Anyone in possession of nothing more sophisticated than a smartphone can nowadays take virtually a limitless number of photos and videos, so getting some bad ones now and again shouldn’t be a problem that major tech players and scientific institutions would be teaming up to solve.
Yet on the face of it, that’s what’s happening now with a group of UAE, Chinese, and US scientists – from the Inception Institute of Artificial Intelligence, the Beijing Institute of Technology, and Stony Brook University – producing a paper proposing a human-aware deblurring model for images distorted by rapid movement.
According to a paper by Ziyi Shen et al, this proposed application of Artificial Intelligence (AI) will be of great use to those people taking photos of foreground objects moving at high speed relative to their background, resulting in blurry photos.
What’s of interest with this model of machine learning – a pattern-matching subset of AI – is that it aims to deblur not just any object, but human faces. The proposed model, human-aware image deblurring, saw scientist introduce what they refer to as a large-scale dataset called HIDE that contains 8,422 blurry and sharp image pairs along with “65,784 densely annotated FG (foreground) human bounding boxes.”
The latter – bounding boxes around subjects – are described as being roughly accurate, and in need of getting “refined by human annotators.”
According to the paper’s co-authors, the model allows for better results than the current state-of-the-art deblurring methods, particularly when it comes to capturing what they call semantic details pertaining to images of human faces. The reason, they said, is that the model deals with the human-related foreground, and non-human, background blurs “separately and explicitly.”
This group of scientists seems eager to tap into what has lately been a fairly busy and eventful scene. There’s great interest by some very big commercial tech players to move forward AI that allows for “cleaning up messy photos.” Among those investing in researching and developing this AI-based tech are Google and Apple, as well as China’s phone manufacturing giant Xiaomi, US chipmaker Nvidia, MIT, and Finland’s Aalto University.
The motive for these efforts on the commercial end of this story is to improve image colors, exposure, and sharpness.
If you're tired of censorship and dystopian threats against civil liberties, subscribe to Reclaim The Net.