High-Fidelity Focal Stack Generation from FFT and MPI-Based Light Field Rendering

Emerging demands in computational photography and immersive imaging have elevated the need for accurate and efficient post-capture refocusing. In this work, the authors explore two state-of-the-art paradigms for light field focal stack synthesis: Fourier Slice Refocusing (FFT-based) and Multiplane Image Rendering (MPI-based). FFT-based methods simulate focus through frequency-domain shearing, preserving global consistency, while MPI leverages depth-layered alpha compositing to enable photorealistic, depth-aware rendering. We process 81-view light fields under both frameworks and assess their performance using no-reference metrics such as Laplacian sharpness, entropy, Tenengrad gradient strength, and edge density. Experimental results show that MPI better preserves edge-rich structures and local gradients, while FFT excels in entropy uniformity and natural focus transitions. This work offers the first metric-grounded comparison of layered and frequency-based light field refocusing and provides a benchmark reference for selecting refocus techniques in AR/VR, robotic vision, and computational optics workflows.