【作者】 周欣；

【导师】 潘志庚；

【作者基本信息】 浙江大学 ， 计算机应用技术， 2006， 硕士

【摘要】 图像的精细分割技术是指把任意形状的前景物体从图像中分割出来的一种技术,如今它已经成为影视特效等多媒体制作中不可或缺的关键技术,因此该技术具有巨大的商业价值。但是,现有的分割技术难以同时满足人们在图像精细分割的速度和效果两个方面的需求,同时,它又难以对自然界存在的大量的颜色变化比较剧烈的复杂图像进行分割处理。本文针对现有的图像精细分割存在的问题,就如何提高分割速度和易用性两方面进行了研究,并介绍了作者在这两方面的解决方案。 首先,本文基于Poisson精细分割算法,提出并实现了基于图像的Poisson方程的快速数值解法——Multigrid Solver,并找到该算法的各个可调参数对算法效率和效果产生影响的规律。通过大量实验证明,该算法可以使得求解迭代快速收敛,在相同的求解误差下,求解时间将大大减少。 在图像精细分割的人机交互方式方面,本文提出并实现了两种基于Scribbling(随手绘制曲线段)的算法。相比传统的Trimap方式,这种方式简化了用户和计算机的交互。 此外,不论在传统的人机交互方式,还是Scriblbing的方式,算法实现上都需要进行一个关键步骤的判断——点在多边形内外的判断。本文提出了两种判断算法,分别可以在O(logn)和O(n)的时间复杂度下,解决点在凸多边形或普通多边形内外的问题。程序验证表明,新算法易于实现,具有运行速度快、稳定性高等优点。更多还原

【Abstract】 Digital image matting is to extract the foreground objects with arbitrary shapes from an image. It is an essential technique to make special effects in film and video production and has enormous potential commercial applications. The existing image matting techniques can’t satisfy the increasing needs for speed and effects of matting simultaneously. This paper proposes optimal and developed algorithms for image matting in the efficiency and user interaction.First, based on the Poisson Matting, we present and implement a fast numeric solver called Multigrid Solver for poisson equation on images. Besides, we also set up the strategies on adjusting the input parameters of the multigrid solver in order to obtain optimal results. Through lots of experiments, this method enables the iteration process to converge quickly, and the time cost by the algorithm will be reduced largely compared with other solvers.Second, in the aspect of image matting user interaction, this paper proposes two image matting algorithms with a new user interaction method called scribbling. Compared with the traditional interaction method "Trimap", our method, scribbling, will simplify the interaction between people and computers.Third, in the implementation of the matting algorithms, we have to employ a test called point-in-polygon test to judge whether the point is in the polygon. Our paper proposes two algorithms to respectively solve the point in-out convex polygon problem in O(logn) time complexity and the point in-out polygon problem in O(n). The two new algorithms are easy to be implemented and experimental results show the high robustness and efficiency of the new algorithm.更多还原