Tone-Mapped Mean-Shift Based Environment Map Sampling.

Area of Science:

• Computer Graphics
• Image Processing

Background:

• Realistic rendering requires significant computational resources.
• Environment map sampling is crucial for reducing rendering costs.
• Existing methods may yield unbalanced importance metrics and lack user control over strata number.

Purpose of the Study:

• To develop a novel and pragmatic approach for environment map sampling.
• To reduce computational cost in realistic rendering.
• To improve rendering quality and robustness.

Main Methods:

• Exploiting adaptive mean-shift image clustering with tone-mapping for initial strata generation.
• Implementing an adaptive split-and-merge scheme to refine strata and balance their distribution.
• Utilizing image quality metrics such as SSIM, RMSE, and HDRVDP2 for evaluation.

Main Results:

• The proposed method yields oversegmented strata with uniform intensities and captured light region shapes.
• The adaptive split-and-merge scheme effectively refines strata for a more balanced distribution.
• Achieved comparable or superior rendering quality compared to state-of-the-art methods.
• Demonstrated robustness to viewpoint variations, environment rotation, and sample number.

Conclusions:

• The novel environment map sampling approach effectively reduces computational cost for realistic rendering.
• The method provides high-quality and robust rendering results.
• The adaptive split-and-merge scheme addresses limitations of initial strata generation.