Image Enhancement Training Using Scene-Guided Unsupervised Image Pairs

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image enhancement methods, particularly those using machine learning, face challenges in efficiently generating high-quality enhanced images without relying on costly and time-consuming manual adjustments by experts, and there is a need for a more efficient method to create diverse training data.

Innovation Solution

A method is developed that utilizes unsupervised image pairs generated through computational degradation models based on scene information, applying random or quasi-random adjustments to high-quality images to create training data, reducing the reliance on manual expert input and minimizing the need for supervised pairs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If manual adjustments by experts are used for image enhancement, then image quality is improved, but time and cost increase

Engineering Contradiction:

Improveimage qualityVSAvoidtime

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The system enables automatic image enhancement through machine learning models that self-adjust visual parameters without requiring manual intervention from experts. The model learns optimal enhancement patterns from training data and autonomously processes images, eliminating the need for time-consuming manual adjustments while maintaining high image quality.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces manual expert adjustment with automated machine learning algorithms. The machine learning model substitutes the mechanical process of manual editing with computational processing, achieving image enhancement through automated prediction and application of visual parameters based on trained patterns.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Manufacturing precision

If manual adjustments by experts are used for image enhancement, then image quality is improved, but cost increases

Engineering Contradiction:

Improveimage qualityVSAvoidcost

Core Design Contradiction:

Manufacturing precisionVSQuantity of substance

Solution Approach 1:

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent uses copying by creating a machine learning model that replicates expert enhancement capabilities. Instead of paying experts directly, the system copies their knowledge and patterns into a trained model that can process images independently, reducing operational costs while maintaining enhancement quality.

Inventive Principle:

Principle #26Copying

3Measurement precision

If supervised training data is used for machine learning, then model accuracy is improved, but data creation complexity increases

Engineering Contradiction:

Improvemodel accuracyVSAvoiddata creation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent inverts the traditional supervised learning approach by using unsupervised learning where the model learns from image pairs without requiring labeled enhancement data. Instead of needing expert-labeled training data, the system uses computational degradation models to create training pairs automatically, simplifying data creation while maintaining model accuracy.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

Inventive Principle:

Principle #25Self-service

4Adaptability or versatility

If computational degradation models are applied to create training data, then data diversity is improved, but processing time increases

Engineering Contradiction:

Improvedata diversityVSAvoidprocessing time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-computing and storing degradation models that can be quickly applied during training data generation. These pre-prepared models enable rapid transformation of images into diverse training pairs without requiring time-consuming real-time computation, thus achieving data diversity efficiently.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system maintains continuity of useful action by using pre-computed degradation models that can be repeatedly applied to generate diverse training data efficiently. Once the degradation models are prepared, they can be continuously applied to create training pairs without repeating the full computational process, reducing processing time while maintaining diversity.

Inventive Principle:

Principle #20Continuity of useful action

Data Source

PatentUS20260024177A1Image enhancement

Publication Date: 2026.01.22 CANVA PTY LTD

AI summary

Methods of training a machine learning model for image processing are described. A method of training includes utilising as a learning objective a reduction or minimisation of a combination of both an image loss and a classification loss. A method of training includes utilising unsupervised images pairs generated by applying a selected degradation model to a target image, the selected degradation model being selected based on classification information associated with the target image. Methods for generating unsupervised image pairs and methods for image processing using a trained machine learning model are also described, together with computer systems and computer-readable storage for performing the various methods.