography, photo quality enhancement is beyond the capability of non-professional users, thus leading to the new trend of automatic techniques of image retouching.”
The study indicates that DeepExposure could also be adjusted in the future to improve image tone and contrast.
Xiaomi isn’t the only company using artificial intelligence to automatically improve images. Earlier this year, researchers with Intel and the University of Illinois Urbana-Champaign detailed a deep neural network able to brighten low-light images without reducing their quality. Similarly, a technology called Deep Image Prior was unveiled last year with the ability to recreate damaged parts of an image based on the image’s existing elements.