Limits and errors of

[11:35 Wed,22.February 2023 by Rudi Schmidts]

The software engineer Will Yager has written an interesting article on PetaPixel about the limits of computational photography. When he tried to digitally share the result of a very small laser engraving, he was amazed. Because the photo he had taken with his iPhone showed a completely different image than what it was supposed to represent.

When every other attempt failed, he finally took the same picture with a Fujifilm X-T5 and the image turned out correctly. The culprit in this case was Computational Photography, which made the image worse as it saw fit. Because what they can&t reproduce correctly, modern smartphone cameras simply invent.

In principle, this also makes sense for many cases: after all, no one notices whether, for example, the high-resolution eyelash structure had the same shape "in real life" as in the portrait photo. The eye in the picture, however, always looks breathtakingly sharp.

In his further remarks, Will Yager then goes into what advantages dedicated cameras can have over a smartphone. And that is, roughly speaking, the "photographic bandwidth", i.e. the information-theoretical limit of the amount of optical data that can be stored on a photo.

This in turn is physically determined by parameters such as the diameter of the objective lens and its lens quality, as well as the sensor and sensel size. This then leads him to the Poisson distribution of the incident photons. So basically a lot of things that we

at slashCAM also ponder from time to time....

The promise of "computer photography" is to overcome these physical limitations with the help of "intelligent" algorithms. But in doing so, of course, it can also make system-related mistakes. And these can go so far that, for example, the incredibly detailed structure of a moon image never reached the surface of the sensor.

For many cases, the signal post-processing now has internal models ready that encode into the image the structures that they "expect" to see in photos. In almost every case, the camera tries to guess what should be in your picture. Whether it was actually there, on the other hand, cannot be said afterwards.

Anyone who uses cameras for documentation should therefore think twice about whether it is a good idea to use their smartphone for these purposes.

more infos at bei petapixel.com

deutsche Version dieser Seite: Computergestützte Fotografie: wenn Handykameras Bildinhalte erfinden