Realize face recognition "dry goods"

preface

San Francisco, California, has passed an anti-tech policy banning government departments from using facial recognition technology, according to news reports. And the prohibition is still very thorough, any police, traffic and other departments are prohibited to use.

We all know that facial recognition is very good at finding suspects, although the technology is not very mature yet. Our learning group are artificial intelligence enthusiasts, learners, face recognition is actually using image recognition technology, according to some features to recognize the face. For example, face recognition technology on mobile phones to unlock, these are the real use of our lives, and inseparable. Today, let’s share with you:

The principle of face recognition

A typical model application of deep learning is face recognition. The deep neural network is used to train thousands of pictures and obtain a model that can recognize faces. The recognition accuracy and accuracy of Huawei’s face detection model rank among the best in the industry.

Face recognition is a three-step process, starting with finding the face in the image: Face detection in cameras uses dichotomy, in which a photograph is first cut into image chunks. Then each image block will pass through the face classifier to determine whether it is a face. If a face is found, the corresponding coordinate value is returned, and the position of the face is marked with a wireframes.

The second step is to analyze the features of the face:

Feature extraction: it is to let the machine “understand”, the face part of the analysis, get the corresponding features of the face, such as five lawsuit feature points, whether to smile, whether to wear glasses and other characteristic information.

The third step is face comparison:

Face comparison: the first two steps to get the information, is used with the face database has been recorded portrait (such as ID photos, face library) in a certain way, that is, to solve the problem of “like who”.

Facial recognition labels faces, including left eye, right eye, left eyebrow, right eyebrow, nose, mouth, ears and outline.