Face Detection The Real Story
Ok first off let's start off by saying facial detection and facial recognition are two separate things. Facial detection is just simply that, it detects if a face is present. It does not however differentiate differences between two faces or identify who that face belongs to. That is where facial recognition comes into play. Facial Recognition is what government agencies and such use to identify suspects of crimes, missing persons, and various other things. That being said facial recognition is an off branch of facial detection, how are you supposed to identify someone if you can detect the face in the first place. Facial recognition has many flaws which have been improved over the years, but that is for another post to get into. In this post, we will only cover face detection and some of the many areas that may be used in today's society.
So the first use of facial detection and recognition was back in the mid-’60s when Woody Bledsoe, Helen Chan Wolf, and Charles Bisson were working on an algorithm to detect human faces. Not too much is known of their work as it was originally sponsored by an unknown intelligence agency and the work was classified. But we do know that they tried using “landmarks” of the face like eyes and nose and the distances between them to identify faces. Also using another algorithm to determine the rotation of the face and pose to determine what the distances would be if the faces were rotated. Throughout the years the technology advanced and so did facial detection. In 2010 Facebook started using facial detection to “tag” people in the photos being uploaded to its social media platform. From that point, Facial detection rapidly became more accurate and prevalent. Apple released the iPhone X in 2017 which include FaceID which was a facial recognition software to unlock the users' phone with their face.
So how does facial detection work you ask? Let's get into that. Yan, Kriegman, and Ahuja presented a classification for face detection methods. There are 4 classifications to which they broke it down to. Facial detection can be apart of more than one classification as well. The 4 groups are as follows: Feature-based, Appearance-based, Knowledge-based, and Template matching.
- Knowledge-based: This is set upon a set of rules to follow. all rules are set by human knowledge. This classification can produce many false positives as the rules can be too general.
- Feature-based: This method as the name applies is based on structural features of the face. It has a much higher success rate as it is used as a classifier first to differentiate the structural features of a face and what is not one of a face.
- Template matching: This is having a set image/images of a face to detect as the rules for matching a face. It compares the scanned image to those in the “templates” for a match.
- Appearance-based: This section is based on AI and Deep Learning algorithms and also statistical analysis. That makes this far superior to other methods.
Now that we know the classifications let’s get into the how. First, the program would take the image and convert it to greyscale. it is much easier to detect things in the picture in greyscale. After converting they will try to enhance the picture by cropping or sharpening up the picture that way segments can be detected better. Then the actual face detection algorithm will run detecting edges and contours that may be defined as a face and if found will identify an area of where the face may be located.
In today’s applications face detection could be used in numerous applications. It helps blind people tell if the person they are talking to is smiling and can vibrate their device to let them know so they have a better understanding of how the person they are talking to is receiving the conversation they are in. Another scenario is on I'm sure you are aware of is on social media, Like tagging friends in pictures or face masks in videos. Security is also a big section that uses face detection. Whether it is at concerts or airports or your local PD, they may all use face detection for security purposes. These are just a few things that may use facial detection, who knows what may come in the future.
The picture at the top of the story comes from the show person of interest where an AI built by the guy on the right detects the probability of a crime taking place. It uses facial detection and facial recognition to detect who is “Admin” to the system and allow them to have the AI make corrective measures to city infrastructures like traffic lights to slow down the criminals. it is a good show and suggest you binge-watch it on Netflix.
Hope this provided you with a brief overview of face detection and here are some links that I used to compile this post.