Apps that promise to reveal the face of your soulmate They often cite "facial recognition" and "advanced algorithms" as technical justification. These technologies exist and are impressive—but they do something quite different from what the announcement suggests. Understanding what they actually do resolves the issue.
This text explains how facial recognition works, why it is unable to identify an unknown person, and where image analysis is genuinely useful.
How facial recognition works
The process has three stages, and none of them involve prediction.
1. Detecção
The system locates where there is a face in the image. It's the square that appears on the phone's camera when you frame someone.
2. Extração de características
The detected face is converted into a list of numbers—a vector that compactly describes geometry and texture. This set of numbers is often called a... facial signature. It's not a photograph; it's a mathematical representation.
3. Comparação
The subscription is compared to other subscriptions. already stored. If the distance between two vectors is small enough, the system concludes that they are the same person.
Step 3 is what answers our question. Facial recognition is an operation of comparison against an existing bank. That's why unlocking a cell phone works: the device has saved your signature in the registration and compares it to whoever is in front of the camera. Without prior registration, there's nothing to compare it to.
Why doesn't this point to an unknown person?
For the system to indicate the person you will be interacting with, it would be necessary for that person's facial signature to exist in some database. marked as your future partner.
That label doesn't exist. Nobody collected it, nobody could collect it, and the meeting hasn't even been determined yet — it depends on the city you're going to live in, the job you're going to accept, the friend who's going to introduce you to someone.
It's important to be precise with vocabulary. Facial recognition answers "are these two images of the same person?". It doesn't answer "who matches this person?", and even less "who is this person going to meet?". These are questions of different natures, and only the first one has available data.
What the apps actually do
- They create a new face. with an image template. The result is plausible and doesn't match anyone.
- They combine traits Several photos are taken of an average face, which tends to be perceived as attractive precisely because it is average.
- They transform their own photo., altering gender or traits. The resemblance to you is interpreted as affinity.
- They display a bank image., drawn, with an analysis animation on top.
A simple test: send the same photo twice, with a few minutes' interval. If the return is different, nothing was read—it was random generation. If it's the same, it's still just the same processing on the same input.
Where is image analysis truly useful?
- Unlocking devices and authentication, comparing it against your own registration.
- Photo organization, grouping images of the same person in their album.
- Accessibility, describing aloud what appears in an image.
- Identity verification In financial services, comparing a self-portrait with the document.
- Edition — remove object, improve lighting, adjust portrait.
All these applications have something in common: they work with information that already exists.
Taking care of your image
Facial features are biometric data, and in Brazil, the General Data Protection Law classifies them as such. sensitive personal data, with stricter treatment rules. Before sending your photo to an unknown app, it's worth checking:
- If the terms grant a broad and irrevocable license over the submitted images.
- Is there an option to delete the photo from the servers after use?.
- Who is the company responsible and in which country is it located?.
- If the app asks for phone number registration for a purely visual function — that's a sign that the goal is contact, not image.
What increases the chance of meeting someone?
No predictive algorithms here: proximity, shared experiences, and the number of possible encounters. That's what research on couple formation consistently shows, and that's what dating apps offer—not prediction, but reach.
happn: dating app
AndroidAn honest profile, a photo with your face uncovered and good lighting, and a description of what you really like to do. Combined with safety precautions on the first date, this changes the outcome much more than any generated portrait.
SEE MORE:
- Why can't any app predict a face?
- How does billing work in these apps?
- Compatibility in love: what can be measured?
Where your face is already being used within the phone itself.
It's important to distinguish between real-world and imagined uses. The phone's facial unlocking feature compares the face captured at that moment with a mathematical model stored on the device itself, in a protected area of the processor, and responds only with yes or no. There is no stored photo or upload to a server, which is why this feature works without internet access.
Gallery apps do something else: they group photos by facial similarity so you can browse by person. It's grouping, not identification—the system only knows that those images contain the same face, and you're the one who assigns the name. In both cases, the reach ends within your collection and your account.
None of this resembles searching for a complete stranger in the world. That would require a database with identified photos of the entire population, something that doesn't exist publicly and whose creation would directly conflict with biometric data protection regulations.
False positive, false negative, and why the rate matters.
Every facial comparison system works with a similarity threshold. If the threshold is too strict, the system makes a mistake by rejecting legitimate people, which is called a false negative. If it is too lenient, it accepts people who shouldn't be accepted, a false positive. There is no adjustment that eliminates both at the same time, only a balance chosen by the designer.
In everyday practice, false negatives occur with a new beard, sunglasses, bright light shining on the face, or a dirty camera. False positives occur between siblings who look very similar and, more frequently, between identical twins, in which case manufacturers explicitly recommend using a password or fingerprint instead of facial recognition.
Public performance evaluations of these algorithms also show that the error rate varies according to demographic group, age, and image quality. This matters because many people treat the result as absolute truth, when it is a similarity score accompanied by a margin of error. It is a statistical measure, not an infallible identification.
Before sending your face photo, please check this.
- Stated purposeThe policy needs to specify what the image will be used for and how long it will be stored.
- For training purposes.Look for the clause that authorizes the use of the submitted material to improve the service; it's common and cannot always be refused.
- Sharing with third partiesAdvertising partners and cloud providers should be listed.
- ExclusionThere needs to be a way to delete the image and account, with a specified timeframe.
- Server and responsibleAn identifiable company, with an address and contact information, is a basic requirement.
- Camera permissionIt's better to send a selected image from your gallery than to grant permanent access to your camera.
A simple precaution greatly reduces the risk: use a photo that is not on any of your public profiles. That way, even if the database leaks, the image cannot be cross-referenced with your social media accounts.
Common mistakes people make when using facial recognition apps.
The first is sending a photo of another person. Sending a picture of a friend, an ex, or a child to a foreign service provides biometric data of someone who did not consent, and consent, in this case, is precisely what the law requires.
The second is to grant access to the entire gallery when the app requests a single photo. Both systems already allow limited image selection, and this option should be the standard in any app that isn't an editor you use every day.
The third mistake is confusing similarity with affinity. Comparison systems measure the distance between points on the face; there is no technical link between facial geometry and romantic compatibility, personality, or destiny. Any result claiming otherwise is presenting entertainment dressed as science.
Frequently asked questions about facial recognition
O desbloqueio pelo rosto é seguro?
On devices with a depth sensor, it's quite secure and withstands printing. On models that only use the front camera, the method is considered convenient, and the manufacturers themselves warn that it's less secure than a password or fingerprint for authorizing payments.
Um aplicativo pode me encontrar em fotos alheias?
Not through standard store tools. Searching for faces in public databases is a restricted, controversial service with legal limitations in several countries. Entertainment apps don't have this access, even if they claim to.
Vale desativar o agrupamento por rostos na galeria?
It's a personal choice. The feature is convenient for finding photos and can be turned off in the gallery app settings. Turning it off doesn't delete photos, it only stops grouping and removes any groups that have already been created.
Filtro de câmera é reconhecimento facial?
It uses facial detection, which is the step of finding where the face is in the image, but it doesn't identify who it is. It's the difference between knowing there's a face there and knowing whose face it is: the first is done by the device and is trivial, the second requires a basis for comparison.
