A system based on artificial intelligence ‘corrects’ the mistakes and elevates them to the category of reflex.
The recent arrival of the Pixel 2 XL Google has done nothing but stir more, if it is, the struggle in the world of photography among the mobile. It is already known that the bulk of the photos that are taken in the balloon are made from a smartphone and the terminal of the Mountain View he was presented with the highest award in the famous ranking DxOMark. Samsung stuck before with the Note 8 and its fabulous camera, but Apple was not far behind with their exceptional iPhone 8 Plus getting, to day of today, no one agrees in crown the king of the moving picture. But of course, we talk about mobile devices that meet or exceed comfortably a thousand euros and what about the rest?
This question was raised a couple of students of the university of Zurich who prepared your doctorate: what could be improved the quality of an image once taken? We’re not talking about the classic settings that are already equipped with automatically the major manufacturers, but for real corrections on the snapshot of the form that would make it seem to be a category very top.Nikolay Kobyshev and Andrii Ahnotov set out to investigate on the matter and with a single goal: to make the pictures taken with mobile phones were not so much the restrictions of cost and size, and to achieve a result similar to that obtained with a reflex.
This team of researchers has been achieved and note, in addition, thanks to a system based on artificial intelligence that corrects the defects of light, sharpness and shape of the photos taken with the mobile with a surprising result. “I think what makes it unique to our approach is that it is done to improve the photos simultaneously in many aspects,” he explains to THE COUNTRY, Luc Van Gool, professor at the Computer Vision Lab of the said university, and tutor of the project.
“The technique used is based on the artificial intelligence or deep learning («deep learning»)”, he argues. “The neural network compares the images taken with different types of cameras and learn how to, automatically make a photo look like same taken with another camera”. In this process, logically, the image is enhanced considerably and have provided a number of examples to do this, although it is best to try the demo online with your own photographs.
This expert explains that the ultimate goal is to achieve a quality of image similar to the one we would get for an slr camera, although limited, that yes, that would give the perspective of the mobile; that is to say, the system corrects, but cannot replace the lack of light or sharpness. The prototype provides already some spectacular results, but the best thing is that this is just the beginning. As they explain, » we may soon see it operating directly in a given mobile: “We’ve been contacted by one of the leading manufacturers of mobile and are considering starting up a development work with our team,” explains Van Gool.