Face recognition on its own is already a powerful technology, but through the integration with AI (Artificial Intelligence), it can reach a new level of automation and efficiency.
AI in face recognition begins with a tagged feature set. In other words, a group of photos and videos with existing, hand-matched correlations to the people/face involved. An initial, manual correlation between a person's face and the rest of their identity is necessary to give the AI a foundation to stand on. Once started, the AI can learn the patterns and steadily become better at identifying faces in images and videos.
The AI can recognize faces through what we call face recognition data. A person's face is broken up into numerous data points such as:
- the distance between their eyes
- the height of their cheekbones
- the distance between the eyes and the mouth
- among others.
The AI uses these face recognition data points and tries to account for potential variations (for example, distance from the camera or slight angle variations of the face) to identify the face in an image or video.
The challenge of training AI
A company often possesses thousands upon thousands of digital and brand assets in the shape of images and videos. Organizing, tracking, and managing them is far from easy, especially if the company mainly works locally − a local news agency is a prime example of such a company.
There is no shortage of AI pre-trained in different datasets available. However, how useful do you think these datasets are for the local news agency that mainly deals in local stories and events? Exactly, next to useless. Their local celebrities or politicians will not be found in any existing dataset. Instead, the AI must be taught from scratch to match their local requirements.
To solve this challenge, we partnered up with DeepVA to create a system that makes deployment and training of AI in a media asset management system intuitive and straightforward - even for companies with little to no experience in data science.