Display Panel Color Shift Correction Using Neural Network Inverse Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing color correction methods for display panels, such as OLEDs, are time-consuming and computationally intensive, requiring significant resources to achieve accurate color calibration due to their inherent display characteristics and the need for regular recalibration over time.
Innovation Solution
A neural network-based method using transfer learning to establish a forward and inverse model for color correction, employing a small number of RGB values to quickly adapt to unknown panels, ensuring luminance and color accuracy through white point correction and inverse model initial guesses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional polynomial functions or look-up table methods are used for color calibration, then color accuracy can be achieved, but the process requires significant time and computational resources
Solution Approach 1:
The patent transforms the color calibration problem from traditional polynomial fitting or iterative LUT methods to a neural network parameter optimization problem. By representing the transformation matrix as learnable parameters in a neural network model, the system achieves accurate color correction while reducing computational complexity and calibration time through efficient gradient-based optimization rather than iterative calculations.
Solution Approach 2:
The patent replaces traditional mechanical/mathematical color calibration methods (polynomial functions, iterative LUT searches) with a neural network-based system. The neural network learns the color transformation through training data, substituting the need for time-consuming iterative calculations with a pre-trained model that can rapidly predict corrected RGB values, significantly reducing calibration time while maintaining accuracy.
2Measurement precision
If traditional color calibration methods are used, then accurate color correction can be achieved, but computational resources and processing complexity increase significantly
Solution Approach 1:
The patent reformulates the color transformation problem by parameterizing the transformation matrix within a neural network architecture. This allows the complex color correction calculations to be performed through efficient neural network forward propagation rather than traditional iterative mathematical methods, reducing computational complexity while maintaining correction accuracy.
Solution Approach 2:
The patent uses a neural network model that is trained on color transformation data and then copied/deployed for actual color correction tasks. The trained network parameters (weights and biases) are transferred from the training phase to the inference phase, allowing accurate color correction without repeating the complex training computations during actual operation.
3Measurement precision
If neural network models are trained on large datasets, then model accuracy improves, but training time and computational cost increase
Solution Approach 1:
The patent performs the computationally intensive neural network training in advance during a setup phase, creating a pre-trained model that can then be deployed for rapid color correction. This preliminary action separates the time-consuming training process from the actual color correction operations, allowing accurate models to be prepared beforehand without impacting the speed of subsequent corrections.
Solution Approach 2:
The patent implements a two-stage approach where a general color correction model is trained on diverse data, and then fine-tuned or adapted to specific display panels using transfer learning. This allows the system to benefit from large-scale pre-training while reducing the amount of panel-specific data and time needed for final adaptation, achieving high accuracy without requiring exhaustive training for each individual panel.
Data Source
AI summary
A method of correcting color shifts in display panels includes converting target RGB values to XYZ values, converting the XYZ values to RGB values using an inverse model, and a panel under test displaying a pixel according to the RGB values. The inverse model is trained based on a neural network model.