Example-Based Color Transfer for 3D Model via MGGD Transformation

The 3D model color transfer technique holds significant value across various fields， such as film editing， game development， and medical image processing. This technique allows artists and researchers to enhance the visual quality and realism of 3D models by transferring the color characteristics of real-world materials and environments onto virtual objects， enabling artistic style transfer and fusion. Traditionally， these techniques consist of two primary steps： extracting color features from an example model， such as hue and saturation， and transferring these features to an input model using specific methods， often based on optimal transport theory. These transformations aim to achieve artistic style transfer while enhancing the overall visual quality of 3D models. By utilizing optimal transport， traditional methods facilitate precise mapping of color distributions from the example to the input model， resulting in visually compelling and aesthetically pleasing outputs. However， conventional color transfer methods primarily focus on simple features， such as the mean and variance of color styles， often neglecting the intricate relationships between color styles and geometric structures. This limitation can result in color distortion and artifacts， particularly when applied to complex scenes or detailed models， compromising color consistency and detail integrity. To address these challenges， this study proposes a novel color transfer method for 3D models based on the multivariate generalized Gaussian distribution （MGGD）. By extending traditional statistical approaches， the proposed method incorporates the interdependence between color and geometric structure information in both the input and example models. Leveraging the robust generalization capability of MGGD， this method achieves a precise fit of the true probability density of the input dataset， effectively capturing the complexity of color distributions and their interaction with geometric structures. Building on this framework， the Monge-Kantorovich transformation from optimal transport theory is employed to map color information between models using the MGGD scatter matrix approach. This ensures consistency in overall model color during the transfer process. Furthermore， a stochastic transformation mechanism， based on the shape parameters of the MGGD， introduces controlled randomness to the color transfer process， enhancing adaptability to complex scenes and subtle variations while maintaining color consistency. Numerical experiments demonstrate that the proposed method significantly outperforms traditional techniques in preserving color consistency and detail integrity. Across seven commonly used quantitative evaluation metrics， including Structural Similarity Index Measure （SSIM）， Learned Perceptual Image Patch Similarity （LPIPS）， and Wasserstein distance， the MGGD-based approach achieves superior visual performance.