Neural Network Data Extraction from Receipt Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Extracting data from electronic images, such as receipts, is challenging due to the presence of various data types and the complexity of accurately identifying and extracting relevant information.
Innovation Solution
A method utilizing neural networks, specifically a combination of optical character recognition (OCR) and recurrent neural networks (RNNs) with masking layers, processes images to recognize and extract specific data types by generating probabilities and translating them into output characters, enabling accurate extraction of data types like purchaser information and vendor details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR methods are used to extract data from images, then the extraction process is simple, but the accuracy of extracting specific data types is low
Solution Approach 1:
The patent segments the data extraction process into distinct functional components: an RNN component that sequentially processes characters to capture contextual relationships, and a masking component that selectively identifies specific data types. This segmentation allows each component to specialize in its function, improving overall extraction accuracy while maintaining manageable system complexity through modular architecture.
Solution Approach 2:
The patent introduces a masking layer as an intermediary component between the RNN and the final output. This masking layer acts as a mediator that filters and selects specific data types from the RNN's sequential processing output, enabling precise extraction of target information while suppressing irrelevant data, thus resolving the contradiction between accuracy and complexity.
2Measurement precision
If neural networks are used to improve data extraction accuracy, then extraction precision improves, but processing time increases
Solution Approach 1:
The patent applies preliminary action by using the RNN to process and understand the sequential structure of characters before the masking operation. This preliminary sequential processing establishes contextual relationships early, allowing the masking layer to efficiently select target data types without requiring complex post-processing, thereby reducing overall processing time while maintaining high accuracy.
Solution Approach 2:
The patent extracts only the specific data types of interest through the masking operation, rather than processing or outputting all recognized characters. This selective extraction approach reduces the computational burden of handling unnecessary data, decreasing processing time while maintaining high accuracy for the target information.
3Productivity
If simple character recognition is used, then processing speed is fast, but the ability to identify specific data types is poor
Solution Approach 1:
The patent maintains continuous useful action by having the RNN sequentially process characters in their natural order, preserving the continuous contextual relationships inherent in the data. This continuous sequential processing occurs at high speed, and the masking layer then applies continuous selection to identify specific data types, ensuring both processing speed and identification accuracy are maintained throughout the pipeline.
Solution Approach 2:
The patent segments the processing into two rapid stages: fast sequential character processing by the RNN, followed by fast selective masking for data type identification. This segmentation allows each stage to operate at optimal speed for its specific function, maintaining overall processing speed while achieving accurate data type identification.
Data Source
AI summary
Embodiments of the present disclosure pertain to extracting data corresponding to particular data types using neural networks. In one embodiment, a method includes receiving an image in a backend system, sending the image to an optical character recognition (OCR) component, and in accordance therewith, receiving a plurality of characters recognized in the image, sequentially processing the characters with a recurrent neural network to produce a plurality of outputs for each character, sequentially processing the plurality of outputs for each character with a masking neural network layer, and in accordance therewith, generating a first plurality of probabilities, wherein each probability corresponds to a particular character in the plurality of characters, selecting a second plurality of adjacent probabilities from the first plurality of probabilities that are above a threshold, and translating the second plurality of adjacent probabilities into output characters.