A Multiple-View Geometric Model of Specularities on Non-Planar Shapes with Application to Dynamic Retexturing.

Area of Science:

• Computer Vision
• Computer Graphics
• Geometric Modeling

Background:

• Predicting image specularities from camera pose and scene geometry (SLAM) is a challenging problem.
• Existing models like JOLIMAS assume planar scenes, limiting their applicability.
• Specularities are crucial for realistic image retexturing and augmented reality.

Purpose of the Study:

• To develop a novel geometric model for predicting specularities on non-planar surfaces.
• To overcome the limitations of previous models by removing the planarity assumption.
• To enhance the accuracy and applicability of specularity prediction in computer vision tasks.

Main Methods:

• Proposed dual JOLIMAS, a model that represents specularities as images of 3D quadrics associated with convex surface patches.
• Utilized virtual cameras to predict specularities, enabling the model to handle surface curvature.
• Leveraged the property that specularities remain elliptical on convex surfaces.

Main Results:

• Dual JOLIMAS effectively predicts specularities on non-planar surfaces, demonstrating improved performance over planar models.
• The model was validated on diverse synthetic and real-world datasets with varying objects and lighting.
• Successful application in an image retexturing task was demonstrated.

Conclusions:

• Dual JOLIMAS offers a robust solution for specularity prediction on complex, non-planar surfaces.
• The model's ability to handle surface unflatness significantly advances the field.
• This work provides a foundation for more realistic rendering and augmented reality applications.