摘要
本文提出了一种运动自适应的帧内插算法,可以用于逐行扫描的数字视频系统中做帧速率转换。该算法首先采用了线性预测块匹配搜索算法来估计内插帧中每一个象素的运动矢量;然后对得到的运动场进行中值滤波和判别,剔除运动不一致或错误的运动矢量;最后沿着运动轨迹线性内插得到内插帧的象素值。仿真分析表明本文提出的算法,在一个大的搜索区域进行快速搜索和判别,可以有效地解决内插帧中快速运动目标的运动模糊和有切换帧时的图象重叠,提高了帧速率转换中内插帧的主观视觉质量。
A motion adaptive frame interpolation algorithm for progressive frame rate conversion of digital video systems is proposed. Firstly a motion vector is searched for each pixel in the interpolated frame using the block-matching search method of linear prediction. Then the resulting motion field is median filtered and differentiated to remove inconsistent or wrong vectors. Finally averaging along the motion trajectory is used to get the interpolated pixel values. Simulation results indicate the proposed method with fast search and differentiation in a large area can solve the motion blurring of fast moving object and the scene overlapping of a scene change in the interpolated frame, and the corresponding objective quality of the interpolated frame is also improved.
基金
国家攀登计划(神经网络)重大项目
东南大学与菲立浦公司消费电子领域合作项目资助
