Efficient representation and coding of dynamic light field: a data-driven approach based on dynamic mode decomposition
Joshitha Ravishankar, Mansi Sharma · Journal of Electronic Imaging · 2025
Dynamic light fields offer an enhanced and immersive three-dimensional representation of a scene, making them ideal for computational multi-view autostereoscopic display applications. However, the large volume of light field videos results in high data rates with excess storage and transmission requirements. We present a framework that integrates joint aperture and pixel-wise exposure coding with a data-driven model to achieve efficient representation and compression of dynamic light fields. Optimized coding patterns are used to derive acquired images that capture the entire video content, and their inherent spatio-angular-temporal dynamics are effectively exploited using dynamic mode decomposition. Subsequently, High Efficiency Video Coding (HEVC) is employed to eliminate remaining intra-frame redundancies in the light field data while maintaining good reconstruction fidelity. The proposed scheme is the first of its kind to conceptualize light field videos as mathematical dynamical systems, leverage on dynamic modes of acquired images, and gain flexible coding at various bitrates. Experimental results demonstrate the competitive performance of our scheme over its other proposed variants and traditional HEVC-based methods, achieving notable gains in compression efficiency, bitrate savings, computational speed, bitstream size, and reconstruction quality.