A New Objective Quality Metric for Frame Interpolation Used in Video Compression

Abstract

A new metric for assessing spatial quality of the temporally interpolated frames is proposed. It is designed to overcome disadvantages found in several widely used metrics for the frame interpolation application. The proposed metric quantifies the severities of several pronounced interpolation artifacts and combines them along with human visual factors to form a global (i.e. frame-wide) quality measurement. Also, the artifact amounts are used to determine additional quality impairment caused by locally aggregated distortion. Finally, these two measurements are combined into a single quality score. Subjective experiments show that the proposed metric significantly outperforms other commonly used metrics.