Talk about face recognition

Artificial intelligence shallow talk

Artificial intelligence has been mentioned too many times this year, and the Internet has several wind gusts of finance, AI, healthcare, education.

AI has moved into a lot of areas and has been known for everything from self-driving cars to smart customer service robots to smart homes. You seem to be able to make any product, such as medical health medicine cabinets, smart speakers, driverless cars, etc., make it smarter and more interesting by the technology of artificial intelligence.

There are three categories of artificial intelligence, technical and product Angle.

In my understanding, this kind of product is artificial intelligence technology driven products, in baidu, tencent, such companies have AI department many PM every day to face a lot of information, technical documentation and some technical terms and traditional PM knowledge structure difference is very big. So there is a lot of confusion and uncertainty about the responsibilities and work of many of the product managers of AI. Every AI company that wants to do and is already working is groping.

Every one of these categories, I've compiled a sentence that I think is more accessible and valuable.

Three categories of technical angles

In many academic articles and in the field of AI, there will be a lot of references to these characters and words. As the first semester began to understand these words and things, we first got to know them. Remember the technical reserve and the position of the technical personnel in the AI field.

Cognitive AI: cognitive computing is the most popular artificial intelligence branch that is responsible for all the "human-like" interactions. Cognitive AI must be able to easily deal with complexity and ambiguity, also continuously in data mining, NLP (natural language processing) and intelligent automation learn the experience.

Machine Learning AI (Machine Learning AI) : Machine Learning is to find some "pattern" in the big data, and then without too much personal interpretation, use these models to predict the results.

Deep Learning (Deep Learning) : Deep Learning is the foundation of many modern speech and image recognition method, and compared the method of Learning from the past, with the passage of time has higher accuracy.

Three categories of product Angle

The product path of artificial intelligence can be divided into three stages or three types:

identify

Judgment in deciding

Create generated

Image recognition, speech recognition, face recognition are part of the identification. Face comparison, the living judgment belongs to the judgment decision stage, the highest order is the generation and creation stage, such as the image phonetic synthesis, the ancient poem automatic composition.

At the current stage of development, the PM felt that the product and business value of judgment and decision making were more important. Not only is there a lot of things that you can do to improve the efficiency of many things.

Artificial intelligence products are divided into images, speech, natural language, text, user portraits and so on.

An architecture diagram for an AI platform

You will find that there are many modules inside. There are also a lot of sub-business lines.

So today I'm going to do a subset of the faces in the image field. Then there's the sequel.

What is a face recognition

Appear in the fast and the furious 7, a intelligent system known as the "third eye", may at any time call the cameras to track all over the city, through the facial features rapid accurate positioning, launch the round the city. In many films, such as "the bourne identity" and "the enemy of the state", there has been a rapid identification of the use of face-recognition technology. Cool technology often makes viewers crazy.

The traditional face recognition technology is based primarily on the face recognition of visible images and has been developed for more than 30 years. But in this way, when ambient light changes, the recognition effect drops sharply. The solution to the lighting problem is a three-dimensional image of face recognition and thermal imaging face recognition, but the two technologies are far from mature and are not very effective. In addition, based on active near infrared image more light facial recognition technology rise rapidly, it can overcome the influence of the light changes, in terms of accuracy, stability, speed of doing well.

Today, facial recognition is very broad in everyday life. There are many cases of security and crime hunting.

The Japanese government, for example, has put face recognition on the safety and security systems of the 2020 summer Olympics in Tokyo.

Summary: face recognition is a biometric authentication technique based on human facial feature information.

Third, face recognition product realization path

1, face detection, find face

Face detection and analysis technology is the first step in face recognition. For any given image, a certain strategy is applied to determine whether or not it contains people.

2, face alignment, find the five official position

The main purpose is to locate the feature point in the face region, has changes in facial expression, head posture change is still able to pinpoint face the main position, such as mouth, nose, eyes, etc.

Face calibration and similarity measure, make judgment and decision

The main one is 1:1 and 1: the recognition problem of N, to confirm whether two faces are one person and a large photo gallery to identify similar faces.

Simple said that face recognition is the basis of the first to recognize the face position from one picture, people face feature points localization to step by step, and then do face check and face recognition.

Here is a case of face recognition:

Test the face in the picture and mark the border for the face. To detect the human face, may carry on the analysis of face, eye, mouth, nasal profile, etc. 72 point positioning accurately identify a variety of facial attributes, such as gender, age, facial expression and other information.