CNN Identity Document Detection Without OCR Text Exposure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data loss prevention (DLP) systems rely on optical character recognition (OCR) techniques that store text data in memory, creating an additional attack vector for data leakage and failing to accurately detect sensitive data in documents with issues like character misdetection and document rearrangement, especially for categories like identity documents.
Innovation Solution
A convolutional neural network (CNN) based identity document detector is trained on augmented image data to identify sensitive categories, avoiding OCR pitfalls by detecting visual features and deployed in a DLP system to flag potentially sensitive documents, with a two-dimensional CNN architecture and data transformations like contrast, rotation, and cropping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If OCR techniques are used to extract text data from documents, then text extraction capability is improved, but data leakage risk increases due to storing text data in memory
Solution Approach 1:
The patent extracts only the essential visual features needed for document classification (layout patterns, text region arrangements, icon positions) while leaving the actual text content unreadable and unextracted. This selective extraction of feature information rather than complete text extraction eliminates the memory storage requirement while maintaining classification capability.
Solution Approach 2:
The patent introduces an intermediary visual feature detection mechanism that acts as a mediator between the document image and the classification system. Instead of directly extracting and storing text, the system uses visual feature detectors to capture structural characteristics, which then serve as the basis for document classification without requiring text data to be stored in memory.
2Measurement precision
If OCR techniques are used for document analysis, then text detection capability is improved, but detection accuracy deteriorates due to character misdetection and document rearrangement issues
Solution Approach 1:
The patent inverts the traditional OCR approach by not attempting to read and interpret text characters at all. Instead of converting images to text and then analyzing the text, the system directly analyzes visual structural features of the document layout. This inversion avoids the inherent limitations of OCR in handling rearranged or distorted documents while maintaining effective document classification.
Solution Approach 2:
The patent replaces the mechanical OCR text recognition system with a visual feature-based detection system. Instead of using character recognition algorithms that are sensitive to distortion and rearrangement, the system employs visual pattern recognition to detect document structure and classify documents based on layout characteristics, which are more robust to document variations.
3Reliability
If visual feature detection is used instead of OCR, then detection robustness is improved, but text extraction capability is lost
Solution Approach 1:
The patent creates a simplified visual representation or copy of the document's structural features rather than attempting to copy the actual text content. By capturing and analyzing visual patterns, layout arrangements, and structural characteristics, the system obtains sufficient information for classification purposes without needing to extract or store the original text data.
Data Source
AI summary
An identity document detector comprising a two-dimensional convolutional neural network is trained to detect categories of identity documents based on inputting unprocessed image data from documents. The documents comprise documents monitored by a data loss prevention (DLP) system across an organization, and each category of identity documents has associated risk levels for DLP. The DLP system performs corrective action to prevent data leakage based on detection of identity documents by the trained identity document detector and risk associated with the detected identity document categories as well as document context within the organization.