Low Complex Block Level Correlation and Registration for Video Frame Interpolation

Abstract

In this paper, we propose a low complex block level correlation and registration for Video Frame Interpolation (VFI) algorithm. The proposed method finds the correlation between the corresponding non-overlapping blocks. The correlation coefficients indicate the amount of similarity between the blocks. Using the median of correlation coefficients as threshold, the blocks are blended using image registration process. Compared to existing block based Motion Compensated Frame Interpolation (MCFI) methods, proposed algorithm is a low complex algorithm. In the proposed, there is no need of Motion Estimation (ME), smoothing of unreliable Motion Vectors (MVs) and Motion Prediction. Extensive experiments are conducted on various benchmark video sequences. Subjective assessment shows that the artefacts like blocky artefacts, holes, blurry artefacts, and ghost effects in MCFI are overcome by this algorithm, as there is no ME, preserving the true motion of the object to a greater extent is possible. The metrics PSNR (Peak Signal to Noise Ratio) and SSIM (Structural Similarity Index Metric) provide the objective assessment.