Light Field Depth Estimation via Stitched Epipolar Plane Images

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Depth estimation is crucial in light field processing.
• Epipolar-plane image (EPI)-based methods face challenges like discretization errors, limited angular resolution, occlusions, and texture-less regions.

Purpose of the Study:

• To enhance the accuracy and robustness of depth estimation in light field images.
• To address limitations of existing EPI-based depth estimation methods, particularly in challenging regions.

Main Methods:

• Proposed stitched-EPI (SEPI) by shifting and concatenating lines from different EPIs for improved slope computation.
• Introduced half-SEPI algorithm to specifically address occluded regions.
• Developed a depth propagation strategy from edges to interior for texture-less regions.

Main Results:

• SEPI significantly enhances slope computation accuracy.
• The half-SEPI algorithm effectively handles occlusions.
• Depth propagation strategy improves estimation in texture-less areas.
• Experimental results show superior accuracy and robustness compared to state-of-the-art methods.

Conclusions:

• The proposed SEPI-based approach offers a significant advancement in light field depth estimation.
• The method provides more accurate and robust depth maps across all image regions.
• This work contributes to overcoming key challenges in EPI-based depth estimation.