Overview: DeepFace is a project being developed and put in use by FaceBook, DeepFace can accurately identify the true identity of an image and can spot users in the 400 Million photos uploaded to Facebook every day. “DeepFace uses a fresh approach to artificial intelligence known as deep learning.” Deep learning is noted as networks of neurons learn to recognize patterns that are in large amounts of data. Deep learning is not a new breakthrough concept, it has been floating around for decades, but our technology is finally able to handle the extensive amounts of equations going on to create the virtual neurons necessary to recognize images and or speech.
Watch Video at this link:
What is DeepFace? 'Human-Level' Face Matching, Explained https://www.youtube.com/watch?v=13FZHiXJSsE Application:
Although the concept of facial recognition is not a breakthrough, the uses for it are. Facebook states that the DeepFace technology will be used to improve user privacy to dispel privacy concerns it has had in the past. DeepFace will be able to identify the user in a photo and notify them before allowing them to be tagged in it without their permission. This is especially useful as it will allow the user to blur out their face in images that may be embarrassing or even incriminating. This will notify uses whether the person uploading the photo is a friend or a stranger but only allows users to see the identities of the people they are already friends. Google and the government have taken a special interest in this application. Google has developed its own variation of this technology called FaceNet that boasts more than 86% accuracy at identifying whether images are the same person, while DeepFace has an accuracy score of 97% using the same recognition data set called Labeled Faces in the Wild. The difference between the FaceNet and DeepFace is that FaceNet can also put a name to a face and also present a group of faces that look the most similar or distinct.
Technology:
The technology behind the DeepFace is intricate yet allows all of the necessary pieces to flow together in a manner that appears effortlessly. The technology works in 4 stages. The stages are as follows: detect, align, represent, and classify. The DeepFace technology revisits the alignment and representation technology using 3D face modeling in order to apply a piecewise affine transformation and obtain a facial representation from a nine layer deep neural network.
The technology behind this must be “on point,” because as stated by Erik Learned-Miller, facial recognition is the “single most impressive thing that the human brain can do.” Deep learning comes into play heavily in the technology behind FaceBook’s DeepFace project. DeepFace takes photos and compares faces to discover features and tie in statistics to allow a cluster to develop. After the cluster has developed, additional layers are applied to discriminate faces based on intricate details of users’ faces. References
Facebook's DeepFace shows serious facial recognition skills http://www.cbsnews.com/news/facebooks-deepface-shows-serious-facial-recognition-skills/ Facebook will soon be able to ID you in any photo http://news.sciencemag.org/social-sciences/2015/02/facebook-will-soon-be-able-id-you-any-photo Facebook can identify your face from any image with Deep Face http://www.techworm.net/2015/02/facebook-can-identify-face-image-deep-face.html Deep learning … moving beyond shallow machine learning since 2006! http://deeplearning.net/ Google: Our new system for recognizing faces is the best one ever http://fortune.com/2015/03/17/google-facenet-artificial-intelligence/ DeepFace: Closing the Gap to Human-Level Performance in Face Verification
https://research.facebook.com/publications/480567225376225/deepface-closing-the-gap-to-human-level-performance-in-face-verification/
