数字图像抠图算法研究

发布时间：2018-04-24 03:21

本文选题：图像处理 + 自然图像抠图　；参考：《电子科技大学》2012年硕士论文

【摘要】：数字图像抠图是一种对任意自然图像中指定前景对象进行精确提取的图像处理技术。例如在电影制作中，导演可以利用抠图技术将演员无缝地转移到任意需要的位置。数字抠图技术已经成为视觉效果制作的关键技术，广泛运用在杂志设计，平面艺术，广告，影视后期制作等领域。抠图问题提出至今已有几十年时间。早期的抠图技术以光学方法为主，直到最近十几年随着计算机和多媒体技术的飞速发展，基于数字的抠图方法才被学者广泛研究。传统的抠图方法往往被限定在解决某个特定物理环境下的抠图问题，，这极大限制了抠图技术的使用范围。针对自然图像的抠图方法没有抠图环境限制，能对任意自然图像进行抠图处理，具有很广的适用范围，但该类方法往往计算复杂度高且抠图准确性难以保证，因此还处于研究和发展阶段。本文针对自然图像抠图技术存在的上述问题，进行了下面的研究工作： 1．系统研究了蓝屏抠图、自然图像抠图、环境抠图，阴影抠图及视频抠图的基本原理和相关方法。实验了鲁棒抠图，贝叶斯抠图和闭合场抠图三种经典的抠图算法。 2．研究了基于样本的前景移除方法，并将其运用到抠图算法中。设计出一种基于背景还原先验假设的抠图算法，该算法能够在不影响抠图准确性的同时，提高抠图效率。该方法还能得到图像在指定前景对象移除后的背景还原图。实验表明，基于背景还原先验的抠图算法，能够准确地提取前景对象，具有良好的抠图效率。 3．研究了样本颜色采样与图像中前景对象边缘梯度间的关系，设计出基于图像纹理方向采样的抠图算法，并结合颜色线模型对前/背景颜色样本进行采集和优化。实验表明，这种基于纹理方向先验信息的抠图方法，具有良好的抠图准确性。
[Abstract]:Digital image matting is an image processing technique for extracting the specified foreground objects from any natural image accurately. In film-making, for example, directors can seamlessly move actors to any desired location using matting techniques. Digital matting technology has become the key technology of visual effect production, widely used in magazine design, graphic art, advertising, film and television post production and other fields. Matting problem has been put forward for decades. In the early years, matting technology was mainly optical. With the rapid development of computer and multimedia technology in recent years, the matting method based on digital has not been widely studied by scholars. Traditional matting methods are often limited to solve the matting problem in a specific physical environment, which greatly limits the scope of application of matting technology. The matting method of natural image has no matting environment restriction, it can process any natural image and has a wide range of application, but this kind of method often has high computational complexity and the accuracy of matting is difficult to guarantee. Therefore, it is still in the stage of research and development. Aiming at the above problems in natural image matting technology, the following research work is carried out in this paper: 1. The basic principles and related methods of blue screen matting, natural image matting, environment matting, shadow matting and video matting are systematically studied. Three classical matting algorithms, robust matting, Bayesian matting and closed field matting, are tested. 2. The foreground removal method based on sample is studied and applied to matting algorithm. A matting algorithm based on background restoration priori hypothesis is designed, which can improve matting efficiency without affecting matting accuracy. The method can also obtain the background restore image after the specified foreground object is removed. Experiments show that the priori matting algorithm based on background reduction can extract foreground objects accurately and has good matting efficiency. 3. The relationship between sample color sampling and edge gradient of foreground object in image is studied. A matting algorithm based on texture direction sampling is designed, and the color line model is used to collect and optimize the front / background color samples. Experiments show that this matting method based on texture orientation priori information has good matting accuracy.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41

【参考文献】