How Facial Recognition Powers Secure Mobile Payments: Inside Ant Financial’s AI

Biometric Authentication Accelerates Mobile Payments

Traditional username‑password‑SMS verification no longer meets the security and convenience demands of mobile finance, prompting a shift toward biometric methods such as facial, fingerprint, and palm‑print recognition, as well as behavioral cues like keystroke dynamics.

Is "Face‑Swipe" Really Reliable?

Early commercial face‑recognition systems faced skepticism over accuracy. Ant Financial’s research focuses on overcoming challenges such as makeup, twins, and facial changes. Their solution uses a multi‑layer convolutional neural network (CNN) on 60‑pixel face patches, combining local perception, weight sharing, and spatial subsampling to achieve robust feature extraction.

Pre‑processing : Detect key facial points, align faces, normalize inter‑eye distance, and crop images before multi‑scale normalization.

Feature Learning : Extract multiple facial region features with separate CNNs, then classify using a Joint Bayesian classifier. Dropout is applied during training to prevent over‑fitting of high‑dimensional features.

Training and Prediction : Training runs on GPUs for large‑scale offline data processing, while inference runs on CPUs for stable, low‑cost online deployment.

Multi‑Layer Closed‑Loop Security Architecture

The facial‑recognition system is part of Ant Financial’s five‑stage security loop: endpoint protection, identity authentication, risk identification & assessment, risk decision & control, and deep analysis. Results feed back in real time to strengthen the loop.

Leveraging Big Data for Fraud Detection

Beyond biometric traits, Ant Financial combines user behavior traces, social relationships, and keystroke timing to build machine‑learning models that enhance trustworthy behavior analysis without disrupting the user.

Liveness Detection

First‑line defense requires users to perform actions such as blinking or turning the head, confirming they are a live person; 3D modeling further thwarts spoofing attacks.

Verification Scheme

Strong identity verification combines remote facial recognition with ID document verification, using public‑key encryption to transmit facial features securely, ensuring the device holder is the genuine user.

Future Applications of Biometrics

Potential scenarios include face‑or palm‑based access control, vehicle unlocking, fashion recommendations based on facial shape, skin‑type cosmetics suggestions, and entertainment features like age‑guessing games.