Low Resolution OCR Using Neural Network Layout Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-resolution cameras in mobile devices pose challenges for optical character recognition (OCR) due to difficulties in binarization and character segmentation, resulting in lower recognition rates and making it impractical to capture and process documents effectively.
Innovation Solution
A system combining convolutional neural networks, dynamic programming, and language models to enhance low-resolution OCR capabilities, allowing for robust character and word recognition by predicting character locations, performing layout analysis, and optimizing word recognition through dynamic programming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional OCR processes are used with low-resolution camera images, then the OCR system can process documents, but the recognition rate deteriorates due to poor binarization and segmentation
Solution Approach 1:
The patent merges the traditionally separate steps of binarization, segmentation, and character recognition into a single integrated neural network model. This unified approach allows the system to process low-resolution images directly without requiring high-quality intermediate representations, thereby maintaining reliable recognition rates despite poor binarization and segmentation quality in traditional pipelines
Solution Approach 2:
The patent replaces the mechanical image processing pipeline (binarization -> segmentation -> recognition) with a neural network-based system that directly maps pixel data to character识别. This substitution eliminates the need for high-quality intermediate processing steps and enables effective OCR on low-resolution images by learning direct feature representations
2Device complexity
If separate binarization and segmentation steps are performed, then the OCR process can be simplified, but the recognition rate deteriorates due to inability to handle low-resolution images
Solution Approach 1:
The patent combines multiple OCR functions (binarization, segmentation, character recognition, and word recognition) into a single integrated neural network architecture. This unified model processes images end-to-end, eliminating the need for separate processing steps while maintaining high recognition rates even on low-resolution inputs
3Length of moving object
If low-resolution cameras are used to capture documents, then the device size and cost are reduced, but the image quality deteriorates making quality binarization and segmentation impossible
Solution Approach 1:
The patent replaces traditional mechanical image processing requirements with a neural network-based system that can directly process low-resolution pixel data. This substitution allows the system to extract meaningful character information from blurry, low-resolution images without requiring high-quality intermediate processing steps
Data Source
AI summary
A global optimization framework for optical character recognition (OCR) of low-resolution photographed documents that combines a binarization-type process, segmentation, and recognition into a single process. The framework includes a machine learning approach trained on a large amount of data. A convolutional neural network can be employed to compute a classification function at multiple positions and take grey-level input which eliminates binarization. The framework utilizes preprocessing, layout analysis, character recognition, and word recognition to output high recognition rates. The framework also employs dynamic programming and language models to arrive at the desired output.