Abstract
As High Frame Rate (HFR) video becomes standard, converting these sources to legacy lower frame rates (e.g., 24 or 30 fps) remains a challenge. Traditional methods often suffer from temporal aliasing (judder) or ghosting artifacts. This paper presents a novel down-sampling framework that utilizes Motion Compensated Frame Interpolation (MCFI) to simulate a "virtual shutter angle." By synthesizing intermediate frames to fill temporal gaps and applying a Gaussian weighting scheme, the proposed method achieves smooth, filmic motion blur while significantly reducing artifacts. Subjective evaluations demonstrate a strong viewer preference for this approach, particularly at very low frame rates.
Introduction
The transition to HFR video (60 fps and above) offers superior motion clarity but creates compatibility issues with legacy display standards and cinematic preferences. Simply dropping frames to reach 24 fps results in "judder," while simple frame averaging creates "ghosting"—where moving objects appear as discrete, overlapping images. These artifacts stem from the "dark" periods in the original capture where the shutter was closed, creating gaps in the temporal information.
Proposed Methodology
The Virtual Shutter Angle
The core of the proposed method is the concept of a virtual shutter angle. Unlike traditional averaging which only uses existing frames, this approach uses MCFI to generate synthetic intermediate frames that fill the gaps between original captures.
By combining original and synthesized frames, the system can simulate a continuous exposure over any desired shutter angle (up to 360°). This ensures a continuous motion path, effectively eliminating the spatial gaps that cause ghosting.
Hierarchical Prediction and Weighting
To manage computational complexity, the framework employs a hierarchical prediction structure. This allows the system to use previously predicted frames as references for higher layers, narrowing the search range for motion vectors. Furthermore, instead of a standard box filter (equal weighting), a Gaussian weighting scheme is applied to the frame sequence. This more accurately mimics the physical light integration of a camera sensor, providing a more natural-looking motion blur that balances detail preservation with aliasing suppression.
Experimental Results
The method was evaluated using 4K sequences from the Ultra Video Group (UVG) database, originally captured at 60 fps with a 180° shutter.
Subjective tests involving 18 participants using a double-stimulus methodology revealed that the proposed method significantly outperforms traditional frame averaging. At a target rate of 15 fps, over 80% of participants preferred the proposed method. Statistical analysis (one-tailed t-test) confirmed these results with high significance ($p \approx 0$).
Conclusion
By integrating motion-compensated interpolation with a virtual shutter model, this method provides a robust solution for high-quality frame rate conversion. While computationally more intensive than simple averaging, the significant gains in visual quality—particularly the elimination of ghosting—make it a compelling choice for professional video processing and archival conversion. Future work will explore the integration of deep-learning-based interpolation to further enhance performance in complex scenes.
Summary updated