Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

May 2020

Joint collaboration by
ADVANCE.AI | HPE APAC Innovation Center

DIGITAL KYC
| NVIDIA

SOLUTION
Human verification through machine and vision intelligence
Contents
INTRODUCTION ............................................................................................................................................. 4

TRADITIONAL TO DIGITAL KYC (e-KYC) ........................................................................................................... 4

ID DOCUMENT RECOGNITION ........................................................................................................................ 5

LIVENESS DETECTION .................................................................................................................................... 6

Liveness SDK Version ................................................................................................................................. 7

H5 Liveness Version ................................................................................................................................... 8

FACE COMPARISION ...................................................................................................................................... 8

HARDWARE BUNDLED SOLUTION ................................................................................................................ 10

HPE Edgeline EL1000 Converged Edge System ......................................................................................... 11

NVIDIA T4 Tensor Core GPU ...................................................................................................................... 11

MEASUREMENT & PERFROMANCE DATA .................................................................................................... 11

QPS on KTP OCR, liveness and face comparison ........................................................................................ 12

ADVANTAGES OF e-KYC BUNDLED SOLUTION .............................................................................................. 12

High Performance .................................................................................................................................... 13

Frequent Model Updates .......................................................................................................................... 13

High Accuracy ........................................................................................................................................... 13

Full Functions ........................................................................................................................................... 13

Multi-national Coverage ........................................................................................................................... 13

CONCLUSION ............................................................................................................................................... 13
List of figures
Figure 1: Workflow of an e-KYC solution ..................................................................................................... 5
Figure 2: Measurement and accuracy of OCR on different fields ................................................................ 6
Figure 3: ID document recognition process...................................................................................................... 6

Figure 4: Liveness detection solution .............................................................................................................. 7

Figure 5: Live photo and ID photo comparision ................................................................................................ 9

Figure 6: Accuracy of live and ID photo comparison ....................................................................................... 10

Figure 8: EL1000 Edge Server ....................................................................................................................... 11

Figure 9: NVIDIA Tesla T4 GPU...................................................................................................................... 11

List of tables
Table 1: Hardware specification for KTP OCR, liveness and face comparison .................................................. 12

Table 2: QPS result for KTP OCR, liveness and face comparison..................................................................... 12
INTRODUCTION

This paper highlights the collaboration testing performed by ADVANCE.AI on e-KYC solution from
ADVANCE.AI, and is powered by HPE EL1000 Edgeline server with embedded NVIDIA T4 GPU modules.
The objectives of the test are to ensure that the e-KYC solution works well in under the environmental
setup, and to understand the latency experienced by the services including KTP OCR, face comparison as
well as liveness detection. Functionality is one of the tests performed to determine that the SDK interface
is running without any problem. Tests are also performed against 1000 relevant data set of images,
measuring the average processing time for OCR, anti-spoofing and face comparison. In addition, the
health status of the system, specifically the CPU and the GPU, is also monitored with concurrent requests
executed in parallel for each service.

TRADITIONAL TO DIGITAL KYC (e-KYC)

Know Your Customer (KYC) has been an essential solution to many enterprises, especially to the financial
institutes. It involves the process whereby a variety of services are executed. These services include
processes to verify the identity of a customer and to perform due diligence of on the customer. The KYC
solution is usually deployed by enterprises to confirm that their customers are verified with the identity
whom they declared and claimed to be. It also ensures that the customers have no records of engagement
in illegal activity during the onboarding period.

While KYC is an essential process to many enterprises, there are many drawbacks in taking the traditional
KYC approach. Traditional KYC may result in significant: high manpower cost and frustratingly slow
journey with tedious paperwork. To tackle these issues, e-KYC has become popular by making the KYC
process digital and online. It facilitates enterprises including banks, e-commerce sites, payment
companies, insurance companies, fintech companies, car renting platforms to onboard customers,
promote customer level and assess the risk.

Leveraging on the in-house AI capabilities in OCR and facial recognition, ADVANCE.AI e-KYC solution
effectively verifies the user identities through technologies including ID card recognition, liveness
detection, face comparison. Identity verification products apply to multiple scenarios and businesses,
lowering manual verification costs and errors, increasing business efficiency.

The solution grants companies the ability to minimize fraud significantly. It is the gatekeeper that
prevents fraudsters from disrupting business cycles. The figure below shows a typical e-KYC workflow
adopted by fintech companies in Indonesia market.
Figure 1: Workflow of an e-KYC solution

ID DOCUMENT RECOGNITION

ID document recognition with Optical Character Recognition (OCR) technology is becoming an


indispensable part of the e-KYC process. To minimize the e-KYC duration, recognition accuracy is critical.
ADVANCE.AI’s OCR model for ID document recognition is trained with local data and has significantly high
accuracy against its competitors.

Take the case of Indonesia market, the accuracy of all the available fields are tabulated on the following
table. It ranges from 92.5% (address field) to as high as 99.6% accuracy.

