针对传统的空域超分辨率重建算法过度依赖配准的精度而限制了算法的适用范围，提出基于自适应运动估计的视频超分辨率重建算法。在非参数估计的核回归模型的基础上，将二维核回归函数扩展到三维，把视频序列中的每一个像素点表示成三维泰勒展开式。通过局部加权最小二乘法得到展示的系数，用核回归的权重捕捉时空局部运动信息，避免了显式的亚像素精度的运动估计。在标准测试视频数据库中的实验表明，该算法具有较好的重建效果和较大适用范围，可用于具有局部和复杂运动的视频。%In view of the traditional airspace super-resolution reconstruction algorithm overly dependent on the accuracy of registration which leads to the problem of limiting the scope of application of the algorithm,a video super-resolution reconstruction algorithm is proposed based on self-adaptive motion estimation.On the basis of the kernel regression model of non-parametric estimation,the two-dimensional kernel regression was extended to 3D and each pixel of video sequence was represented into three Taylor expansion.The coefficients were obtained by local weighted least squares method.The temporal local motion information was captured with kernel regression weights and explicit sub-pixel precision motion estimation was avoided.Experiments in the standard test video database show that the proposed algorithm has better reconstruction effect and wider application range,which is able to be applied to videos with local and complex motion.
展开▼
