Why can the iPhone distinguish the human face ... "Roofing" mechanism that distinguishes the unexpected difference between the brain and the computer and the common face

notebook-laptop
13/07/2022
520 Views

iPhoneの「Face ID」など、いまでは日常生活のさまざまな場面で顔認証技術が用いられている。NECフェローの今岡仁さんは「顔認証の仕組みは人間の脳にも備わっている。コンピュータによる認証の仕組みは、脳の仕組みと共通な部分が多く、決してブラックボックスではない」という――。

* This article is a re -edited part of Hitoshi Imaoka's "Facial Authentic Textbooks Creating a Test AI to Create Tomorrow's Business" (President).

写真＝iStock.com／Zephyr18※写真はイメージです全ての画像を見る（4枚）

The components are defined for what is the face authentication system, and the mechanism is briefly explained (Chart 1).

出所＝今岡仁『顔認証の教科書　明日のビジネスを創る最先端AIの世界』（プレジデント社）

For example, in airport immigration management, people who are going to pass the border confirm whether they are the same person as the name and nationality person written in the passport.At this time, the face authentication system shoots the face image of a person who is about to pass through the border, compares the face image embedded in the passport, and determines whether it is the same person.

Therefore, two images are required to perform face authentication.The first one is an image of a person on a certified terminal placed on the border when trying to pass the border.This is called a "collation image" and is shot every time you try to pass the border for "collation" of your face image.The other one is a "registered image", a face image embedded in the passport.

なぜiPhoneは人間の顔を見分けられるのか…脳とコンピュータの意外な違いと共通点顔を見分ける｢物差し｣の仕組み

The collation image is taken with a device called "authentication terminal" located at the border (immigration and exits).

Facial authentication also requires a "face recognition algorithm" in which AI (artificial intelligence) determines whether it is the same person or not.

Furthermore, facial authentication algorithms have three important features: face detection, feature extraction, and face correspondence.

The first function, the face detection, finds the face in the registered image and the collation image, and identifies the position of the face.

The second functional extraction, which is a variety of personal faces in the collation images and registered images, is, for example, (0.5, 0.2, 0.3, ..., 0.2, 0.7).It is extracted as a numerical column (feature) of a hundred to thousands of elements.

This is because if the computer handles the image as it is, it takes a lot of time to "judge whether the same person is the same person", and it is not possible to achieve speedy identity verification.Therefore, in order to compress the amount of information and make it easier to judge, the face image is replaced with the characteristics of numerical data.Depending on the design of this feature, the accuracy and speed of the face authentication system will change significantly.

The third function, which is the third function, compare the features extracted from the collation image with the features extracted from the registered image on the database, and finalize whether they are the same person.