Communications of the ACM
Deep Learning Technique Reveals 'Invisible' Objects in the Dark
Engineers at the Massachusetts Institute of Technology trained a computer to reconstruct transparent objects from images captured in almost total darkness.
Credit: George Barbastathis et al.
Massachusetts Institute of Technology (MIT) engineers have used a deep neural network to train a computer to reconstruct transparent objects from images captured in almost total darkness by associating certain inputs with specific outputs.
The researchers trained a computer to identify more than 10,000 integrated circuit etchings, using a "phase spatial light modulator" that displayed the pattern on a single glass slide in a manner that reproduces the same optical effect that an actual etched slide would have.
The computer was fed these images, as well as corresponding patterns, or "ground-truths," then shown a new grainy image it was never trained on.
The system learned to rebuild the transparent object obscured by the darkness.
MIT's Alexandre Goy said, "This result is of practical importance for medical imaging to lower the exposure of the patient to harmful radiation, and for astronomical imaging."
From MIT News
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found