新生研讨课报告

新生研讨课报告

Image processing is used to correct the aesthetic homogeneity and exaggeration caused by single image modification of beauty camera

电子科技大学 格拉斯哥学院 2017级 王菁怡 2017200601036

Abstract

Selfies have become a popular way of socializing in today’s society, and people all have the desire for beauty. Its working principle is based on image processing and filtering, which is simple and efficient, however, the popularize of the beauty camera, the problem of aesthetic homogeneity and exaggeration has arosen, to takcle this problem, the machine learning could be introduced.

Antistop

Beauty Camera, Picture Processing, Filter, Machine Learning

Investigation object

Selfies have become a popular way of socializing in today’s society, and people all have the desire for beauty. In order to make the photos posted on social media eye-catching enough, many people used to choose plastic surgery or makeup to beautify themselves. However, with the gradual improvement of the role of mobile phones and the increasing progress of image processing technology, more people tend to choose relatively convenient beauty camera or beauty software to achieve the goal of becoming beautiful.

Camera or beauty camera, by identifying the photographer faces when autodyne, by further identifying features and face, and other characteristics, size, shape, etc, to adjust, has reached to adjust the state of the skin, facial features face, or with different styles of stickers filters, to beautify the photographer facial or exaggerated effect thereby the purpose of the tool or software.

Survey Content and Research Hypothesis

It has to be said that the advent of beauty cameras and beauty software has given many people who did not like to take selfies more opportunities and intentions to share themselves, and is a more convenient way to beautify themselves for people who like to take photos and love beauty. However, with the popularity of beauty cameras, some problems have come along.

First, let’s talk about the working principle of beauty camera. The working principle of beauty camera or beauty software is actually very simple. It is a combination of face recognition technology and filtering.

The application of face recognition technology in beauty tools is based on the recognition of facial features and key points of contour, the modification and processing of specific positions of faces, and the realization of interactive entertainment functions such as special effect beauty, special effect camera and stickers. Face recognition technology is a kind of biometric recognition technology based on human face feature information. The core technologies are face detection and tracking, facial key point detection, face pixel analysis, facial expression, gender, age, race analysis, in vivo detection and verification, face recognition retrieval. he key points of its technology are:

1. After detecting the face, analyze the face and obtain 72 key points such as eye, mouth, nose and contour, etc.;

2. Store coordinates and features of key points according to face tracking, and monitor and track every expression and action in real time;

3. Capture some micro-expressions and micro-movements (blink, open mouth, head, posture and rotation Angle changes) of a person through living body detection;

4. Determine the classification (which organ) of each pixel in the face by semantic segmentation. Different colors represent different categories in the segmentation diagram;

5. Accurately identify various facial attributes through face attribute analysis, such as gender, age, attribute, race, whether to wear glasses, etc.;

6. Face recognition: 1:1 comparison &1: Nverification.

And the whitening that we often say, grind skin function more simple, it is simple to raise image brightness and undertake local blemish only, the principle that removes wrinkle dispel blain to wait for a function is relatively complex a few. The pictures displayed in real time are skinned and whitened, not the original video data. At this time, there is no algorithm related to face detection. After shooting, the images will be detected by face and processed according to face, such as slight lean face. And the original picture that saves is the original data in camera, did not pass beautiful appearance processing. Real time beauty processing requires GPU acceleration, and CPU performance cannot achieve real-time processing. With the use of filtering, the color mutation of the picture corresponds to the high-frequency component, such as moles and other wrinkles change rapidly with the change of location. The high-frequency component of this region is rich, and the digital method to filter the high-frequency component can make the picture look rounder, less spots, smooth and delicate.

Investigation Procedures and Procedures

Today’s problems stem from the rough principles of beauty cameras and software. The process of face recognition is relatively complex, but after recognizing the face, the general beauty software often simply narrates the user’s chin and enlarges his eyes after recognizing the user’s facial contour. The so-called skin color adjustment can only simply brighten the picture by adjusting the brightness and contrast. Through this method of beauty treatment out of the portrait, is often the same as the big eyes pointed chin white skin, it is difficult to exist in a different place. In the long run, the definition of beauty by camera users and people who view beauty pictures on the Internet has remained limited to a pattern of large eyes, pointed jaws and white skin, which in turn has spawned more people who want to turn themselves into this pattern of beauty. In the long run, the mass aesthetic growing in this mode will become homogenous due to the promotion of beauty cameras, and other diverse aesthetic feelings will no longer be regarded as beauty. In the United States to achieve standardization, it is difficult to form visual stimulation, people may have a nearly abnormal pursuit of it, want to transform it, also prompted the emergence of more exaggerated performance. It is because if this background that the appearance of excessive plastic surgery or retouching characters, known as “snake-like-monster” on many social media platforms today, takes place.

Because skin care effects of intensity is often adjust the user, and the intensity of limit this regulation to reduce the homogeneity exaggerated way seem unreasonable, so the best solution in alleviate this kind of phenomenon, is will be incorporated in the beauty of processing, machine learning automatically for the user to design more diversified, close to the user’s own beautification.

Machine learning is a field that enables computers to learn without explicit programming. Machine learning algorithms can be divided into two categories: supervised learning and unsupervised learning. Here is a simple introduction, supervised learning is how we hand over the computer to do something, unsupervised learning is to let the computer learn how to do a thing. Terms like reinforcement learning and recommendation systems are all part of machine learning algorithms, but the two most common types of learning algorithms are supervised earning and unsupervised learning.

Analysis and Discussion of Survey Results

Using machine learning to beautify the image, first of all we need to provide materials for machine learning, character, that is, different standards of appearance picture, through learning, let the machine can automatically according to the user’s facial features to make customized beautification, is suitable for the fine-tuning square face, square face for phoenix eye will also not only rigid users into eyes huge double-fold eyelid, make skin care no longer limited to enlarge eyes and thin face, but a targeted adjustment, so as to provide users with a variety of beautification, rather than to simply method of homogeneity beautification.

But there are also some problems with this scheme, first, the definition of beauty is inherently different from person to person, so the supply of machine learning to these as a template, beauty, how to choose, how to ensure the adjusted according to its effect to the user satisfaction, second, when on the beauty of the original features of adjustment of user, how to determine which as a reference and what will be adjusted.

Conclusions and Recommendations

The design of beauty cameras should be more reasonable and diversified. However, there are many unconsidered factors in our survey, which require us to think carefully and make rational analysis.

References

[1]Qin yongyuan, Kalman Filtering and Combined Navigation Principles, northwestern university of technology press;

[2]Chris Okasaki, Purely Functional Data Structures, Cambridge University Press;

[3]Li Hang,* Statistical Learning Method*, Tsinghua University Press.