End-to-End Deep Learning OCR Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition (OCR) processes are time-intensive and prone to errors due to sequential steps, leading to potential information loss and distortion, and require significant computational resources.
Innovation Solution
An OCR system utilizing a neural network architecture that processes images of text into characters in a single step, incorporating global context and reducing computational costs by using a convolutional neural network (CNN) for character recognition, segmentation masking, and bounding box generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sequential steps are used in OCR processing, then comprehensive character recognition is achieved, but processing time increases and computational costs increase
Solution Approach 1:
The patent combines multiple sequential OCR steps (de-skewing, de-speckling, binarization, line removal, layout analysis, and character recognition) into a single integrated neural network model. This consolidation processes the entire document in one end-to-end pass, eliminating the time delays associated with sequential processing while maintaining comprehensive recognition capabilities through the model's ability to perform all functions simultaneously.
2Reliability
If multiple sequential steps are used in OCR processing, then comprehensive character recognition is achieved, but computational resources increase
Solution Approach 1:
The patent merges multiple computational steps into a single neural network inference operation. By integrating de-skewing, de-speckling, binarization, line removal, layout analysis, and character recognition into one unified model, the system performs all these functions in a single computational pass, reducing the cumulative computational resource consumption that would occur if each step were executed separately as independent processing stages.
3Ease of operation
If intermediate processing steps are applied sequentially, then document analysis is performed, but information loss or distortion occurs
Solution Approach 1:
The patent implements continuous end-to-end processing through a single neural network model that maintains information integrity by processing the document in one uninterrupted pass. The model receives the original document image and continuously transforms it through all processing stages (de-skewing, de-speckling, binarization, line removal, layout analysis, and character recognition) without intermediate saving or separate handling, thereby preventing information loss or distortion that would occur through multiple discrete processing steps.
4Productivity
If a single step process is used, then processing time is reduced, but recognition accuracy may decrease
Solution Approach 1:
The patent creates a universal neural network model that performs multiple functions simultaneously - de-skewing, de-speckling, binarization, line removal, layout analysis, and character recognition - all within a single processing step. This multi-functional approach maintains high processing speed while achieving accurate character recognition because the model is trained to handle all these tasks comprehensively, rather than requiring separate specialized steps for each function.
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for optical character recognition using end-to-end deep learning. In an embodiment, an optical character recognition system may train a neural network to identify characters of pixel images and to assign index values to the characters. The neural network may also be trained to identify groups of characters and to generate bounding boxes to group these characters. The optical character recognition system may then analyze documents to identify character information based on the pixel data and produce a segmentation mask and one or more bounding box masks. The optical character recognition system may supply these masks as an output or may combine the masks to generate a version of the received document having optically recognized characters.