Advances in Light Field Spatial Super-Resolution: A Comprehensive Literature Survey

Super-resolution reconstruction of light field images has recently become a central focus in the fields of computational photography and computer vision. We present a systematic review of 17 mainstream light field spatial super-resolution techniques, evaluating their performance across seven public datasets. Integrating experimental results, we specifically analyze the performance of deep learning-based super-resolution algorithms at various magnification levels. Although these models have made significant progress at lower magnifications (e.g., $2\times $ and $4\times $ ), current methods exhibit clear limitations at higher magnifications (e.g., $8\times $ and $16\times $ ), particularly in maintaining structural integrity and disparity consistency. Our experimental findings indicate substantial differences in robustness and adaptability among methods: approaches such as DistgSSR and DPT perform exceptionally well at high magnifications, while others, like HLFSR, exhibit comparatively poorer performance in complex scenes. Additionally, the unique characteristics of light field images add complexity to the super-resolution task. Future research should focus on enhancing the robustness, generalization, and capability of algorithms to handle complex scenarios. This review offers valuable direction for future research on light field image super-resolution and provides a solid foundation for its applications in virtual reality, augmented reality, and autonomous driving.