Adobe Patents an AI System That Erases Glass Reflections From Photos
Taking a photo through a window almost always means fighting a ghost of yourself in the glass. Adobe is filing a patent for an AI system that removes those reflections automatically, using a second photo of whatever is causing the glare.
What Adobe's reflection-removal system actually does
Ever tried to photograph a storefront display, only to see your own face staring back at you in the glass? That ruined shot is what Adobe is trying to fix with this filing.
The system works by taking two images: the one you're shooting through the glass, and a separate photo aimed directly at whatever object is creating the reflection. A machine learning model then compares the two, figures out which parts of the main image are the reflection, and strips them out. You're left with a clean version of whatever was on the other side.
The output is a RAW image, the high-quality uncompressed format that professional photographers prefer, which means the cleaned-up photo doesn't lose quality in the process. The result appears directly in your editing interface, ready to use.
… separating, by the processing device, the base image from the overlay image by identifying at least one portion of the object in the image and in the additional image using a machine learning model …
Translation: The software uses artificial intelligence to tell the difference between the actual photo and the unwanted glare.
How the model separates reflection from the real scene
The patent describes a reflection removal system that takes two inputs: a primary image and a secondary image captured while pointing the camera at the object causing the reflection.
A machine learning model performs what the patent calls segmentation, meaning it identifies and separates two layers inside a single image. One layer is the base image (the scene you actually wanted to photograph), and the other is the overlay image (the reflected object layered on top). The second photo, aimed directly at the reflecting object, gives the model a reference point: it finds matching features between the reflection layer and that reference shot, then uses those matches to decide what to remove.
The system works on RAW digital images, the unprocessed files that cameras capture before any compression or color adjustments are applied. That matters because editing RAW files preserves far more detail than working on a compressed JPEG.
The cleaned image is then displayed in a user interface, so the user sees the corrected output without needing to manually clone, patch, or retouch the reflection away.
The reflection removal system generates an output RAW digital image that includes the base image and displays the output RAW digital image in a user interface.
Translation: The system then saves and displays the cleaned up photo without the distracting reflection.
What this means for everyday photography
If this makes it into Adobe's editing software, it could save photographers a significant amount of manual retouching work. Removing reflections today typically means cloning pixels by hand or using content-aware fill tools that often make mistakes. An AI that uses a dedicated reference photo of the reflecting object could be far more precise.
Adobe's steady investment in computational photography AI suggests the company sees this kind of automatic correction as a core part of its editing tools going forward. For anyone who regularly shoots through glass, whether that's street photographers, real estate photographers, or someone trying to capture a museum exhibit, a reliable automated fix would be a meaningful upgrade.
That makes this Adobe's 29th filing we've tracked since May in the AI photo editing race, a list that already includes one on filling 3D model holes with text and one on scoring video edit flow.
The two-photo requirement is the real cost buried in this design. To make it work, you have to photograph your subject through the glass and then deliberately turn around and shoot whatever is causing the reflection. That extra step means planning ahead, and most people do not think about reflections until they are already home.
That friction matters because the people most likely to benefit are the least likely to follow a two-step ritual. The system trades convenience for accuracy, and the accuracy only holds because the second photo gives the software a clear reference to work from. Without it, the whole approach likely falls apart.
For photographers who already shoot carefully and edit in professional file formats, that trade is probably acceptable. For anyone else, it asks too much of the moment.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
8 drawing sheets from US 2026/0268559 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →