Image Enhancement Training With Scene-Based Unsupervised Pairs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image enhancement methods, particularly those using machine learning, struggle to effectively balance aesthetic appeal and clarity while minimizing the need for manual expert input and large data collection, often leading to style bias and high costs.

Innovation Solution

A method of training a machine learning model using unsupervised image pairs generated by applying degradation models based on scene information, combining visual parameter adjustments with classification tasks to enhance images, allowing for automatic and diverse enhancement.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning models are trained using traditional supervised learning with large datasets of high-quality images, then image enhancement accuracy is improved, but data collection time and cost increase significantly

Engineering Contradiction:
Improveimage enhancement accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system generates its own training data by applying degradation models to target images, creating unsupervised image pairs automatically. This eliminates the need for manual collection and curation of large datasets of high-quality images, as the model trains on synthetically generated pairs where the degraded image is transformed from the target image through controlled degradation processes

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Instead of requiring real high-quality images for training, the system creates synthetic copies by applying degradation models to target images. These synthetic degraded images serve as training data, replacing the need to collect and store large amounts of actual high-quality image data

Inventive Principle:
Principle #26Copying

2Measurement precision

If machine learning models are trained using traditional supervised learning with large datasets, then image enhancement accuracy is improved, but training cost increases

Engineering Contradiction:
Improveimage enhancement accuracyVSAvoidtraining data quantity
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system generates its own training data by applying degradation models to target images, creating unsupervised image pairs automatically. This eliminates the need for manual collection and curation of large datasets of high-quality images, as the model trains on synthetically generated pairs where the degraded image is transformed from the target image through controlled degradation processes

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Instead of requiring real high-quality images for training, the system creates synthetic copies by applying degradation models to target images. These synthetic degraded images serve as training data, replacing the need to collect and store large amounts of actual high-quality image data

Inventive Principle:
Principle #26Copying

3Ease of operation

If manual editing is used for image enhancement, then control over visual parameters is improved, but processing time increases

Engineering Contradiction:
Improvecontrol over visual parametersVSAvoidprocessing speed
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The system generates its own training data by applying degradation models to target images, creating unsupervised image pairs automatically. This eliminates the need for manual collection and curation of large datasets of high-quality images, as the model trains on synthetically generated pairs where the degraded image is transformed from the target image through controlled degradation processes

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces manual editing processes with an automated machine learning model that learns from unsupervised image pairs. The model automatically adjusts visual parameters by analyzing the relationship between degraded images and their corresponding target images, eliminating the need for manual intervention while maintaining control over enhancement outcomes

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4432217B1Image enhancement
Publication Date: 2026.05.06 CANVA PTY LTD
  • EP4432217B1 patent drawingFigure 1~2
  • EP4432217B1 patent drawingFigure 3
  • EP4432217B1 patent drawingFigure 4

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.