KTP OCR
0.986 0.983 0.996 0.996 0.993 0.995 0.995 0.993 0.979 0.98 0.98 0.994 0.992
1.000
0.950 0.925
0.900
0.850
0.800
0.750
0.700
0.650
0.600
0.550
0.500
Figure 2: Measurement and accuracy of OCR on different fields

The diagram below shows the process of the recommended OCR ID document recognition.

Figure 3: ID document recognition process

With ADVANCE.AI’s OCR technology, it highlights the following key strengths.

1. High accuracy: key fields accuracy is 97%+


2. Low latency: response within 1 second
3. High concurrency: support high Query Per Second (QPS) with auto scaling (30M+ accumulated API
calls)
4. High robustness: support complicated backgrounds, different lighting conditions, various rotation
angles (0° , 90° , 180° , 270° ) ± 10° .

LIVENESS DETECTION

Fraudsters exploring various ways to fake the e-KYC process. The most common cases are using static
photos or recorded videos to exploit the system. Below is a fraud scenario when a user fakes the system
with a static photo.
To ensure business security, verifying if the user is a living person is recognized as an important step
during the e-KYC process. ADVANCE.AI’s liveness detection product is specifically designed to detect
this kind of fraud.

ADVANCE.AI provides two versions of the liveness detection product. They are the
• Liveness SDK version
• H5 Liveness version

Figure 4: Liveness detection solution

Liveness SDK Version


This version is suitable for enterprises with existing mobile apps. It composes of two key modules. They
are the mobile liveness detection SDK and server-side anti-spoofing API.
Mobile liveness detection SDK
In order to detect if a user is a living person, the SDK verifies three movements from the camera’s video
stream. The three movements include head movement, eye blinking and mouth opening. The motion
detection algorithm is embedded in the SDK to prevent round-trip verification from the server.

Once the user passes the movement verification, the SDK returns the best image from the video stream
for anti-spoofing checking.

Anti-spoofing
With pre-recorded video, the fraudsters can still fake the motion verification, if they replay the video
carefully by aligning with the exact moment of each movement verification request.

To detect this kind of fraud, ADVANCE.AI designed the anti-spoofing module to verify if users use pre-
recorded video.

H5 Liveness Version
This version is suitable for enterprises who do not have any mobile apps and would like to make
verification on their websites. It takes 2 - 8 seconds video of the user, and then sends the video to the
anti-spoofing service for verification.

With ADVANCE.AI liveness detection technology, it highlights the following key advantages

1. Excellent post-loan performances with a 2% reduction of the overdue rate based on feedbacks from
existing enterprise customers
2. Customizable UI allows an enterprise customer to have flexible designs to fit in their app styles
3. Minimum SDK size with a base SDK size of about 2MB

FACE COMPARISION

Inspired by top tier research publications, ADVANCE.AI has improved existing image recognition systems
optimized for localized regions, which is highly competitive among the industry. With an accuracy of 95%
on local in the regions, this enhanced image recognition system will curb identity-fraud and also act as a
tool for faster business decisions.

Face comparison is used for identity authentication, which effectively calculates the similarity of two
images of the face, one from the liveness detection and the other from the ID document. ADVANCE.AI
model is trained with millions of local faces, achieving a high accuracy even with low-quality ID
documents.
Figure 5: Live photo and ID photo comparision

The following table illustrates the performance of the face comparison model amongst players in the
market. From the result, it clearly shows that the face comparison model from ADVANCE.AI has
performed significantly superior with higher accuracy of greater than 5% points.
Face Comparison(TAR@FAR)
100.00%

90.00%

80.00%

70.00%

60.00%

50.00%

40.00%
Amazon Microsoft Competitor 1 Competitor 2 Advance AI

0.001 0.0001

Figure 6: Accuracy of live and ID photo comparison

With ADVANCE.AI face comparison technology, it highlights the following key advantages

1. Catered to the local market - the model is trained with over 2 million local faces in Indonesia
2. Customized for e-KYC - the model is fine-tuned for selfie and ID document face comparison
3. Robust algorithm - the algorithm mitigates the impact of shabby and damaged ID documents
4. Effective in different environment conditions - the algorithm supports various lighting environments,
different ages and different head dressing conditions and scenarios

HARDWARE BUNDLED SOLUTION

The e-KYC solution from ADVANCE.AI provides strong advantages that fulfill the condition of the digital
KYC requirement in the region and particularly well-tested in Indonesia using the Kertu Tanda Penduduk
(KTP) which is the national identity card for Indonesian. The e-KYC solution from ADVANCE.AI has been
proven to work at its optimal performance running on an EL1000 Edgeline server from Hewlett Packard
Enterprise (HPE) and accelerated by with Graphics Processing Unit (GPU), the Tesla T4 GPU from NVIDIA.

