IEEE Future Directions
Related Articles
Macphun has announced the new release of Luminar, a digital photo editor that is trying to compete against Adobe Lightroom, a well established photo editor. It is planning to win the market leveraging on Artificial Intelligence applied to photo post processing.
What has Artificial Intelligence to do with a photo? Aren’t we supposed to capture what is out there when we take a photo? In the end we would like to see in the photo what we are seeing with our eyes.
Well, this is the point. We are not “seeing” with our eyes but with our brain! When you look at a landscape our eyes swipe over it and continually change the focus point. The data reaching the brain are a sequence of “images” with different exposure and focal point. The brain decides what is relevant and based on that decision reconstruct an image that we are perceiving. The brain is not just good in this reconstruction, it creates it “knowing” what we like and what matters to us.
How many times have you seen a gorgeous landscape, took a snapshot and felt dismayed once looking at it since it was quite different from what you saw? Actually, your eyes saw exactly what is recorded in the photo (assuming you used the right exposure and white balance) but your brain changed it and you saw the reconstruction made by your brain.
The Luminar photo Editor, as Photolemur and a few others, uses AI to do what your brain does. First to recognise objects in the image and then to render those objects in the best possible way.
A filter (applied to the image in the figure) insert Sun rays lighting objects in their path in the correct way. To do that Luminar needs to understand what kind of objects would be affected by the light (since different surfaces and textures render illumination differently).
Artificial Intelligence is taking the upper hand in digital imaging and not just in post processing. Advanced DSRL have a powerful processor that analyses the image framed by the lens “before” the picture is taken and adjust the exposure parameters before hand. These cameras no longer work on an average exposure metering, rather they “understand” what is in the image (a face, a car…) and set the exposure accordingly. High Dynamic Range, HDR, is also going to take advantage of AI, since it will no longer try to extend the range of exposure but by understanding what is in the photo and what is relevant will mix different exposure in those specific points, exactly what our brain does.
Today HDR often results in photos that look fake, because the extended exposure range is applied uniformly to the all photos, and this is not what our brain does. Our brain focusses on what matters and tweak the images on that.
Digital photography has changed the way photos are being taken. It is no longer about capturing a fair balanced image, rather to capture as many details as possible and then applying post processing to use those details in the best possible way to reconstruct an image. The shift towards AI changes the use of these details by first extracting information on what is there, then understanding the relations among the objects and the intention of the photographer. Only at this point the rendering takes place.
