Custom Neural OCR for Accurate Mobile Check Field Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current OCR processes for financial document processing, such as mobile check deposit, suffer from high error rates, leading to inefficiencies and resource wastage, especially when handling millions of transactions.
Innovation Solution
Implementing customized neural network models trained for specific data field features, allowing for real-time or near-real-time extraction of check data fields using active OCR on mobile devices, which can process live camera imagery and improve extraction accuracy and efficiency by processing in parallel.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If standard OCR processes are used for financial document processing, then the system can handle large volumes of transactions, but the error rate increases significantly
Solution Approach 1:
The patent segments the document processing task by dividing the document into multiple regions of interest (e.g., header, body, footer, specific data fields) and applies specialized processing to each region. This allows the system to maintain high accuracy for critical fields while processing large volumes of documents efficiently.
Solution Approach 2:
The patent applies different processing qualities and methods to different parts of the document. Critical data fields receive enhanced processing with customized neural network models and multiple validation checks, while less critical areas use standard OCR. This local quality approach maintains high overall accuracy without proportionally increasing processing time for entire documents.
2Measurement precision
If customized neural network models are implemented for specific data field extraction, then extraction accuracy improves, but system complexity increases
Solution Approach 1:
The patent implements a universal document processing framework that can handle multiple document types and data fields through a common architecture. The system uses a standardized pipeline with configurable components that can be adapted to different financial documents (checks, invoices, receipts) without requiring completely separate systems, thus managing complexity while maintaining high accuracy.
Solution Approach 2:
The patent introduces intermediary components such as pre-processing modules that prepare documents for extraction, post-processing modules that validate and clean extracted data, and configuration layers that manage model parameters. These intermediaries simplify the overall system architecture by creating clear separation between data input, processing logic, and output generation.
3Speed
If parallel processing is used for real-time extraction, then processing speed increases, but resource consumption increases
Solution Approach 1:
The patent applies partial parallel processing by selectively processing only the most critical data fields in parallel with high-priority models, while less critical fields use sequential processing. This approach achieves real-time performance for essential extraction tasks without the full resource cost of parallel processing all fields, optimizing the balance between speed and resource consumption.
Data Source
AI summary
A computer implemented method, system, and non-transitory computer-readable device that may be used in a remote deposit environment. A plurality of differing neural networks, such as customizable neural networks, are selected, based on an architecture and features of the data field, to extract data from specific data fields on a document. The selected customizable neural networks are trained by historical or synthetic data corresponding to the data fields. Upon receiving, from a neural network Optical Character Recognition (OCR) system, a selected first trained customizable neural network model and at least a second selected trained customizable neural network model, the data fields are extracted, based on a series or parallel configuration of the customizable neural networks, and the extracted data communicated to a remote deposit process.