Applications of Machine Learning in eKYC’s identity document recognition
Bui, Duy Tuan (2021)
2021
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2021060414178
https://urn.fi/URN:NBN:fi:amk-2021060414178
Tiivistelmä
Know your customer (KYC) is a client on-boarding process by which financial institutions obtain customers’ information to verify their credentials. The objective of the thesis was to present the applications of Machine Learning and Deep Learning models in disrupting the traditional identity document verification task done manually in a KYC process. Financial institutions can leverage the proposed AI-based solution named eKYC’s identity document recognition in this thesis to save time and incurred cost associated with the customer onboarding procedure.
The pipeline for identity document recognition solution included five modules as following: Card Detector, Card Rotator, Text Detector, Text Classifier, and Text Recognizer. Each module had its own models and each model also had its own training, evaluating, and testing process.
The proposed solution achieved an accuracy of > 98% on average, which was tested on 1000 ID images collected from public sources such as Google Images and Facebook. In terms of inference speed, the full pipeline takes a normal laptop less than one second to extract all information from an input image. Vietnamese identity documents were chosen to evaluate the solution and the results proved its applicability in real-world usages.
The pipeline for identity document recognition solution included five modules as following: Card Detector, Card Rotator, Text Detector, Text Classifier, and Text Recognizer. Each module had its own models and each model also had its own training, evaluating, and testing process.
The proposed solution achieved an accuracy of > 98% on average, which was tested on 1000 ID images collected from public sources such as Google Images and Facebook. In terms of inference speed, the full pipeline takes a normal laptop less than one second to extract all information from an input image. Vietnamese identity documents were chosen to evaluate the solution and the results proved its applicability in real-world usages.