In this blog post, we’ll examine the principles of pattern recognition technology and machine learning using the “Pudding Face Recognition” app—which was once hugely popular—as a case study.
Do you remember the “Pudding Face Recognition” app that caused a huge sensation in South Korea in 2011? At the time, “Pudding Face Recognition”—a game-style app that found celebrities who looked like you simply by uploading a selfie—was first released on the iPhone App Store in 2010 and recorded over 2 million downloads; an Android version was released the following year. Buoyed by its immense popularity, it later expanded into overseas markets by introducing multilingual versions. At the time, the app went beyond being a simple game to become an engaging information service that provided estimates of users’ appearance, age, and gender. So, how was this app able to provide such information based on just a single selfie?
The core technology behind this app is pattern recognition. Pattern recognition refers to the technology that enables computers to identify features in various types of data—such as text, images, speech, and pictures—and use them to classify or make judgments. Today, this technology has established itself as a key field within machine learning, and the term “pattern recognition” is often used alongside machine learning, data mining, and KDD (Knowledge Discovery in Databases). However, strictly speaking, they are not identical concepts, and each field differs in its scope and purpose. The learning methods used in pattern recognition can be broadly categorized into supervised learning and unsupervised learning.
Supervised learning is a method in which a computer learns rules using data for which the correct answers are already provided. For example, when inputting a company’s historical sales data to predict future sales, the system learns from the existing data and then predicts new results. This method includes not only regression analysis—which predicts continuous values—but also classification, which categorizes data according to specific criteria.
In contrast, unsupervised learning is a method in which a computer identifies common features on its own from data for which no correct answers or classification criteria are provided. Since it involves grouping items with similar characteristics while analyzing the data, it is also called clustering. For example, in supervised learning, when you input a photo of a cat into a computer, you also tell it that it is a cat; however, in unsupervised learning, you simply provide multiple photos without such an explanation and let the computer identify common features and classify them on its own. In this process, the computer extracts features from the data and distinguishes between necessary and unnecessary information. Various algorithms have been developed for this feature extraction and learning, and the widely known deep learning is a prime example of a technology that utilizes this learning method. Currently, in addition to supervised and unsupervised learning, various other learning methods, such as self-supervised learning, are also being utilized.
These pattern recognition technologies are already being used in various fields without us even realizing it. Numerous services—including speech recognition, handwriting recognition, internet search, image search, recommendation systems, facial recognition in photos, and online ad recommendations—operate based on these technologies. As information filtering technologies become more sophisticated, artificial intelligence will also be able to interact with humans more naturally.
Meanwhile, the “Pudding Face Recognition” service ceased operations following a 2013 court ruling related to the infringement of celebrities’ portrait rights. However, the technology for analyzing human faces has continued to advance and is now utilized in various services, such as facial authentication on smartphones, automatic photo categorization, and face recognition features on social media.
Nowadays, it’s not hard to find AI services that, when asked, “Do I look like Kim Tae-hee?” analyze voice data and facial images to provide a response based on reasonable evidence. Pattern recognition technology will continue to serve as a core technology for various AI services in the future.
