Document ROI Anomaly Detection for Explainable Fraud Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital identity verification systems face challenges in detecting document anomalies due to high variability across document types, noise levels, lack of genuine reference documents, reliance on fraudulent data, and inefficiencies in human verification, leading to vulnerabilities in fraud detection.
Innovation Solution
A modular anomaly detection method that segments document images into regions of interest (ROIs), applies transformations to generate transform-specific and region-specific features, and computes anomaly scores using machine learning models, enabling flexible, scalable, and interpretable fraud detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a comprehensive fraud detection model is trained across thousands of different document classes, then the model can detect various fraud types, but the model performance is severely affected when new fraud types are incorporated and global model updates are required
Solution Approach 1:
The patent segments the document verification system into separate modules for different document classes and fraud types. Instead of using a single comprehensive model that requires global updates, the system divides detection into independent components that can be updated locally. Each document class has its own trained models, allowing new fraud detection capabilities to be added without affecting other document types or requiring complete model retraining.
2Adaptability or versatility
If general fraud detection models are used, then coverage across multiple document types is achieved, but explanations about the decision to mark documents as fraudulent are not provided
Solution Approach 1:
The patent segments the detection process into multiple independent modules, each responsible for specific document classes or fraud types. Each module provides its own predictions and anomaly scores with associated explanations. This modular architecture allows the system to maintain comprehensive coverage across multiple document types while providing transparent, explainable decisions for each specific detection, as each module operates independently with its own trained parameters and decision logic.
3Measurement precision
If fraud-specific anomaly detection models are developed for specific fraud types, then detection accuracy for known fraud patterns is improved, but the models cannot detect unknown fraud types and require continuous updates
Solution Approach 1:
The patent implements a dynamic system where detection models are continuously updated and adapted. Instead of static fraud-specific models that require complete redesign for new fraud types, the system uses dynamic updating mechanisms that allow models to learn from new data and adapt to emerging fraud patterns. The modular architecture enables selective updates to specific document class models without affecting the entire system, maintaining both high accuracy for known fraud types and adaptability to new threats.
Data Source
AI summary
Described are methods and systems for training a system for detecting anomalies in images of documents in a class of documents. A plurality of training document images of training documents in a class of documents are obtained. For each training document image, the training document image is segmented into a plurality of region of interest (ROI) images, each ROI image corresponding to a respective ROI of the training document. For each ROI image, a plurality of transformations are applied to the ROI image to generate respective transform-specific features for the ROI image and respective transform-specific anomaly scores from the transform-specific features. Based on the respective anomaly scores of the plurality of training document images, a transform-specific threshold is computed for each transformation to separate document images containing an anomaly from document images not containing an anomaly.