基于PCA的帧频提升算法

Algorithm for Frame Rate Up-conversion Based on PCA

提出基于主元分析(PCA)的帧频提升算法,首先由运动补偿得到初始内插图像并确定重构区域(重叠像素区域和空洞区域)。然后训练处理过程从前后帧的多分辨力图像中提炼训练块创建训练库和相应的PCA空间。最后应用海森矩阵值(HMDV)决定各个搜索块的重构填充顺序,逐块填充重构区域,从而解决内插图像的重叠和空洞问题。与传统的帧频提升算法相比,获得较好的内插图像主、客观质量。

A novel video frame rate up-conversion algorithm is proposed based on Principal Component Analysis （PCA）. First, the initial interpolated frame is obtained from motion compensated interpolation and identified the reconstructing regions including the overlapped pixels and holes. Then the training process is used to extract training patches from multi-resolution images of the previous frame and the following frame in order to create a training base with corresponding PCA space. Finally, the Hessian Matrix Decision Value （HMDV） is applied to determine a priority order of each search patch, and the matching patch is filled in the reconstructing regions patch by patch, thereby the overlapping problem and the holes problem are resolved. Compared with the conventional algorithm, the proposed algorithm provides better subjective and objective quality.