A team of researchers from the Massachusetts University of Technology (MIT) has developed a way to recover lost data on images and details on video.

The algorithm was called the “visual deprojection model”; it is based on the convolutional neural network, which is aimed at pattern recognition. The network was trained in pairs of low and high quality images so that it could check its answers.

When a model is used to process low-quality images with blurry elements, it analyzes them to find out what in the video could cause blur and what image could be there.

Then it synthesizes new images that combine data from both sharper and fuzzy parts of the video. For example, the algorithm sees the sidewalk with pedestrians and can “guess” that the walking person is blurred.

During the tests, the model was able to recreate 24 videos - in one of the cases, it even restored in detail the size and position of the human legs. Researchers are confident that this technology will help improve medical devices.

They believe that technology can be used to restore medical images, such as x-rays. So scanning can become much cheaper, which will make it affordable even for developing countries.

hightech.fm