AI and Facial Recognition reached the digital asset management market a few years ago. With Canto adding the free Facial Recognition feature to their cloud solution in 2019, and it has been improving ever since. Based on Amazon Rekognition technology, this technology matches faces “based on their visual geometry, including the relationship between the eyes, nose, brow, mouth, and other facial features”.
In Canto, due to privacy concerns of some organisations, your Canto account manager needs to request this feature turned on (please contact databasics). Next, your Canto admin needs to enable this feature in Settings. To set up Facial Recognition in Settings, go to Configuration Options – Recognition. Here you will have to select different options:
- Confidence Level (0-100%)
The Amazon Rekognition uses what is called “probabilistic system”, where determinations cannot be made with absolute precise accuracy, it is instead, a prediction. Thus, when setting the Confidence Level, you need to determine how important accuracy of these predictions is for your organisation. For example, law enforcement and public safety might require a 99% accuracy. For most companies a level of 80% might be acceptable.
- Maximum number of people recognised in an image
By default, maximum number of people is set to 10. You can change this number according to your preference. The higher you have set the required confidence level the fewer faces may be recognised.
Best Practices
To achieve best results, the following is recommended:
- Upload a portrait of a person, by doing that the FR feature will be able to identify and allow for a person to be tagged.
- Use images with one face only large enough in proportion to the whole image.
- If you only have images with multiple faces, make sure the face used is large enough proportionally to the image and not too close to other faces or other objects.
- Otherwise crop and upload single face images.
To better illustrate this, please take a look at the images below:
* People faces marked in red squares could not be tagged (returned the information message listed below).
* People faces marked in red circle could be tagged successfully.
* People faces marked in red squares could not be tagged (see information message below)
* Other faces already tagged in the system
*Face to small
*cropped, can be tagged
Just to clarify, this only refers to the feature allowing manual entry of names for unrecognised faces (screenshot below) and is a measure designed to ensure the quality of the reference images used for facial recognition.
It does not relate in any way to the capability of facial recognition technology in Canto or AWS.
Once you have entered a name manually the system will recognise that face in other images even if there are multiple faces, the face is smaller that 10%, on angle, next to other objects, or sharp, etc.
*once the faces have been picked out, you can tag them with appropriate names
In summary, every organisation might have slightly different needs when using facial recognition. For one, it might help identify images of individuals that have requested strict privacy protection, making sure their images don’t get shared or published. For another, it might be the timesaving need to avoid manually adding metadata.
The facial recognition technology is not always 100% accurate, however one of the advantages of this technology is that it continuously learns and improves, so false positives can be reduced over time.
A recent study concluded that even older technologies could outperform human facial recognition capabilities.
