High-Order Model and Dynamic Filtering for Frame Rate Up-Conversion.

Area of Science:

• Computer Vision
• Image Processing
• Video Enhancement

Background:

• Traditional frame rate up-conversion methods rely on simplified assumptions of constant brightness and linear motion.
• These assumptions limit the accuracy and visual quality of interpolated video frames.

Purpose of the Study:

• To develop a novel frame rate up-conversion method (HOMDF) that overcomes limitations of traditional approaches.
• To enhance the visual quality of interpolated video frames through advanced modeling and filtering techniques.

Main Methods:

• Proposed a high-order model and dynamic filtering (HOMDF) approach for video pixels.
• Modeled pixel intensity and position using high-order polynomials over time.
• Estimated polynomial coefficients by minimizing auto-regressive prediction and reconstruction errors.
• Implemented dynamic filtering for efficient coefficient estimation, leveraging temporal coherence.
• Utilized motion-compensated interpolation based on estimated pixel variations and motion trajectories.

Main Results:

• HOMDF achieved superior visual quality in interpolated frames compared to state-of-the-art methods.
• Experimental results on natural and synthesized videos demonstrated significant improvements in both subjective and objective comparisons.
• The method accurately captures pixel intensity variations, velocity, and acceleration.

Conclusions:

• The proposed HOMDF method offers a significant advancement in frame rate up-conversion.
• The high-order modeling and dynamic filtering effectively enhance video interpolation quality.
• HOMDF provides a robust solution for generating high-fidelity interpolated video frames.