https://mp.weixin.qq.com/s/y_uCmLzAuqOhxUa8ksfobw
By 超神经
场景描述：选礼物是千古大难题之一，母亲节来临之际，借助机器学习、大数据分析、图像增强等 AI 技术，可以帮你为母亲送上一份走心的礼物。
关键词：母亲节 机器学习 大数据 图像增强 神经网络 There are only two days left before Mother's Day. Have you chosen a gift for your mother?
It is really a tangled problem to make the gift beautiful, practical and attractive to mothers, and it is even more daunting for programmers who are usually in the ocean of code.
Programmers who are unable to extricate themselves in the ocean of code
According to unreliable intelligence, on the occasion of Mother's Day, programmers behave differently:
The first type: what? Mother's Day is coming soon?
The second type: Ao, Mother's Day is coming soon, please call my mom and ask her what gift she wants.
The third type: open an e-commerce app, search for "Mother's Day Gifts", and choose one from the first three recommended lists (usually massage tools).
Fourth: Open an e-commerce app, search for "cleaning tool set", and then buy a set of brooms, mops, and feather dusters to send to the mother... 
"Mom hit me again" gift set
And programmers who don’t even have an e-commerce app may choose the simplest and most straightforward method: sending red envelopes.
So, how to choose a gift? I heard that AI is better than you.
Option 1: Use algorithm to recommend favorite gifts
There is a small program called "Book of Rites" that uses big data and machine learning to recommend a tailor-made gift list by answering 5-6 questions about the gift recipient.
"Book of Rites" Election Questionnaire
After a personal test, the recommendation matches the preferences of the giver pretty well. If you don’t like the most, you can also find inspiration through the recommendation list.
In addition, L'Oréal has made a similar attempt. In 2017, it launched a chat robot, Beauty Gifter, to recommend the most suitable cosmetic gift based on the question and answer of the gift giver and the receiver.
This robot is a cooperative project between L'Oréal and Automat. The way Beauty Gifter chooses gifts is to find the most suitable gift through a well-designed questionnaire.
The recipient will receive a card asking about their skin care and beauty preferences. If you are worried about spoiling the surprise, as long as the party giving the gift knows enough about the information, you can also answer the card questionnaire by yourself.
Option 2: Send your mother a beautiful outfit
On Mother's Day, you can also consider giving your mother a dressing assistant.
At F8, Facebook's annual developer conference in 2017, trend and beauty service robots Epytom Stylist and Sephora appeared. Since then, Amazon has also launched Echo Look, a robot that combines graphic recognition and Alexa to help people choose clothing.
Echo Look combines graphics recognition and voice recognition
Echo Look has a depth-sensing camera, LED lighting for flash, and computer vision technology. Its working principle is that after wearing the clothes, through the voice control, help yourself to take a full-length depth-of-field photo or short video, and then establish a user's personal clothing style preference data set.
After that, it will use machine learning algorithms, combined with the advice of fashion experts, to evaluate whether the user's outfit is appropriate. It even provides guidance and suggestions on the colors, styles and trends of users.
Option 3: Take mom to relive the good old times
Another trick is to use AI to make a gift that combines technology and art.
A small program called "You and Me Back" can use AI technology to restore old photos from a low-definition version to an ultra-high-definition version.
Using it to fix the topic of actresses such as Lin Qingxia and Wang Zuxian has also become a hot spot on Weibo.
The restored picture of the goddess has clear facial features instantly and is more beautiful
Think about it, it should be regarded as a very thoughtful gift to restore the photos of my mother's youth to clear images and help her find her youthful appearance.
"You and Me" used image enhancement technology. Use AI to improve image resolution and optimize it mainly for human faces.
The specific principle is to provide a large number of low-resolution images to the algorithm, and then input the high-resolution original images of these pictures, let the algorithm compare the differences between the two, and learn the skills to increase the resolution. Finally, the blurred picture can be restored.
In terms of image enhancement, in 2018, NVIDIA proposed a method based on deep learning to remove photo noise. The neural network they trained can make blurry images clear in a few seconds. The advantage of this new model called noise2noise is that there is no need to look at a large number of high-resolution sample examples.
Picture on the left: Noisy, unknown creature; Picture on the right: Noisy, a koala
The research team extracted 50,000 images from the ImageNet data set, trained the noise2noise model, and added a random noise distribution to each image. The system must estimate the size of the noise in the image and remove it, and finally the image can be repaired with high quality.
Option 4: Take time to spend more time with your mother
With the help of AI, you may soon be able to pick out gifts that satisfy your mothers, or maybe not. Because every mother is unique.
Using big data analysis and machine learning methods, personalized recommendations will be given, and this method full of technological sense can bring novel experiences.
However, in fact, the unbreakable truth is very simple. The best gift is to spend more time with your parents.
After all, going home is the best gift.
HyperNeuropedia
Image Enhancement
The method of image enhancement is to add some information or transform data to the original image by certain means, selectively highlight the features of interest in the image or suppress (mask) some unwanted features in the image, so that the image matches the visual response characteristics .
Image enhancement can be divided into two categories: frequency domain method and space domain method.
The former regards the image as a two-dimensional signal and performs signal enhancement based on two-dimensional Fourier transform. Using low-pass filtering (that is, only passing low-frequency signals) can remove the noise in the picture; adopting high-pass filtering can enhance high-frequency signals such as edges and make blurry pictures clear.
The representative algorithms in the latter spatial domain method include local averaging and median filtering (taking the middle pixel value in the local neighborhood), etc., which can be used to remove or reduce noise.