Scientists from the University of Maryland have developed a new algorithm based on artificial neural networks, allowing a wide range of corrections to be applied to damaged digital images. Researchers tested their algorithm by taking high-quality intact images and intentionally damaging them, and then used a damage restoration algorithm. In many cases, the algorithm outperformed competitive methods, virtually returning images to their original state.

Digital images from camera phone shots to those used in medical examinations play an important role in the transmission of information, but these images are subject to a number of drawbacks including blurriness, granular noise, missing pixels and incorrect color rendering, aiportal.ru reports. A new algorithm developed by experts can be "trained" on how the image should look ideally, so it can simultaneously eliminate several errors in one image.

The specialists "trained" their algorithm with the help of a large database consisting of high-quality intact images widely used for research with artificial neural networks. Because the algorithm can take a large amount of data and identify complex parameters that define images, including changes in texture, color, light, shadows and edges, it is able to predict what an ideal intact image should look like. Thanks to this, the algorithm can recognize deviations from these ideal parameters and correct them, eliminating flaws.