The researchers achieved image correction through the use of a new algorithm. The algorithm operates artificial neural networks simultaneously, so that the networks apply a range of different fixes to corrupted digital images. The algorithm was tested on thousands of damage digital images, some with severe degradations. The algorithm was able to repair the damage and return each image to its original state.
The application of such technology crosses the business and consumer divide, taking in everything from everyday camera snapshots to lifesaving medical scans. The types of faults digital images can develop includes blurriness, grainy noise, missing pixels and color corruption.
The developed algorithm embraces machine learning since the algorithm can be “trained”, via the neural networks, to recognize what an ideal, uncorrupted image appears as. Understanding this allows the program to tackle multiple flaws within a single image. Neural networks are modelled on the human brain., especially the way the brain learns something new through repeated tasks.
This point is taken up by development team lead Professor Matthias Zwicker, who states: “Traditionally, there have been tools that address each problem with an image separately. Each of these uses intuitive assumptions of what a good image looks like, but these assumptions have to be hand-coded into the algorithms.”
The scientist adds: “Recently, artificial neural networks have been applied to address problems one by one. But our algorithm goes a step further — it can address a wide variety of problems at the same time.”
The computer technologists have presented their algorithm together with examples of successful image correction at the 31st Conference on Neural Information Processing Systems in Long Beach, California, U.S. The conference took place during December 2017.
