- Immich's facial recognition system uses DBSCAN-based algorithms to automatically group similar faces.
- It allows advanced library management through profile merging, naming, and filtering of irrelevant faces.
- Setting detection and distance thresholds is fundamental to optimizing the accuracy of machine learning.
If you've spent your life accumulating thousands of photos from cameras, social media, or endless WhatsApp chats, you'll know that Bringing order is a real nightmareSometimes we find that our gallery is full of screenshots or receipts that are useless, and what we really want is... a system that allows us to filter out the noise and keep only the memories that matter.
To solve this chaos, Immich presents itself as a powerful tool that not only saves your files, but is also capable of Recognize faces in images and videosThanks to its machine learning engine, the software automatically groups faces so that you You just have to name them and find them in the blink of an eye from the Explore section.
How facial recognition works
The process begins with face detection, where the application sends a preview image to the Machine Learning service. If the necessary model is not installed, the system download the model automatically to process the image. After decoding and preprocessing the capture (using hardware acceleration if you have it configured), bounding boxes are generated that serve to crop the face and pass it to a facial recognition model.
The most interesting thing happens in the recognition process, which is based on a variant of the DBSCAN algorithmBasically, it treats each detected face as a point on a graph and looks for other nearby points. If the system finds a group of similar facesThey are associated with an existing person or, if there is no one, it is assessed whether that face is a "central point".
For a new person to be created, the algorithm requires by default that there be at least three similar facesIf this number is not reached, the face is put on hold or marked as an outlier, as might be the case of a stranger who happens to appear in the background of a photo, thus preventing your contact list from being filled with irrelevant people.
Configuration and adjustment options
For those who want to get the most out of the system, in the Administration menu, within the Machine Learning Settings, There are several key options. facial recognition model It can vary; although the default is the best, there are lighter options for less powerful devices.
A critical point is the minimum detection score. It's tempting to lower this value to detect more faces, but you could end up with frustrating false positivesIdeally, it shouldn't go below 0.5. On the other hand, the maximum recognition distance controls how similar two faces must be to be considered the same person; if you have twins in the family, you might be interested in this. reduce this threshold to avoid confusion.
There is also a minimum face recognition threshold. Increasing this value makes the system more robust and prevents vaguely similar faces from being grouped together, although this will cause... some people disappear from sight until there are enough photos of them to validate the group.
Library management and available actions
Once Immich has done the dirty work, you have complete control. From a person's detailed view, you can merge two or more profiles If the algorithm has made a mistake and created duplicates, you can also change each person's cover photo, add their date of birth so the system can calculate their age at the time the photo was taken, or simply hide faces you don't want to see on the Explore page.
If you have a massive library, as happens when migrate from Google Photos to Immich When importing hundreds of thousands of images from external folders, you may experience some errors where certain people cease to be visible despite having been named. In these cases, it is essential to verify that the indexing processes and the overnight reconnaissance work have been completed correctly.
Installation using Docker for a smooth deployment
To get all this up and running, the easiest way is to use Docker, which ensures that the software is isolated from other applications and won't cause dependency conflicts on your operating system. You just need to install Docker and Docker Compose, clone the official repository from GitHub, and configure the .env file with your storage path and passwords.
When executing the command docker-compose up -dThe backend, frontend, PostgreSQL database, Redis, and machine learning models are automatically started. Once inside the web interface on port 2283, you can begin to Upload your media and activate face detection to organize your personal files privately and securely.
Immich can be integrated with other external tools via its REST API. For example, you can schedule the download of specific images for processing in document security or digital signature software, using API keys generated on the server to authenticate requests and manage multimedia assets programmatically.
I am a technology enthusiast who has turned his "geek" interests into a profession. I have spent more than 10 years of my life using cutting-edge technology and tinkering with all kinds of programs out of pure curiosity. Now I have specialized in computer technology and video games. This is because for more than 5 years I have been writing for various websites on technology and video games, creating articles that seek to give you the information you need in a language that is understandable to everyone.
If you have any questions, my knowledge ranges from everything related to the Windows operating system as well as Android for mobile phones. And my commitment is to you, I am always willing to spend a few minutes and help you resolve any questions you may have in this internet world.