Accurate Frame Rate Up-Conversion for Advanced Visual Quality

Abstract

Motion-compensated frame rate up-conversion (FRUC) is a method for providing better video image quality than non-motion-based methods. However, incorrect motion vector (MV) prediction can lead to continued induction of debris and artifacts. In this paper, a motion compensated FRUC algorithm is proposed, which is able to acquire more accurate MVs by referencing neighboring frames with hierarchical motion filed construction block size and an MV mapping stage. The block size of hierarchical MV filed construction has been enlarged in order to acquire more motion information, and the size of the MV mapping and image interpolation stage has been reduced to obtain a more accurate interpolated result. In order to smooth the MV, a median filter is considered with sum of absolute difference value and constructed with an adaptive MV smoothing method. For the purpose of suppressing the debris, a noise removal filter is applied after the interpolation operation. According to the experimental result, the proposed approach is able to improve visual quality performance by 0.82 dB on average. Subjective analysis shows that annoying artifacts are significantly reduced.