Abstract

This paper proposes fast and accurate frame rate up-conversion algorithm for both software and hardware. The proposed algorithm first construct Gaussian pyramid structure using 3×3 Gaussian mask. After constructing Gaussian pyramid, forward and backward unilateral motion estimation is implemented at the top level of the pyramid. Then, motion vector field that is obtained at the top level is projected onto the middle level and the bottom level of the pyramid with unilateral motion vector refinement. Next, on the bottom level, we implement motion vector smoothing and proposed motion vector shifting with bilateral motion vector refinement. By shifting unilateral motion vector to bilateral motion vector, we could avoid generating holes and overlapped regions. Finally, we select final MV between forward and backward MVs according to their reliability. After all the processes, we interpolate new frames by using overlapped block motion compensation. Experimental results show that proposed algorithm is up to 80 times faster than conventional algorithm with 0.1dB peak signal-to-noise gain in average.

Highlights

  • Frame rate up-conversion (FRUC) is the technique that interpolates virtual frame which exists in the middle of two original frames’ time interval

  • FRUC was implemented by copying original frames

  • MCFRUC is composed of two parts, motion estimation (ME) and motion compensated interpolation (MCI) [1]

Read more

Summary

Introduction

Frame rate up-conversion (FRUC) is the technique that interpolates virtual frame which exists in the middle of two original frames’ time interval. When interpolating frames, blocking artifact occurs because of BMA To solve this problem, overlapped block motion compensation (OBMC) was proposed [3]. Miss match between unilateral MV and bilateral MV caused loss of quality despite of their refinement process To overcome their weakness, Yoo et al proposed direction-select ME (DSME) which use both forward and backward unilateral MV as an initial MV to estimate accurate MV [8]. They couldn’t propose the solution to minimize error that occurs when shifting unilateral MV to bilateral MV.

Proposed algorithm
Gaussian pyramid
Forward and backward motion estimation
Motion vector projection
Weighted motion vector smoothing
Motion vector shifting
Motion vector selection
Experimental results
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call