Enhance image: Google AI makes tired TV trope a reality
Often in TV in cinema, especially in the crime and espionage genres, protagonists identify perps from grainy, low resolution footage after enhancing it into an ultra high definition still, this process until recently had little grounding in reality.
Google Brain samples
However, research from Google has shown that artificial intelligence programmes are to some extent capable of restoring detail to irreparably images using neural networks.
During research, 8X8 pixel images were enlargened to the more detailed 32X32 versions, using an AI to “hallucinate” missing data by filling in the gaps with likely features while using pre-exisitng images entered into the system as reference.
Google said the system can “synthesize realistic details into images while enhancing their resolution". To test the performance of the system, a study was performed among members of the public, they were asked "Which image, would you guess, is from a camera?”
Google Brain upscaled 50% of the study's celebrity images, the processed pictures were found to fool respondents 10% of the time. The modest returns is representative of a leap in image enhancement, building upon bicubic scaling, a less intensive process that fooled no respondents during the same study.