Contactless Fingerprint Scanner
Upgrading fingerprint technology to the contactless 3D era
Intelligent Fingerprint/Palmprint System
Making identification smarter and safer
MOQI· ID Identity Authentication Platform
Providing flexible, convenient, and privacy-preserving API/SDK for various industries
Invention of Multi-Scale Vector and Graph Representation Framework
Full-stack technology for unstructured data processing
Processing over billions of fingerprints and palmprints
Identifying tens of billions of fingerprints and palmprints in seconds
Moqi Tech Talk
Linpeng Tang: High-speed and Accurate Matching to 2 Billion Images in Seconds
We are able to achieve accurate search across 2 billion fingerprints in seconds with multi-scale image representation, high-speed heterogeneous matching framework and distributed parallel processing.
Qingdi Zhang: Heterogeneous Parallel Computing and Performance Optimization
Applying parallel computing algorithms using CUDA has significantly improved matching performance due to optimizations in areas such as memory access, local storage, code branches, instruction optimization, data transfers, etc.
Qin Liu: Application of Persistent Memory in Image Search System
By applying the latest NVDIMM technology and memory mapping technology (mmap) in Linux, the system can store all image features necessary for heterogeneous parallel computing algorithms in a distributed memory database. This significantly lowers costs for massive image searches and enhances stability of the distributed image search system.
Linpeng Tang: Contactless Fingerprint Acquisition and Privacy-Preserving Biometrics
By upgrading fingerprint acquisition technology to contactless 3D era, we provide secure, reliable and privacy-preserving biometrics technology to billions of people.
Cheng Tai: Technological Breakthroughs in Traditional Machine Learning Algorithms Revolutionize Fingerprint Recognition Technologies
Transforming the fingerprint matching problem into high-accuracy image search problem that surpasses the limitations of traditional machine learning in accuracy, training data and performance, truly a first for unlabeled matching.