Image Lighting Correction via Dual Neural Network Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional white balancing techniques, especially machine learning-based methods, assume uniform lighting and require large amounts of high-quality annotated training data, which limits their effectiveness in handling non-uniform lighting scenarios and data efficiency.

Innovation Solution

A method using two neural networks to generate predicted lighting data and embedding data for an input image, calculating a lighting weight, and applying it to basis lighting data to produce correction lighting data, which is then used to create a white-balanced image, without relying on uniform lighting assumptions or extensive annotated data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning based white balancing methods are used, then color correction capability is improved, but requirement for high-quality annotated training data increases

Engineering Contradiction:

Improvecolor correction accuracyVSAvoidamount of training data

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent segments the white balancing task into two separate neural networks: one for lighting estimation and another for color correction. This segmentation allows each network to be trained independently with less data, reducing the overall requirement for large-scale annotated training data while maintaining color correction accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces lighting estimation data as an intermediary between the input image and the final color correction output. The first neural network estimates lighting conditions, and this estimation serves as intermediate input to the second neural network which performs the actual color correction, enabling accurate correction without requiring extensive annotated training data.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Speed

If conventional white balancing methods are used, then processing speed is maintained, but ability to handle non-uniform lighting deteriorates

Engineering Contradiction:

Improveprocessing speedVSAvoidhandling of non-uniform lighting

Core Design Contradiction:

SpeedVSAdaptability or versatility

Solution Approach 1:

The patent applies local quality by enabling the neural networks to handle different lighting characteristics in different regions of the image. The lighting estimation network captures spatial variations in lighting conditions, allowing the system to adapt to non-uniform lighting while maintaining efficient processing through automated neural network inference.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20240233080A9Method and device for correcting lighting of image

Publication Date: 2024.07.11 SAMSUNG ELECTRONICS CO LTD

AI summary

A method, implemented by a processor, of correcting lighting of an image includes inputting an input image to a first neural network and generating predicted lighting data corresponding to lighting of the input image and embedding data corresponding to a feature of the input image, inputting the generated predicted lighting data, the generated embedding data, and sensor data to a second neural network and generating a lighting weight corresponding to the input image, and generating correction lighting data for the input image by applying the generated lighting weight to preset basis lighting data corresponding to the input image.