全自动抠图技术的研究

发布时间：2018-02-05 00:23

本文关键词： 单图像光流自动抠图景深　出处：《山东师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：抠图是一种精确提取任意图像或视频中前景物体的图像处理技术。该技术最早可以追溯到19世纪50年代的光学抠图,现在随着计算机工业的迅猛发展,数字抠图取代了传统的光学抠图并成为当今计算机视觉领域的一个重要课题。由于数字抠图模型存在欠约束性,现有的抠图算法往往需要用户事先标记出图像的前景和背景部分。而这些人为的标记再增加了操作的复杂程度的同时,也会影响抠图的精度。针对上述问题,本文提出了一种新的基于单幅图像实现自动抠图的方法。本文主要做了以下工作:1、详细介绍了常见的抠图算法。本文将现有的抠图算法分为两类,交互式抠图算法和自动数字抠图算法。交互式抠图算法中主要介绍了贝叶斯抠图、Graph Cuts、Easy matting和全封闭抠图等经典算法,并对基于学习的抠图方法进行了详细分析;自动数字抠图算法中主要介绍了Flash matting、光谱抠图算法、自动三分图生成算法和自动肖像抠图,并分别说明了它们的优点和不足。2、详细介绍了光流算法:分别介绍了HS法和LK法及其改进算法,并对上述算法进行了分析。3、针对传统抠图方法的缺点,并结合现有的光学原理及光流机制,本文提出了基于单幅图像的自动抠图技术。该技术系统搭建简单,仅需要一面平面镜和一只相机。通过该系统获取所需图片,并利用光学分析来得到一个景深计算等式。然后使用改进的LK-HS法对图像中的实景和虚景部分进行光流计算,同时结合景深计算等式来获得目标场景的景深图,再根据此景深图获得三分图;最后,将其与现有的抠图算法进行计算生成a值,从而实现自动抠图。
[Abstract]:Matting is an image processing technique for extracting foreground objects from any image or video accurately. The technology can be traced back to optical matting in 1850s. Now with the rapid development of computer industry. Digital matting has replaced the traditional optical matting and has become an important subject in the field of computer vision. The existing matting algorithms often require users to mark out the foreground and background of the image in advance. These artificial markers not only increase the complexity of operation, but also affect the precision of matting. This paper presents a new method to realize automatic matting based on single image. In this paper, we mainly do the following work: 1, introduce common matting algorithms in detail. In this paper, we divide the existing matting algorithms into two categories. Interactive matting algorithm and automatic digital matting algorithm. The classical algorithms such as Easy matting and fully closed matting are analyzed in detail, and the learning-based matting method is analyzed in detail. The automatic digital matting algorithm, spectral matting algorithm, automatic three-point algorithm and automatic portrait matting algorithm are mainly introduced, and their advantages and disadvantages are explained respectively. The optical flow algorithm is introduced in detail: HS method, LK method and their improved algorithm are introduced, and the above algorithms are analyzed. 3, aiming at the shortcomings of traditional matting method. Combined with the existing optical principle and optical flow mechanism, this paper proposes an automatic matting technology based on single image, which is simple to build. Only a plane mirror and a camera are needed. Get the desired picture through the system. An equation of depth of field is obtained by optical analysis, and then the improved LK-HS method is used to calculate the light flow in the real and virtual parts of the image. At the same time, the depth of field map of the target scene is obtained by combining the depth of field equation, and then the three-point map is obtained according to the depth of field map. Finally, it is calculated with the existing matting algorithm to generate a value, thus realizing automatic matting.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】