Spatial-Temporal Correlation Based Multi-resolution Hybrid Motion Estimation for Frame Rate Up-Conversion

摘要

This paper presents a method applied in frame rate up-conversion (FRUC) to estimate the motion vector (MV) based on spatial-temporal motion vectors, which have a high correlation with the interpolated block. Meanwhile, a hierarchical structure is employed to generate a series of low-resolution images for motion estimation (ME) aimed to reduce computation complexity. After using ME in the lowest resolution image, there are several relatively accurate MVs picked to transmit down to the next level for further calculation. Then the initial motion vector field (MVF) is generated by calculating the sums of absolute difference (SAD) of the candidate MVs. To refine the MVF, an outliers detecting criterion is set up to smooth the MVF and obtain a more accurate MVF at the same time. The motion compensated (MC) will not perform until the termination of MVF refinement level-by-level. Experimental reveals that the proposed method find a tradeoff between accuracy and complexity, because the peak signal-to-noise ratio (PSNR) is promoted to 4.69 dB and the processing time is quite fast.