Personalized Image Compression Using User Preference Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Lossy image compression degrades image quality, leading to a suboptimal user experience, as storage space on devices and servers is finite and users have varying preferences for image importance.

Innovation Solution

A machine-learning-based approach using a feature-detection model to identify image features and a user-specific model to determine personalized compression settings, allowing for targeted compression based on user preferences.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If machine learning models are trained to predict user preferences for image compression, then compression settings can be optimized for individual users, but the complexity of the system increases due to additional processing requirements

Engineering Contradiction:
Improvecompression setting adaptabilityVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by training machine learning models offline to predict user preferences before actual image compression occurs. The models are pre-trained on user behavior data and stored for rapid inference during compression operations, separating the complex training phase from the execution phase to reduce real-time system complexity

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Machine learning models serve as intermediaries between user preferences and compression parameters. Instead of directly implementing complex preference analysis, the patent uses trained models to translate user behavior patterns into optimal compression settings, simplifying the decision-making process while maintaining adaptability

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If multiple machine learning models are used to predict different compression parameters, then prediction accuracy improves, but the processing time and computational resources increase

Engineering Contradiction:
Improvepreference prediction accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the compression parameter prediction into multiple specialized machine learning models, each trained to predict specific parameters (e.g., quality factor, format selection, resolution). This segmentation allows each model to be optimized for its specific task, improving overall prediction accuracy while enabling parallel execution to reduce total processing time

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system applies partial action by selectively invoking only the necessary prediction models based on the input image characteristics and user profile. Not all models are executed for every compression task - the system determines which subset of models is needed, reducing processing time while maintaining accuracy for the parameters that matter most for each specific case

Inventive Principle:
Principle #16Partial or excessive action

3Adaptability or versatility

If user preference data is collected and processed in real-time, then personalization accuracy improves, but the computational load and energy consumption increase

Engineering Contradiction:
Improvepersonalization accuracyVSAvoidcomputational energy consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

User preference data is processed and models are trained in advance during offline periods when computational resources are more readily available. This preliminary processing converts raw preference data into structured models that can be rapidly applied during actual compression operations, reducing real-time energy consumption while maintaining high personalization accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses trained machine learning models as copies of user preferences that can be applied repeatedly without reprocessing the original preference data. Once a model is trained on user behavior, it serves as a lightweight copy that requires minimal computational resources to execute during compression, preserving personalization accuracy while reducing ongoing energy consumption

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3942467B1Machine learning-based image compression setting reflecting user preferences
Publication Date: 2026.04.29 GOOGLE LLC
  • EP3942467B1 patent drawingFigure 1
  • EP3942467B1 patent drawingFigure 2
  • EP3942467B1 patent drawingFigure 3

AI summary

Embodiments described herein relate to methods, devices, and computer-readable media to determine a compression setting. An input image may be obtained where the input image is associated with a user account. One or more features of the input image may be determined using a feature-detection machine-learning model. A compression setting for the input image may be determined using a user-specific machine-learning model personalized to the user account based on the one or more features in the input image. The input image may be compressed based on the compression setting.