A no-reference quality metric for evaluating deinterlaced video frames

摘要

Video fields are commonly labeled "odd" and "even", depending on the order of field. Odd fields contain scan lines that correspond to the odd lines of the video frame. Even fields contain scan lines that correspond to the even lines of the video frames. One odd field and one even field are used to create a video frame. Deinterlacing algorithms convert video from the interlaced scan format to the progressive scan format. This paper presents a deinterlace metric which evaluates the performance of a deinterlacmg algorithm. The metric is based on a frame processed with a deinterlacing algorithm. The described approach is a no-reference metric, meaning that the quality measure is not dependent on previously deinterlaced frames. In previous literature, the mean squared error (MSE) is frequently used as a performance metric for deinterlacing. MSE does not necessarily correlate with the effectiveness of the deinterlacing. Rather than using MSE, our metric is based on high frequency components of deinterlaced frame. We found that the metric corresponds well with subjective testing and is therefore suitable for quick qualitative characterization of deinterlaced frames.