AmericaScientists have developed an artificial neural network that can reconstruct images of faces with near-true color from infrared images.
The photo in visible light (left) and the image reconstructed by AI from the infrared image (right). Photo: UC Irvine Department of Ophthalmology
Humans cannot see in the dark and invented infrared cameras to do this job. Now, a team from the University of California Irvine (USA) has developed a new method to make these images more realistic. The study was published in the journal PLOS ONE on 6/4.
Some night vision systems use infrared light that is imperceptible to humans, and the recorded image is converted to a digital screen, which displays as a monochrome image in the visible spectrum.
The team sought to develop an imaging algorithm powered by deep learning structures. Accordingly, infrared spectral illumination can be used to infer an image of a scene similar to what humans see in visible light.
The key factor here is not the camera, but the algorithm the team uses to reconstruct the image. Experts have created a special form of AI called an artificial neural network – a type of deep learning algorithm that simulates how the human brain learns. They then trained this neural network to detect the association between images in infrared and visible light.
Although the reconstructed images are impressive, the team of scientists admits that this is only a “proof of concept” study (implementation of an experiment to prove the feasibility of an idea), using Use prints with limited pigments. So far, its success has been limited to the face.
“The human face is of course a very limited group of subjects. Currently, if you apply the method developed based on the face to another scene, it may not be effective and cannot be generated. whatever makes sense,” said Professor Adrian Hilton, director of the Center for Vision, Voice and Signal Processing (CVSSP) at the University of Surrey.
For example, an AI that is trained on a bowl of fruit instead of a face will be fooled by a blue banana, since its training only has yellow bananas.
Andrew Browne, the study’s lead author, cautions that these results are very stub. However, he also thinks that with further research, the new technique could become extremely precise. “I think this technology can be used for accurate color assessment if the amount and variety of data used to train the artificial neural network is large enough,” he said.
Thu Thao (According to IFL Science)
at Blogtuan.info – Source: vnexpress.net – Read the original article here
