A Hierarchical Superpixel-Based Approach for DIBR View Synthesis.

Area of Science:

• Computer Vision
• Computer Graphics
• Image Processing

Background:

• View synthesis enables virtual exploration of static scenes using color and depth maps.
• Depth-image-based rendering (DIBR) is an efficient approach requiring only one color-depth pair, saving storage and bandwidth.
• Existing DIBR methods often suffer from artifacts like cracks, disocclusions, and ghosts during 3D warping.

Purpose of the Study:

• To propose a novel DIBR pipeline for view synthesis.
• To address and mitigate common artifacts generated during 3D warping in DIBR.
• To enhance the structural integrity and visual quality of synthesized views.

Main Methods:

• Developed a novel Depth-Image-Based Rendering (DIBR) pipeline.
• Adapted and utilized a hierarchical image superpixel algorithm for image reconstruction.
• Incorporated artifact mitigation techniques for cracks, disocclusions, ghosts, and out-of-field areas.

Main Results:

• The proposed DIBR pipeline effectively tackles various artifacts from 3D warping.
• The use of hierarchical superpixels preserves scene structural characteristics during reconstruction.
• Achieved superior average results compared to state-of-the-art methods on public datasets.

Conclusions:

• The novel DIBR pipeline offers improved view synthesis quality by reducing artifacts.
• The hierarchical superpixel approach is key to maintaining scene structure.
• Demonstrated potential for real-world applications in virtual exploration and content creation.