Image Quality Assessment for OCR Using Machine Learning Patches

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image quality assessment methods for optical character recognition (OCR) are inefficient and computationally costly, particularly when dealing with diverse and unexpected image types, and often require reference images or sophisticated techniques that are impractical for client devices with limited resources.

Innovation Solution

The implementation uses multiple machine learning models (MLMs) to segment images into patches, generate feature vectors representing image quality and content, and assess suitability for OCR, allowing for efficient and accurate evaluation of image quality without reference images, with the option to train models on client devices for improved performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple machine learning models are used to segment images and generate feature vectors, then measurement precision of image quality assessment is improved, but device complexity increases

Engineering Contradiction:

Improveimage quality assessment accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The image is segmented into multiple patches, and each patch is processed by the machine learning model to generate feature vectors. This segmentation approach allows the system to assess local image quality characteristics (such as blur, noise, and text content) in different regions, thereby improving overall measurement precision without requiring an excessively complex global model.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Feature vectors serve as intermediaries between the input image patches and the final quality assessment. The ML model transforms raw image data into compressed feature representations that capture essential quality attributes, enabling accurate assessment while reducing the computational complexity of direct full-image analysis.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If image quality assessment is performed using traditional methods, then reliability of quality evaluation is improved, but productivity decreases due to computational cost

Engineering Contradiction:

Improvequality evaluation reliabilityVSAvoidprocessing speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

Traditional mechanical/image processing methods for quality assessment are replaced with a machine learning-based system. The ML model learns to predict image quality metrics directly from image patches, substituting complex computational image analysis algorithms with a trained neural network that provides both reliability and computational efficiency.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The machine learning model is trained in advance on a large dataset of images with known quality characteristics. This preliminary training action enables the model to perform rapid quality assessment during actual use without requiring complex real-time computations, thereby improving productivity while maintaining reliability.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If reference images are used for quality assessment, then measurement precision is improved, but ease of operation deteriorates due to additional requirements

Engineering Contradiction:

Improvequality assessment accuracyVSAvoidoperational simplicity

Core Design Contradiction:

Measurement precisionVSEase of operation

Solution Approach 1:

The machine learning model is trained to perform both image quality assessment and text content analysis simultaneously using only the input image itself. The model generates feature vectors that capture quality degradation and text characteristics without requiring external reference images, thereby maintaining measurement precision while significantly improving ease of operation.

Inventive Principle:

Principle #25Self-service

4Measurement precision

If comprehensive feature vectors are generated for each patch, then measurement precision of quality reduction is improved, but use of energy increases

Engineering Contradiction:

Improvequality reduction detection accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The machine learning model extracts only the most relevant features from each image patch that are indicative of quality degradation and text content. Rather than processing all possible image characteristics, the model selectively extracts key features (such as blur amount, noise level, and text presence), thereby achieving high measurement precision with reduced computational energy consumption.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS11893784B2Assessment of image quality for optical character recognition using machine learning

Publication Date: 2024.02.06 ABBYY DEVELOPMENT INC

AI summary

Aspects of the disclosure provide for systems and processes for assessing image quality for optical character recognition (OCR), including but not limited to: segmenting an image into patches, providing the segmented image as an input into a first machine learning model (MLM), obtaining, using the first MLM, for each patch, first feature vectors representative of a reduction of imaging quality in a respective patch, and second feature vectors representative of a text content of the respective patch, providing to a second MLM the first feature vectors and the second feature vectors, and obtaining, using the second MLM, an indication of suitability of the image for OCR.