By Lulingzicun
The Lunar New Year has arrived, and predicting fortunes for the Year of the Rooster is a popular pastime among young people during their leisure time. Although the act of “fortune-telling” may seem more like a recreational activity, it can boost confidence or provide warnings, effectively “changing fortunes” through psychological suggestion.
Interestingly, this year during the Spring Festival, Baidu’s image search launched a “Selfie Fortune Teller” tool. By opening the Baidu app and taking a selfie or uploading a photo, the image search will provide a fortune analysis for the New Year, enticing many netizens to upload their photos to check their fortunes. This new feature may seem ridiculous at first, with many joking that smartphones can now tell fortunes. However, this is not laughable, as the “Selfie Fortune Teller” employs advanced image recognition technology, which is backed by substantial expertise.
What Advanced Technology Lies Behind “Fortune Telling”?
The “Selfie Fortune Teller” primarily utilizes Baidu’s image search capabilities, relying on Baidu’s image recognition technology. It evaluates key facial features such as eyebrows, eyes, lips, and face shape through facial recognition technology to provide a fortune analysis based on traditional Chinese face reading.
Frankly, the “fortune-telling” feature lacks substantial scientific basis but is indeed a form of “black technology.” This is because human facial information is incredibly rich and unique, with distinct features such as face shape, eyebrows, eyes, and nose contours referred to as key points.
Baidu’s facial recognition technology involves multiple rounds of processing each person’s facial image, including contour positioning, global rough positioning, and local fine positioning, tracking 72 key facial points to create a facial expression map. This allows the system to identify individual expressions and features, which are then used to infer the so-called “fortune,” entertaining the public during the Spring Festival and bringing joy.
In addition to “fortune-telling,” the Baidu Brain-powered Xiaodu robot recently defeated the previously unbeatable “Water Brother” Wang Yuhang on the stage of “The Strongest Brain,” competing in facial recognition. Before the recording of the show, Wang Yuhang requested to lower the brightness of the video by 20%, significantly increasing the challenge, hoping to defeat Xiaodu.
Wang Yuhang mentioned that he wondered if a darker environment would increase noise for the computer. Lowering brightness might make it harder for the computer to recover the brightness during processing. However, the accuracy of image recognition is far superior to that of humans. Machine recognition relies on image contours and is not affected by reduced brightness.
Image Recognition Seems Simple But Is Extremely Challenging
When we see the fortune-telling or Xiaodu defeating Water Brother, it often appears to happen in just a few seconds. The tracking of 72 key facial points seems like an instantaneous event.
However, recognizing faces is deceptively complex. Human faces exhibit similarities, with minimal differences between individuals. The structures of all human faces are similar, and even the shapes of facial features are quite alike. This characteristic aids in facial positioning but complicates distinguishing between individuals.
Additionally, individual facial features are highly variable. A person’s face can change significantly with various expressions, and the visual image of a face can differ greatly from different angles. Moreover, facial recognition is affected by lighting conditions (e.g., day vs. night, indoors vs. outdoors), various facial coverings (e.g., masks, sunglasses, hair, beards), and age.
Consequently, the challenges of image recognition test the algorithms and the richness of data. An international image database called LabelMe contains images meticulously labeled, including details of building shapes, contours, windows, cars, grass, and roads. This database has about 100,000 images, with around 10,000 clearly labeled. These images are manually input one by one.
In the recent competition on “The Strongest Brain” between Xiaodu and Wang Yuhang, Xiaodu’s algorithm faced various environmental lighting challenges, as well as difficulties related to facial posture and accessories, camera image issues, and even frame loss during dynamic monitoring. The robot must undergo a series of processes including face detection, preprocessing, feature extraction, and matching and recognition.
These challenges test the accuracy of machine vision. However, with Baidu’s in-depth research in this field, image recognition technology will continue to improve in accuracy.
What Practical Applications Can Facial Recognition Have?
While fortune-telling and entertainment programs may seem distant from our daily lives, what practical applications does image search have in our everyday lives? This is likely the real question that impacts us.
Image search and facial recognition technologies can be applied in many areas, including the following four main categories:
1. They can provide services such as photo translation, answering questions from images, checking medications, and identifying food recipes, solving many problems where text cannot describe what images can. When we are out and see a product we like, we can open a shopping app, recognize it, and purchase it directly. For instance, when the Baidu app recognizes a picture of braised pork, it will directly provide the recipe.
2. Because image recognition technology is far more reliable than humans, it is used in internet finance for credit assessments, shortening intermediary processes while increasing security and even enabling facial payment. Baidu Finance has already integrated facial recognition into its credit assessment processes. This method allows for instant approvals and prevents identity fraud, significantly enhancing financial security.
3. The widespread adoption of the second-generation ID card in China has provided digital photo data for over 1.2 billion people’s identity information. The millions of surveillance cameras in safe cities generate massive amounts of data daily, allowing law enforcement to enter the era of big data applications. Facial recognition technology, supported by this vast amount of data, has become a crucial technical foundation for helping police find missing children and solve cases.
4. Preparing for future technologies such as autonomous driving, Baidu’s self-driving car can automatically avoid other vehicles and pedestrians because it continuously collects real-time road conditions through radar, with onboard computers analyzing and processing these images and videos. It can be said that image recognition technology will be a vital foundational support for future autonomous driving.
In Conclusion:
While facial fortune-telling may seem like mere entertainment, the underlying technologies are genuinely transforming our lives. Nevertheless, facial recognition remains one of the most challenging problems in pattern recognition and computer vision. Continued efforts from all parties are necessary, and we can expect more and better facial recognition products in the future.
================
Lulingzicun, technology columnist. Awarded the Top Ten “Freelance Journalist” by Titanium Media in 2013 and the New Author of the Year by Donews in 2013. Currently contributing to over 40 media platforms including Baidu Baijia, Huxiu, Titanium Media, iBlack Horse, Geek Park, Chuangjian, Jiemian, Everyone is a Product Manager, Blue Whale TMT, iResearch, Zhihu, Toutiao, Yidian Zixun, Tencent News, Sina News, Blog China, etc.
Statement: All works of this public account require authorization from Lulingzicun for any reproduction or citation (WeChat: lulingzicun).