Image Color Correction Using Machine Learning Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image color transfer methods face challenges in accurately matching color characteristics between images with varying color proportions and compositions, often resulting in undesirable shifts or artifacts, especially when dealing with multiple distinguishable colors.
Innovation Solution
The approach involves identifying color candidates of mean color correspondences between a reference image and a correctible image, training a classifier to provide a metric for the degree of difference, and iteratively determining optimal color correspondences using this metric as an objective function, with recursive re-training to refine the color correction and generate a transformation field in color space for enhancing the correctible image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional color matching methods (matching average color and standard deviation) are used, then the process is simple and fast, but the color accuracy deteriorates when images have varying color proportions
Solution Approach 1:
The patent segments the color space into multiple distinguishable color regions using clustering algorithms (e.g., k-means). Instead of treating the image as a single average color, it identifies and processes multiple color clusters separately, allowing accurate color transfer even when color proportions differ between images.
Solution Approach 2:
The patent transforms the color matching problem from matching global statistics (average color, standard deviation) to matching local color distributions within segmented regions. It uses parameters such as cluster centers, cluster sizes, and color distances to achieve accurate color correspondence while maintaining computational efficiency.
2Measurement precision
If color space segmentation is used to improve color matching accuracy, then the color transfer quality improves, but the computational complexity increases
Solution Approach 1:
The patent divides the color space into a limited number of clusters (e.g., 5-10 clusters) representing distinguishable colors. This segmentation reduces the complexity by focusing computation on a small number of representative color regions rather than processing all pixels individually, while still achieving accurate color matching.
Solution Approach 2:
The patent reduces computational complexity by transforming the problem from pixel-level processing to cluster-level processing. It uses parameters such as cluster centers and color distances to represent entire regions, significantly reducing the dimensionality and complexity of the color matching computation.
3Productivity
If greedy color pairing is used to match colors between images, then the algorithm is computationally efficient, but the color transfer quality deteriorates due to limited matching options
Solution Approach 1:
The patent replaces the greedy single-best-matching approach with a probabilistic framework that considers multiple matching candidates. It uses parameters such as color distances, cluster sizes, and probability distributions to evaluate and select optimal color correspondences, achieving both computational efficiency and high color transfer quality.
Data Source
AI summary
Systems, methods and computer program products for image color correction are described herein. An embodiment includes identifying color candidates of mean color correspondences between a first image having desired color characteristics and a second image to which the desired color characteristics are to be applied, training a classifier to provide a metric that corresponds a degree of difference between the first image and the second image, and iteratively determining mean color correspondences between the first image and the second image using the metric as an objective function and generating a color-corrected image having the desired color characteristics of the first image using the determined mean color correspondences.