Hierarchical Extended Bilateral Motion Estimation-Based Frame Rate Upconversion Using Learning-Based Linear Mapping.

Area of Science:

• Computer Vision
• Image Processing
• Video Enhancement

Background:

• Frame rate upconversion (FRUC) is crucial for enhancing video smoothness.
• Existing methods often struggle with motion estimation accuracy and blur reduction.
• The need for efficient and effective FRUC techniques remains high.

Purpose of the Study:

• To develop a novel learning-based frame rate upconversion (FRUC) scheme.
• To improve motion estimation accuracy and frame interpolation quality.
• To achieve superior performance compared to existing FRUC methods.

Main Methods:

• A hierarchical extended bilateral motion estimation (HEBME) method for accurate motion vector prediction.
• A light-weight motion deblur (LWMD) method utilizing linear mapping for blur reduction.
• A synthesis-based motion-compensated frame interpolation (S-MCFI) method for generating high-quality interpolated frames.

Main Results:

• The proposed HEBME method significantly enhances motion estimation accuracy.
• The integrated FRUC scheme achieves an average of 1.42 dB higher peak signal-to-noise-ratio than state-of-the-art non-deep learning methods.
• Comparable performance is demonstrated against state-of-the-art deep learning-based schemes.

Conclusions:

• The developed learning-based FRUC scheme offers a significant advancement in video quality.
• The combination of HEBME, LWMD, and S-MCFI provides precise frame interpolation and contrast enhancement.
• This approach represents an effective solution for high-performance frame rate upconversion.