A Comprehensive Method for Example-Based Color Transfer with Holistic-Local Balancing and Unit-Wise Riemannian...

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Color transfer is vital in image editing but struggles to balance global style with local detail preservation.
• Existing methods face challenges in achieving both holistic color transformation and fine detail refinement simultaneously.

Purpose of the Study:

• To propose an innovative color transfer method, BHL (Balanced Holistic and Local), that effectively balances holistic style transfer and local detail refinement.
• To introduce an efficient parameter estimation method (uRIG) to mitigate computational complexity.

Main Methods:

• Utilizes optimal transport theory within a generalized Gaussian modeling framework for holistic color transformation.
• Employs a Gaussian Mixture Model (GMM) for per-pixel local color and texture refinement.
• Introduces the unit-wise Riemannian information gradient (uRIG) method for efficient parameter estimation, leveraging the Fisher information metric.

Main Results:

• The BHL method demonstrates superior performance over state-of-the-art techniques in both visual quality and objective metrics.
• Achieves the fastest processing time, averaging 4.874 seconds for high-resolution images.
• Successfully balances holistic transformation with local detail preservation.

Conclusions:

• BHL offers a significant advancement in color transfer, providing a balanced approach combining global and local adjustments.
• The method maintains high efficiency and visual quality, outperforming existing techniques.
• BHL represents a robust solution for complex image editing tasks requiring precise color control.