结合Kalman滤波和均值偏移的多视点视差估计

MULTI-VIEW DISPARITY ESTIMATION COMBINING KALMAN FILTERING AND MEAN SHIFT

针对多视点视频编码中视差估计运算量大的问题,提出一种用Kalman滤波器预测运动矢量和Mean Shift算法视差匹配的视差估计方法。对宏块的运动模型建立Kalman滤波器,预测了宏块的运动矢量。分析视差矢量和运动矢量的几何关系进而确定宏块预测视差矢量。计算得到的预测视差矢量为下一步的视差匹配提供了初始搜索位置,用Mean Shift迭代计算完成宏块的最佳视差匹配。实验表明,该方法和全搜索方法相比,在率失真性能几乎不变的情况下编码时间减少了93%以上。与其他方法相比,率失真性能和编码效率获得提高。

Aiming at the large amount of disparity estimation in multi-view video coding,a method of Kalman filtering to predict disparity matching between motion vector and Mean Shift algorithm was proposed. A Kalman filtering was built on the macroblock motion model to predict the macroblock motion vector. The geometric relationship between the disparity vector and the motion vector was analyzed to determine the macroblock prediction disparity vector. The calculated predictive disparity vector provided the initial search position for the next disparity matching,and the best parallax matching of the macroblock was completed by Mean Shift iteration calculation. Experiments show that compared with the full search method,the encoding time is reduced by more than 93% when the rate-distortion performance is almost constant. Compared with other methods,the rate-distortion performance and coding efficiency are improved.