Optical Character Recognition Denoising With Back-Propagated Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional optical character recognition (OCR) systems struggle with accurately converting low-quality images of documents into editable text, leading to inefficiencies and errors in document management systems.
Innovation Solution
A method involving a document management system that processes an image of a document to produce high and low resolution versions. The system performs OCR on the high-resolution image and applies a machine-learned model, such as a denoising autoencoder, to the low-resolution image to produce a denoised version. OCR is then performed on the denoised image, and the results are compared to improve the model's performance through back propagation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional OCR techniques are used on low-quality images, then processing speed is maintained, but text recognition accuracy deteriorates due to image artifacts and noise
Solution Approach 1:
The patent applies a machine-learned denoising model to preprocess low-resolution images before OCR recognition. This preliminary denoising action removes artifacts and noise from the input image, improving the quality of text representation and thereby enhancing OCR accuracy without requiring changes to the OCR engine itself
Solution Approach 2:
The patent introduces a machine-learned denoising model as an intermediary component between the low-quality input image and the OCR recognition system. This intermediary processes the noisy image to produce a cleaner representation, effectively mediating the harmful effects of image artifacts on text recognition accuracy
2Measurement precision
If high-resolution images are used for OCR, then text recognition accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent changes the quality parameter of the input image through denoising processing. Instead of using the original low-quality image or upsampling to high resolution, the system transforms the low-resolution image into a denoised version with improved quality metrics, achieving better OCR accuracy without the computational cost of high-resolution processing
Solution Approach 2:
The patent creates a denoised copy of the low-resolution image that serves as an improved representation for OCR processing. This denoised copy preserves the original image's resolution and size while removing artifacts, providing an efficient alternative to processing actual high-resolution scans
3Measurement precision
If a machine-learned denoising model is applied to low-resolution images, then image quality improves for OCR, but system complexity increases
Solution Approach 1:
The patent employs a universal machine-learned denoising model that can process various types of low-resolution document images (scans, photographs, faxes) through a single unified system. This multi-functional approach handles different document types and noise patterns without requiring separate specialized models, managing system complexity while improving image quality across multiple use cases
Data Source
AI summary
A document management system uses an objective loss function to improve the performance of optical character recognition (OCR) processes on images of documents. The document management system performs OCR on a high resolution version of the image of the document, obtaining a first set of text representative of the text of the document. The document management system applies a machine-learned model on a low-resolution version of the image of the document, producing a denoised image that is of a higher resolution than that input into the machine-learned model. The document management system performs OCR on the denoised image, obtaining a second set of text representative of the text of the document. The document management system subsequently generates a feature vector from the comparison of the sets of text and retrains the machine-learned model with the generated feature vector.