Page 119 - AI for Good-Innovate for Impact Final Report 2024
P. 119
AI for Good-Innovate for Impact
Use Case – 26: Multimodal Bio-verification and Anti-counterfeiting
Platform 26 - MSXF
Country: China
Organization: Mashang Consumer Finance Co., Ltd. (MSCF)
Contact person: Weihong Deng, Yufeng Xie, yufeng.xie@ msxf .com
26�1� Use case summary table
Domain Security and Biometrics
The problem to be Identify people and detect the fake face by using multimodal features
addressed such as images, video, voice, and ultrasound. Protect us from fraud and
false identity information
Key aspects of the Using the AI models to identify real and fake faces, it makes comprehen-
solution sive use of video, image, voiceprint, ultrasound, and other features to
ensure the security of identification.
Technology Biometrics, Computer vision, multimodal fusion, Anti-Counterfeiting,
keywords Privacy and Data Protection.
Data availability Facial images, video, voice, and ultrasound. Parts of our data were
licensed to be made public, which we will make public to promote
research in this area.
Metadata (type of Image, Video, Biometric personal data, template format, voice frequency
data) metadata
Model Training We use BP algorithm to train our anti-counterfeiting model which is
and fine-tuning based on transformer architecture. The data used for training were
collected by ourselves and purchased from data companies. All data
were licensed to be taken by subjects and used for model training.
Testbeds or pilot The platform is deployed as a part of the financial services system of our
deployments company, which used to ensure the security of financial transactions
26�2� Use case description
26�2�1 Description
The multi-modal biometric verification and anti-counterfeiting platform (MmBvAcP), with
its three-tier defense design system, is an exceptionally forward-thinking and innovative
technological solution that plays a crucial role in ensuring security, enhancing efficiency, and
improving user experience.
To begin with, the platform integrates multi-modal biometric verification technology, combining
biometric feature recognition with various modes such as facial recognition and voice
103
