Video de-interlacing method based-on optical flow

This paper proposes to improve the accuracy of motion vectors for video de-interlacing by selectively using optical flow results to assist the traditional block-based motion estimation used in video codec. Firstly, a framework for performance evaluation is proposed so that the subjective performance of de-interlacing will not matter in applications such as video compression. Then pre-filtering is performed to remove potential noise, and a hierarchical structure is introduced for same-parity field pictures or opposite-parity field pictures to get more reliable optical flow results for compression. Finally the optical flow results are used to improve the accuracy of block-based methods for each partition of the to-be-interpolated image and the final motion vectors are determined. Experimental results show that the proposed solution can achieve both better peak signal-to-noise ratio (PSNR) performance for video de-interlacing and higher coding efficiency for interlaced videos.