Bidirectional Optical Flow-Guided Spatio-Temporal Super-Resolution Reconstruction for Integral Imaging

Integral imaging， recognized as one of the most promising autostereoscopic 3D video display technologies， continues to encounter a fundamental trade-off between frame rate and spatial resolution. This necessitates a trade-off between spatial clarity and temporal smoothness during dynamic video generation， thereby compromising the final visual effect and user experience. To address this limitation， this study introduces a bidirectional optical flow-guided spatio-temporal resolution enhancement framework， which enables dynamic 3D display of integral imaging with simultaneously high spatial resolution and frame rate. Inter-frame motion information is robustly estimated using bidirectional optical flow techniques， combined with a bidirectional optical flow-guided feature alignment module， thereby effectively mitigating boundary artifacts caused by parallax discontinuities in conventional methods， facilitating precise cross-frame fusion， and preserving fine-grained spatial details and parallax coherence. Additionally， the spatio-temporal super-resolution reconstruction module performs high-fidelity image reconstruction， substantially enhancing the spatio-temporal resolution of integral imaging videos， thereby improving overall visual quality. Experimental results demonstrate that the method significantly outperforms existing approaches， achieving an average improvement of 8.5 dB in Peak Signal-to-Noise Ratio （PSNR）， a 15% increase in Structural Similarity Index Measure （SSIM）， and a 70% reduction in Learned Perceptual Image Patch Similarity （LPIPS） compared with state-of-the-art 2D and 3D methods. Meanwhile， in terms of the PSNR on Epipolar Plane Image （EPI-PSNR） and the SSIM on Epipolar Plane Image （EPI-SSIM）， which characterize the 3D structure preservation capacity， the proposed method achieves average improvements of 12.77 dB and 35.4%， respectively， thereby demonstrating its superior capability to preserve disparity continuity and angular consistency. For the reconstructed video sequences， under the Motion-based Video Integrity Evaluation （MOVIE） metric， the proposed method attains an average score of 0.850， which is substantially higher than that of other methods， indicating perceptual quality more closely approximating real-world imagery and enhanced temporal stability. These results collectively demonstrate that the proposed approach provides significant advantages in improving video reconstruction quality， mitigating motion artifacts， and strengthening temporal consistency， thereby offering a robust and effective technical framework for dynamic 3D display applications based on integral imaging.