Perceptual Color Space Learning for Image Processing Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based image processing systems face inefficiencies in learning due to the lack of consideration for human visual perception, particularly in distinguishing subtle color differences that are significant to humans but not accurately represented in RGB color space.
Innovation Solution
Converting images from RGB to a color space with a lightness component, such as CIELAB, and using color differences as errors in neural network learning and inference to improve alignment with human vision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RGB color space is used for neural network learning, then the learning process is simple and fast, but the color difference representation does not align with human visual perception
Solution Approach 1:
The patent applies parameter changes by transforming the color space from RGB to a perceptual color space (such as CIELAB or CIEDE2000) that better represents human visual perception. This changes the parameter system used for color difference calculation, allowing the neural network to learn color differences that align with human perception while maintaining computational efficiency.
2Reliability
If more learning data is used to improve model accuracy, then the inference performance improves, but the learning time and computational cost increase
Solution Approach 1:
By changing the color space parameter to a perceptual color space, the patent enables the neural network to achieve better inference performance with fewer learning data points. The transformed color differences provide more meaningful gradients during training, allowing the model to converge faster and achieve high accuracy with reduced learning time and computational cost.
Data Source
AI summary
The image processing device (during learning) converts RGB color space of an input image and a label image into CIELAB color space and performs learning using a color difference between an inferred image and the label image as an error. The image processing device (during inference) converts RGB color space of an input image into CIELAB color space and creates an inferred image using the CIELAB input image.