Computer vision: what it is
Quick facts
- Definition
- The area of computing that teaches machines to interpret images and video.
- Example on the face
- Face detection, reference points (landmarks) and contours.
- Difference from generative AI
- Computer vision interprets the image; generative AI creates new images.
- In Harmy
- The app detects the face on the device and shows labeled points for marking.
- Caution
- Facial data is sensitive and deserves protection under the LGPD (Brazil's data protection law).
Computer vision is the field of computing that teaches machines to interpret images and video: recognize objects, locate people, measure positions. On the face, this includes detecting where a face is, finding reference points such as the eyes, nose and mouth, and tracing contours. It is the technology behind face unlock, camera filters and, in the case of HOF, analysis and area-marking tools.
How does it differ from generative AI?
| Technology | What it does | Example |
|---|---|---|
| Computer vision | Interprets the existing image | Finding the face and the reference points |
| Generative AI | Creates a new image | Generating an "after" projection |
In practice, a product can combine the two: first it detects and locates, then it generates. The difference matters for talking to the patient and for the LGPD (Brazil's data protection law), because the data and risks involved are different.
How does Harmy use computer vision?
In the app, after the frontal photo, face detection happens on the professional's own device. The app shows labeled points over the areas that can be marked: Glabella, Temple, Tear Trough, Cheekbone, Nasolabial Fold, Mouth Corner, Lips, Chin and Jawline. The professional taps the points of the areas to treat and sets the ml for each one. The facial contours are calculated on the device and are not stored. This concept is described in on-device processing.
Limitations
- Poor photo: low light, a cropped face, large glasses or hair over the face reduce accuracy.
- Pose: a strongly tilted or profile face makes detection harder; the app works with a frontal photo.
- Human variation: the tool may get the position of a point wrong, and the professional should check what they see on the screen.
- Not a diagnosis: locating points is not assessing health or recommending a procedure.
Privacy and data
The face is personal data and, when used to identify a person, it can be biometric data, a topic covered in facial biometric data. The clinic must inform the patient how the image is used, obtain consent and protect the photo, in line with the LGPD. Patient photos should not circulate in uncontrolled channels.
Honesty about the result
Detecting the face well does not mean predicting the result of a procedure. The projection is an illustrative simulation and may contain imprecisions typical of a generated image, as in hallucination in image AI. The professional should present the image as an illustration, never as a guarantee.
In Harmy
Computer vision is the foundation of the app's marking step. Harmy does not diagnose or recommend procedures: it only locates points, and the decision is always the professional's.
Also called: image analysis, machine vision.
Frequently asked questions
Are computer vision and generative AI the same thing?
No. Computer vision interprets what is in the image, such as where the face is and its contours. Generative AI produces a new image. Harmy uses the first to detect the face and the second to generate the projection.