Image Decomposition Using Temporal Illumination Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods struggle to accurately decompose images into illumination and reflectance components, especially under varying lighting conditions, leading to degradation in image quality and recognition performance.

Innovation Solution

A computer-implemented method using a neural network-based image decomposition model that includes an illumination extraction model and a reflectance extraction model, trained with loss functions considering temporal correlation and brightness changes, to estimate illumination and reflectance maps, allowing for improved image decomposition and white balancing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional image processing methods are used to decompose images into illumination and reflectance components, then the processing speed is maintained, but the decomposition accuracy deteriorates under varying lighting conditions

Engineering Contradiction:
Improvedecomposition accuracyVSAvoidperformance under varying lighting conditions
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent employs a recurrent neural network (RNN) architecture that dynamically adapts to varying lighting conditions by processing temporal sequences of images. The model learns temporal correlations and brightness changes across frames, enabling dynamic adjustment to different illumination scenarios rather than relying on static decomposition methods

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The invention changes the parameters used for image decomposition by incorporating temporal correlation information and brightness change characteristics into the neural network training process. This allows the model to capture dynamic lighting variations and improve decomposition accuracy under varying conditions

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If a neural network-based decomposition model is used to improve image decomposition accuracy, then the decomposition quality is improved, but the computational complexity increases

Engineering Contradiction:
Improveimage decomposition accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image decomposition task into two distinct components: an illumination extraction model and a reflectance extraction model. This segmentation allows each model to specialize in extracting specific characteristics, improving overall accuracy while enabling modular computation that can be optimized independently

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The invention performs preliminary extraction of temporal correlation information and brightness change characteristics from image sequences before feeding them to the decomposition models. This preprocessing step prepares the data in an optimized format, reducing the computational burden during the actual decomposition process

Inventive Principle:
Principle #10Preliminary action

3Reliability

If temporal correlation information is used to train the decomposition model, then the accuracy under varying lighting is improved, but the training time increases

Engineering Contradiction:
Improverecognition performanceVSAvoidtraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts and utilizes temporal correlation information and brightness change characteristics from existing image sequences during the training process. By taking out these temporal features from the input data, the model learns to adapt to lighting variations more effectively, improving recognition performance under varying conditions

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4152273B1Method and device with image processing
Publication Date: 2025.12.31 SAMSUNG ELECTRONICS CO LTD
  • EP4152273B1 patent drawingFigure 1
  • EP4152273B1 patent drawingFigure 2
  • EP4152273B1 patent drawingFigure 3

AI summary

A device with image processing includes: an image acquirer configured to acquire a plurality of images each having a different brightness; and one or more processors configured to extract an illumination map for an input image of the images and an illuminant color of the input image from the input image and temporal correlation information of the plurality of images, based on an illumination extraction model.