面向运动分割优化的变分光流计算方法研究

面向运动分割的需求,围绕变分光流计算中的运动边缘保留问题,对反应-扩散式光流计算模型中的扩散张量设计与改进方法进行系统深入研究。 在分析已有设计方法的基础上,提出了一种融合图像和流场信息驱动的扩散张量设计方法,并利用该扩散张量建立光流计算模型,然后在多尺度计算框架下给出了相应的偏微分方程数值计算方法。 理论分析与对比实验结果表明,这种设计方法能有效克服现有方法提取运动边缘不精确的缺点,更准确地刻画出运动边缘,提高光流的计算精度。

Variational optical flow computation towards motion segmentation

Aiming at the problem of preserving the discontinuities in optical flow field for motion segmentation, a meaningful research has been done. The key of the research focuses on the improvement of the diffusion tensor in reaction-diffusion model. By analyzing the existing methods, we proposed a new method which design diffusion tensor by joint image- & flow- driven. Furthermore an efficient coarse-to-fine numerical scheme about the partial differential equation has been present at the same time. our experiments proved that our proposed approach can effectively overcome the motion blur problem of other methods while and preserving the discontinuities in the optic flow field more accurately.