What dark technologies are hidden behind the lightweight application Microsoft selfie

At a student gathering last week, when the girls enthusiastically helped themselves to take selfies, the author recommended a photography application called "Microsoft selfie" to them. At that time, this app was still in the testing stage, which won the unanimous praise of the girls and thought that the original selfie gods were weak. Now that this app has landed in Apple's App Store, you can give it a try.

When it comes to the strength of "Microsoft self portrait", people's reaction is generally "more natural". The automatic beauty function of "Microsoft Self portrait" will not make people look fake at a glance, and can well identify the hierarchical relationship between characters and background.

As a product developed by a team at the level of Microsoft Asia Pacific Research Institute, although the function of Microsoft Self portrait seems very simple, there are various "black technologies" accumulated by Microsoft for many years behind each function. Now we will analyze their technical principles while using them.

Microsoft selfie interface

　　 Black technology I: natural beauty

The first is the main natural beauty function. Open Microsoft Self portrait and choose to take a photo. When the photo is taken, it will be automatically beautified, and then you can save and share it. Is it that simple? Yes, it's that simple. However, there are many technologies involved to achieve the effect it has achieved.

The first is Microsoft's face recognition technology. Through the JDA face detection and LBF face comparison program, Microsoft can determine the age, gender, skin color, lighting and other factors of the portrait in the picture according to the face characteristics, and intelligently adjust the facial decoration of the portrait through image processing technology and intelligent information recognition technology.

The advantage of this first analysis and then targeted beautification is that it can retain the facial features of the target population to the maximum extent. For example, black people will not be converted into white people, men will also retain the masculine features they should have, and the elderly will not have no wrinkles. These features are all aesthetic data sets of all kinds of people sorted out by big data, making the facial beautification effect more natural.

Keep masculine details of men after beautifying

The second is the face tracing point location technology. The beautification of Microsoft's selfie is not like the overall beautification of most similar applications, but only beautifies the face part through face tracing point positioning. Other backgrounds, clothes, etc. will not be "skinned" together, which is P at first sight.

With such technical support, even a group photo can beautify everyone according to their age, gender and other characteristics. Can you say that this is not black technology?