The intense-zoom moon shot on Samsung’s latest Galaxy S23 ultra was faked, in keeping with a Reddit user named ibreakphotos.

This generated numerous attention.

But how did he discover?

The claim was that an AI could discover an identical photo to that of the moon after which process it with pre-trained images of the moon. To check this he took a stock image of the moon, reduced the standard to 170 x 170 pixels and applied gaussian blur to it.

Which means that it’s not recoverable because it is digitally blurred and there may be very less information within the image.

He then took the photo of the blurred image together with his Galaxy S23 Ultra on a monitor and this was the result.

It’s introducing information that wasn’t there. It’s not deconvolution, sharpening, and even “multiple frames or exposures,” neither is it anything like super resolution. It’s generating data.

Here’s a comparison of the digitally blurred image and captured image.

Due to this fact, it was evident that some processing was going down and that the optics weren’t those working their magic. The user later discovered that a texture of the moon was being added to photographs that resembled the moon primarily due to Samsung’s scene optimizer.

Let’s put ethics aside and focus on how this was achieved.
Samsung has been working on this for some time, and a Korean blog detailed their process for doing so.

How was this achieved?

With the intention to help users take higher photos, Samsung has incorporated AI technology into its Galaxy S10 and all newer models. AI is utilized by the Scene Optimizer feature to discover the topic being photographed and deliver the most effective results. From the Galaxy S21, the AI may even discover the moon because the goal and use deep learning-based technology and multi-frame synthesis to boost details and create a clearer image.

Of their blog, Samsung makes it clear that this was only a feature and that it could possibly be disabled by the user.

Samsung was attempting to boost the moon image by synthesising greater than 10 images under conditions of high magnification, which allowed it to take low resolution images and switch them into high resolution images. They were calling it .

While they might achieve higher results with just the Super Resolution technology, they paired it with a deep learning model to enhance the standard drastically.

As you’ll be able to see optimal shooting was making it significantly better. The model they were using was a Convolutional neural network, the most effective for images.

Samsung’s moon recognition engine uses an AI deep learning model that was created by learning various moon shapes from full moon to crescent moon based on real-life images that individuals actually see with their eyes on Earth. The engine can detect the presence of the moon in a picture and create a square box around it. The AI model has been trained to detect lunar regions even when presented with lunar images that weren’t used for learning.

If the moon was covered by clouds or other objects, non-lunar objects it did not detect it.

Moon photography is just possible at zoom levels of 25x or more since it is difficult to detect small moons in AI moon photography.

Once the moon has been identified using a high magnification zoom, the screen brightness is regulated to dim in order that the moon will be seen clearly on the camera screen, and the perfect brightness is maintained.

The main focus is modified to infinity when the moon is found to maintain it in focus.

How Does the phone detect the moon?

Before applying the moon detail improvement engine with AI technology, they first run the to reconfirm.

The image is fed right into a convolution neural network, which produces a binary output indicating whether or not it’s a moon.

If the user is attempting to capture a moon, several pictures are taken and combined into a transparent, noise-free image of the moon using multi-frame processing.

As a result of the vast distance from the moon and the dearth of sunshine, the moon shooting environment has physical restrictions, making the high-magnification actual image output from the sensor insufficient to offer the most effective quality experience even after composite quite a few shots.

To get around this, the Galaxy Camera uses a deep learning-based AI detail improvement engine () on the very end of the method to efficiently reduce noise and optimize the small print of the moon, leading to an entire and vivid image of the moon.

the model takes in low resolution image of the moon, then it uses Detail enhancement engine (Convolutional Neural Network) to supply a improved image which is then compared with a reference image from learning data and might update accordingly.

And VOILA there may be your AI enhanced moon image.

To conclude image processing(Computational photography) is getting used by every other smartphone, It’s not all the time the lens that gets the magic. The explanation why Samsung is facing this case is due their deceptive marketing. While some people mistakenly imagine that Samsung can add a texture of the moon to your moon photos using ML, this just isn’t a feature of your camera or lens to be specific.

Due to this fact, there is no such thing as a sharpening or augmentation of detail from additional frames on this experiment since the level of detail in each picture is identical. The true issue was with it’s promotion.

Samsung responded to all of this chaos by stating that it’s not only a overlay.

Thanks for reading!

listed below are the sources:

1. Samsung’s official Korean blog — here
2. Official reddit post — here
3. Indian Express article — here
4. The Verge article — here
5. MrWhoseTheBoss(YouTube) — here