How Artificial intelligence can re-illuminate any portrait without the need for a depth map.

The future of photography is mainly focused on the development and evolution of image processing. Artificial intelligence is and will remain one of the key elements for these systems. And the next field in which I could participate is in the light design; thanks to a study conducted among researchers and engineers at the University of California San Diego and Google.

How Artificial intelligence can re-illuminate any portrait without the need for a depth map.
How Artificial intelligence can re-illuminate any portrait without the need for a depth map.


This artificial intelligence is inspired by the use of filters created by platforms such as Instagram and recent advances in mobile photography as the function of 'portrait light' found in iPhones. However, while Apple's algorithm changes contrast and brightness settings according to the given scene to create a 'beauty' lighting sensation under study, the algorithm of this artificial intelligence has been trained in how the position of light affects the volume of the face and therefore is able to reconstruct the image information to obtain controlled lighting.

For this study, the researchers used 22 subjects. Eighteen of them were used to train artificial intelligence. The training consisted of illuminating each item from 304 directions of light and seven different points of view. With this training, the AI learned to recognize the bony structures, the direction of light, and how this light affects and modifies the volume of the face. The remaining four subjects were used to validate the training and progress of the study. Although the results are not 100% perfect, the effect achieved so far is very close to achieving total naturalness.

For now, the process used to indicate the lighting of the scene is through an environment map, very similar to some lighting processes in 3D modeling software. Artificial intelligence makes it possible to extrapolate the background subject or maintain the same environmental conditions. This second method still produces bizarre artifacts. By just changing the lighting on the issue, a sensation of disparity of light is generated in the scene.

Being a study conducted in conjunction with Google, it is very likely that this artificial intelligence will be implemented first in the company's mobiles. However, as DIY Photography reports, there is no indication of when the window of the arrival of technology would be to the commercial world.