摘要
图像插帧,是在输入两幅图像a、b间,产生一系列中间帧的过程。提出一种针对近似图像进行图像插帧的快速算法。该算法对Mahajan提出的基于路径的插值算法进行了优化,能更快速产生视觉连贯的中间图像,不需要用户人工干预。算法使用更快速直接的dinic算法来使全局能量最小化,计算转变点和转变路径,并且在三维泊松重建部分使用一种能够快速收敛、基于对角线的多重网格算法。实验结果证明,算法能快速准确生成视觉效果连贯的中间图像。
Image interpolation is the process that produces a series of intermediate frames between two input images A and B.In this paper,a quick algorithm of image interpolation for approximate images is presented.The algorithm optimises the path-based interpolation of Mahajan,which can generate successive intermediate frames more quickly without manual intervention from users.This method uses dinic,a quicker and more direct algorithm,to minimise the global energy and compute the transition point and the transition path.And in 3D Poisson reconstruction,a multi-grid solution using diagonally oriented grids which can converge rapidly is introduced.Experimental results show that the improved algorithm can quickly generate successive intermediate frames.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第11期148-152,共5页
Computer Applications and Software
基金
上海市科委项目(10511516005
10dz1500107)
高等学校博士学科点专项科研基金项目(20100071120033)
关键词
图像插值
能量最小化dinic泊松重建
Image interpolation Energy minimisation dinic poisson reconstruction
