Human Bas-Relief Generation From a Single Photograph

Area of Science:

• Computer Vision
• Computer Graphics
• 3D Reconstruction

Background:

• Generating 3D human models from 2D images is challenging.
• Existing methods struggle with complex poses and multiple individuals.

Purpose of the Study:

• To develop a semi-automatic method for creating human bas-relief from single photographs.
• To improve the accuracy and realism of 3D human shape reconstruction from images.

Main Methods:

• Estimates 3D skeletons from input photos.
• Fits SMPL models to generate a 3D guide model.
• Uses 2D warping and non-rigid registration for contour alignment.
• Integrates base shape with fine-scale normal maps for final bas-relief.
• Employs skeleton-level occlusion resolution and sparse correspondence matching.

Main Results:

• Successfully generates perceptually realistic human bas-relief from single images.
• Handles complex intra- and inter-body interactions effectively.
• Achieves accurate contour registration using non-rigid point matching.
• Outperforms state-of-the-art methods on challenging single and multi-person images.

Conclusions:

• The proposed method offers a robust approach to 3D human bas-relief generation.
• It demonstrates significant improvements in handling complex scenes and occlusions.
• The technique shows potential for various computer graphics and vision applications.