Bank Cheque Dataset

Project Overview:

Objective

As a leading data collection and annotation company, we embarked on a project to create a comprehensive dataset for bank cheque processing. Our aim was to support advancements in Optical Character Recognition (OCR), fraud detection, and the automation of cheque clearance processes.

Scope

We successfully collected a diverse range of bank cheque images, ensuring a variety of handwriting styles, bank designs, and conditions, including wear-and-tear. Our expert team meticulously annotated critical components such as payee names, amounts, signatures, and MICR codes, catering to the intricate needs of machine learning models.

Sources

Personal cheques
Business cheques
Cheques from various banks and financial institutions
Photocopies of cheques (to simulate fraud scenarios)
Old and worn-out cheques

Data Collection Metrics

Total Cheques Collected: 250,000 (an increase from the initial 200,000)
Personal Cheques: 150,000
Business Cheques: 70,000
Photocopied Cheques: 15,000

Annotation Process

Stages

Cheque Pre-processing: We standardized images for brightness and alignment, ensuring consistency.
Component Annotation: Our team highlighted key cheque elements for precise data feeding.
Text Transcription: Manual transcription of written elements was done for accurate OCR training.
Forgery Tagging: We tagged photocopied and potentially fraudulent cheques to enhance fraud detection capabilities.
Validation: Utilized existing OCR tools for preliminary verification, followed by thorough manual reviews.

Annotation Metrics

Total Annotations: 1,250,000
Payee Names: 250,000
Amounts in Words: 250,000
Amounts in Figures: 250,000
Date Annotations: 250,000
Signature Annotations: 175,000
MICR Annotations: 75,000

Quality Assurance

Existing OCR Verification: We cross-checked transcriptions against current OCR solutions.
Banking Expert Review: Consultation with banking professionals ensured real-world relevance and accuracy.
Inter-annotator Agreement: Multiple annotators reviewed the same images for consistent annotations.

QA Metrics:

Annotations Reviewed by Banking Experts: 50,000 (20% of total annotations)
Inconsistencies Identified and Rectified: 25,000 (10% of total annotations)

Conclusion

Our proficiency in data collection and annotation has culminated in the creation of the Bank Cheque Dataset, poised to transform various aspects of banking automation, particularly in the realms of cheque clearance and fraud detection..With an extensive range of cheque samples and detailed annotations, it stands as an indispensable tool for fintech advancements.