How to show the United States - Interpretation of the behind the Apparel APP

I believe many of my friends have used APPs such as Mito Xiuxiu. They can use the mobile phone to instantly photograph the photos taken by the mobile phone, such as skinning and whitening, removing freckle and acne, and many live broadcasters are also passing similar apps. Real-time cosmetology, so that they can show their face before us. Today let us interpret how these APPs implement beauty and the techniques behind beauty.

Not new technology, talk about image beauty

When it comes to image beauty, you have to mention beauty salon boss PhotoShop. Many of my friends know that with the help of PS, it is possible to turn rot into something magical and the character P to be beautiful. However, PS operation is cumbersome, and with the popularity of mobile devices, more and more similar APP APP gradually became popular on mobile phones. For example, Mito Xiuxiu, Variety Magic Chart, etc., some mobile phones will be beauty functions integrated into the mobile phone system, such as Mito mobile phone can access the camera by calling the system API, and then capture the screen through the beauty treatment (Figure 1) .

Figure 1 Mito M6 mobile phone with beauty features

With the rise of live webcasts, APP beauty technology is constantly evolving. It is no longer limited to the beauty of still photos. More and more live platforms can use APP to perform real-time grooming of live images of anchors, such as changing the eyes of anchors. Big, turning a round face into a melon face almost achieves the "cosmetic" effect in real life. So how do these beauty apps achieve these amazing beauty effects?

Behind the Beauty - Understanding APP Beauty Principles

For the average user, it is very simple to perform beauty operations on the mobile phone. For example, if you want to select a corresponding menu on the function panel, then you can simply click and select to achieve rapid beauty (Figure 2). .

Figure 2 The beauty operation of Mito Xiuxiu is very simple

But behind this seemingly simple operation is a powerful image processing technology. Take a simple anti-acne operation as an example. We all know that many of my friends have acne on their faces, so that the pictures taken will have a gray value change on your face. If there is a large difference between the gray values ​​of the two adjacent areas on a photo, noise will be formed on the photos. Therefore, the acne operation is actually the "noise reduction" commonly used in image processing. Noise reduction is mainly achieved through a certain algorithm, such as filtering algorithm, which can replace the original value with the average around the point, making the noise on the picture not so obvious, and the reaction is in the photo that the face of acne disappears. It looks like the skin is smoother.

Therefore, when we easily select the anti-acne operation in the beauty app, the mobile phone APP actually performs the noise determination in the background (by comparing the brightness of one point with the brightness of the surrounding points to find the noise), and then calls a certain algorithm. Perform noise reduction, and finally mix the processed image with the original image (because it is very easy to lose detail if you just use the processed image, making the processed image appear distorted), and by adjusting the mixed weight of the two images. This makes the processed photos look both cosmetic and not too “fake”, which is also a common flow for general beauty app processing pictures (Figure 3).

Figure 3 APP to improve the picture flow chart solution

The dynamic beauty of network anchors is even more complex. It must first realize "face recognition." APP can only conduct beauty by accurately identifying faces. The core technology of face recognition is deep learning technology. This technology uses the Internet's image data for learning and training, grabs the core features, and generates an algorithm. This way, Beauty App can use these algorithms to achieve dynamic beauty of the anchors. For example, for the anchor skin tone adjustment, the deep learning technology first compares the original image with the skinned image, and then analyzes the differences so that after a large amount of learning, it can be used after the current anchor image is captured. The specific algorithm quickly completes the treatment of the skin color of the anchors, making the effect of beauty more realistic and effective (Figure 4).

Figure 4 Deep Learning Technology Beauty Diagram

The same beauty treatments for face-lifting and small-eye enlargement are similarly applied to beautify. In this way, during live broadcasting, from the camera's acquisition of each frame of the picture, deep learning technology will perform face recognition, then mark the position of the key points, and then combine the image technology to obtain the final cosmetic effect. For a simple example, such as the popular online red face, the face recognition technology through deep learning can be used to compare the red face after grabbing the current anchor face data, and then optimize with a specific algorithm, and the anchor can be facelifted. As the favorite face and skin color for netizens, we saw all sorts of sisters and handsome guys on the live broadcast platform. In fact, it was the result of the facial recognition technology of the background.

Written in the final: beauty app, more worth the wait

With the popularity of webcasting, beauty technologies based on face recognition are now evolving. On the one hand, as hardware vendors gradually implant various deep learning technologies into their devices, starting from iOS 9, Apple has provided deep learning APIs. In iOS 10, related APIs have been further updated. After the iPhone 7, the development of deep learning has gradually matured. Google released the Facenet network as early as 2015, and this network can achieve a recognition rate of more than 98% for human faces. On the other hand, third-party face recognition technology based on mobile platforms is also evolving, such as Caffe (Convolutional Neural Network Framework), a clear, highly readable and fast deep learning framework that can make mobile devices more powerful. Deep learning ability.

People's pursuit of beauty is endless, and everyone wants to have a more beautiful face in front of the public. Due to the continuous improvement of these technologies and the continuous improvement of the hardware capabilities of mobile devices, in the near future, whether it is the photos taken by mobile phones or all kinds of webcasts (including video chat), we can all appear in people's eyes with a more beautiful image. before.