Each HPE EL1000 Edgeline server is able to mount up to 2 modules of NVIDIA Tesla T4 GPU, which could
deliver the required QPS performance embattled by most enterprises. The following figures show the
components of the key enablers that have been bundled to power the e-KYC solution of ADVANCE.AI to
achieve the superior performing results as indicated in the earlier sections.
Figure 7: EL1000 Edge Server Figure 8: NVIDIA Tesla T4 GPU

HPE Edgeline EL1000 Converged Edge System

The HPE Edgeline EL1000 System allows deployment of a single HPE ProLiant server cartridge, along
with associated storage and I/O options to locations both inside and outside the traditional datacenter
(e.g. factories, branch office and retail outlet). It is a unique system with an industry first, integrating
unprecedented levels of edge compute, precision data capture and control, data center-class security,
device and systems management, as well as large and blazingly fast storage capabilities – all in one
converged box. This rugged and compact system is designed to perform in harsher edge environments
with higher shock, vibration and temperature levels.

NVIDIA T4 Tensor Core GPU

NVIDIA T4 enterprise GPUs supercharge the HPE mainstream servers. Its low-profile, 70-watt (W) design
is powered by NVIDIA TuringTM Tensor Cores, delivering revolutionary multi-precision performance to
accelerate a wide range of modern applications, including machine learning, deep learning, and virtual
desktops. Its revolutionary multi-precision performance accelerates deep learning and machine learning
training and inference, video transcoding, and virtual desktops, and it supports all AI frameworks and
network types, delivering dramatic performance and efficiency that maximize the utility of at-scale
deployments.

MEASUREMENT & PERFROMANCE DATA

QPS is the key parameter that determine the performance of the system as a whole. The following tables
indicate the total QPS performance demonstrated by the various key modules of e-KYC solution of
ADVANCE.AI on HPE EL1000 server and accelerated by NVIDIA T4 modules. The result is measured using
the national ID of Indonesia - KTP.

QPS on KTP OCR, liveness and face comparison

ADVANCE.AI has tested the QPS on HPE EL1000 server with NVIDIA T4 on the following hardware
specifications.

Table 1: Hardware specification for KTP OCR, liveness and face comparison

Item Description

CPU Intel® Xeon® E3 - 4 core with workstation-class CPU, 2.8 GHz

GPU 1 x NVIDIA Tesla T4

RAM 32 GB DDR4

HDD 1 TB SATA-II

The QPS results are shown in the table below.

Table 2: QPS result for KTP OCR, liveness and face comparison

Total QPS running (with 1


Service Name Remark
service type)

Liveness (anti-spoofing) Up to 11 Result for only 1 instance of this service

KTP OCR Up to 7 Multiple instances of this service

Face Comparison Up to 5 Multiple instances of this service

ADVANTAGES OF e-KYC BUNDLED SOLUTION

The bundled solution of e-KYC from ADVANCE.AI, with HPE server and NVIDIA GPU hardware, contributes
to significant advantages of the digital KYC process in the region. Some of these strengths are highlighted
in more details below.
High Performance
The compact form factor of the bundled digital KYC solution runs unmodified enterprise-class analytics
software, with each system being engineered to deliver very high performance. This solution is energy-
efficient solution, achieving the highly demanded AI models that need to be executed as inference
engines.

Frequent Model Updates


The bundled digital KYC solution is supported by professional teams from ADVANCE.AI, HPE and NVIDIA,
that possesses computer vision development team with leading global talents, and allows consistent
improvement in model performance by accumulating data for model optimization.

High Accuracy
The core algorithms of each product along the e-KYC process are specially designed for e-KYC scenarios,
with a large amount of training data collected locally, constantly iterating, ensuring the accuracy
sufficiently meets the needs of the enterprises.

Full Functions
The bundled digital KYC solution provides a range of product and services that include OCR, Liveness
Detection, Face Comparison and other essential functions for e-KYC.

Multi-national Coverage
The bundled digital KYC solution supports multi-national markets including Indonesia, Philippines,
Vietnam, India, and more, and reducing barrier of entry for enterprises into the new digital market.

CONCLUSION

With the measurement and testing performed on the bundled solution, it shows the joint effort to deliver
a leading age digital KYC solution that is highly demanded in the age of the virtual and digital revolution
for today and the future. It helps to transform the digital business of many enterprises and enable a
smoother intake of transactions to be performed in parallel and simultaneous widely across the region. It
helps to defend against fraud and protect potential risk and losses that could have been suffered by the
enterprises.
The bundled solution of the digital KYC solution consists mainly of the following
• e-KYC technology from ADVANCE.AI
• GPU accelerator from NVIDIA
• Compute server from HPE

For more details, please refer to the following


• EL300 Edgeline server from HPE - https://support.hpe.com/hpesc/public/docDisplay?docId=emr_na-c05387745
• Tesla T4 GPU module from NVIDIA - www.nvidia.com/T4
• e-KYC products from ADVANCE.AI - https://in.advance.ai/apply

HPE and NVIDIA worked on the bundled digital KYC solution with ADVANCE.AI who was established in
2015. ADVANCE.AI is a leading big data and AI company in Asia helping to solve digital transformation,
fraud prevention and process automation for enterprise clients. The product suite covers three areas:

• AI solutions, including e-KYC, intelligent process automation and chatbot


• Risk management solutions, including alternative credit scoring, fraud detection and prevention
• Digital lending solutions, including digital onboarding, Smart Decision engine, Smart Collection system

Learn more at

www.advance.ai
www.hpe.com
www.nvidia.com

You might also like