Facebook is becoming more effective as it tries to determine exactly who you are trying to tag in your images. Lest you think you are being plagued by sentient computers, it is important to note that Facebook has developed a new algorithm called “DeepFace” which is designed to be more effective at determining who is who in your Facebook photos.
In testing, Facebook’s DeepFace software had 97.25 per cent accuracy. What is interesting about this is that when it comes to facial recognition, humans as a rule only have 97.53 per cent accuracy, which means that as far as accuracy goes, the DeepFace algorithm has an accuracy that is at least on par with humans, which is a staggering level of accuracy that researchers may not have originally anticipated. A member of the California Institute of Technology told New Scientist that through the software, Facebook could identify all your friends in all your Facebook photos.
Researchers from TechnologyReview.com say that DeepFace uses a 3D model for rotating faces virtually so that more angles of the person’s face can be figuratively covered. DeepFace creates a simulated neural network in order to create a numerical description of the adjusted face that was based on the original, forward facing image and the new, re-oriented image. Some four million images were drawn upon, which then meant that every single person had something on the order of at least 1,000 images for studying. The overall performance of the final software was tested against a standard data set used by researchers to benchmark face-processing software.
It is ultimately hoped that the facial recognition software will help Facebook further refine its efforts to suggest people involved in their users’ lives, and suggest people based on algorithmic matchups that occur. With the accuracy that the program is enjoying currently, it is no real surprise that the program is continuing to be further refined as it racks further successes.
There have been attempts to use DeepFace with YouTube, but one of the chief issues is that using video that is not as clear as their photographic counterparts will have an overall impact on how successful the DeepFace program will ultimately be. Generally speaking, while DeepFace has been used on most social media outlets on photos, the basic lack of clarity with YouTube videos is precisely why DeepFace may not be as helpful as one may have previously thought. In spite of this, however, there are a range of reasons why DeepFace is really interesting, not the least of which is how effective the algorithm is as it attempts to recognize who is in the image.
DeepFace is an algorithm which continues to grow and become more modern thanks to its intuitive knowledge as it works through the process of facial recognition. Certainly, the ability for the program to recognize facial recognition software will continue to “learn” while the algorithm decides how to exactly go about recognizing the faces in the range of photos that any person seems to have on their Facebook page.
One of the first considerations to the DeepFace is the distance between a person’s eyes; Facebook programmers have believed that this would be an accurate measure of how to recognize a person’s face. Regardless, DeepFace is an interesting component that can get users working on identifying the new faces on their Facebook pages. The DeepFace project and paper is due to be presented at the Computer Vision and Pattern Recognition conference in Ohio this June.