差异图像驱动的多聚焦图像融合性能提升研究

发布时间：2018-03-27 13:25

  本文选题：多聚焦图像融合　切入点：多尺度邻域技术　出处：《昆明理工大学》2017年硕士论文

【摘要】：由于光学镜头景深具有的局限性,大部分的透镜成像设备在获取场景图像时无法使所有物体都清晰成像。但是,在某些情况下,人们往往需要得到一幅在一个场景中所有物体都聚焦的清晰图像。在上述背景下,多聚焦图像融合技术引起了学者们广泛的关注。随着科学技术的飞速发展,多聚焦图像融合已经在医学成像,计算机视觉,模式识别技术,军事图像等一系列重要的任务中发挥着举足轻重的作用。多聚焦图像融合算法大致分为两种方法:基于变换域的融合法和空间域融合法。基于空间域的方法大致分为为基于块、基于区域和基于聚焦区域检测的方法。基于聚焦区域检测的方法最为受到学者的关注,但是由于图像的复杂性,在聚焦区域和散焦区域之间,由于许多图像没有明确的纹理和明确的边界,聚焦区域的边界会存在一条明显的边界接缝。在基于变换域的方法中,最具有代表性的方法就是多尺度融合方法。在多尺度变换对多聚焦图像进行融合过程中发现,不同子带的融合规则是影响图像融合效果的一个关键性的因素。但是,近年来研究者提出的新的融合规则下对图像进行融合,融合的结果效果并没有显著的提升。为了解决上述问题,本文主要贡献如下:(1)提出了基于多尺度、多方向领域距离(MMND)的两种融合规则,这两种融合规则都分别实现了对多聚焦图像融合结果性能的提升。提出此方法的理论依据是:低质量融合图像和源图像的差异比高质量融合图像和源图像的差异更加明显。(2)基于上述理论依据,在优化融合图像过程中,源图像上的像素点被分为三种,聚焦区域像素点、离焦区域像素点,和过渡区域像素点。基于源图像的三种像素点,就可以对融合图像的结果在基于MMND变换域融合规则与MMND空间域融合规则上通过差异图构造出的决策图进行图像的优化提升。本文经过大量的实验结果证明,本文提出的两个融合规则比一些近年来提出的先进的算法表现更加优异。
[Abstract]:Because of the limitations of the depth of field of the optical lens, most lens imaging devices are unable to get a clear image of all objects when they get the scene image. However, in some cases, People often need to get a clear image in which all objects are focused in one scene. In the above background, multi-focus image fusion technology has attracted wide attention of scholars. With the rapid development of science and technology, Multi-focus image fusion has been used in medical imaging, computer vision, pattern recognition technology, The multi-focus image fusion algorithm is divided into two kinds of methods: the fusion method based on transform domain and the method based on space domain, and the method based on space domain is roughly divided into blocks. The methods based on region and focus region are the most concerned by scholars, but because of the complexity of image, they are between the focus region and defocus region. Because many images have no clear texture and boundary, there is an obvious boundary seam in the boundary of the focus region. The most representative method is multi-scale fusion. In the process of multi-scale image fusion with multi-scale transformation, it is found that the fusion rules of different sub-bands are a key factor affecting the image fusion effect. In recent years, the new fusion rules proposed by researchers have not significantly improved the result of image fusion. In order to solve the above problems, the main contributions of this paper are as follows: 1) based on multi-scale. Two kinds of fusion rules for multidirectional domain distance (MMND), These two fusion rules can improve the performance of multi-focus image fusion respectively. The theoretical basis of this method is that the difference between low quality fusion image and source image is higher than that between high quality fusion image and source image. Based on the above theoretical basis, In the process of optimizing the fusion image, the pixel points on the source image are divided into three types: the focus region pixel, the defocus region pixel point, and the transition region pixel point, which are based on the three pixel points of the source image. The result of image fusion can be optimized by using the decision graph constructed by difference graph on the basis of MMND transform domain fusion rule and MMND space domain fusion rule. The two fusion rules proposed in this paper are better than some advanced algorithms proposed in recent years.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】

中国期刊全文数据库 前2条

1 屈小波;闫敬文;肖弘智;朱自谦;;非降采样Contourlet域内空间频率激励的PCNN图像融合算法(英文)[J];自动化学报;2008年12期

2 刘从义;敬忠良;肖刚;杨波;;Feature-based fusion of infrared and visible dynamic images using target detection[J];Chinese Optics Letters;2007年05期

中国博士学位论文全文数据库 前2条

1 赵恒军;图像多尺度邻域距离分解及其应用[D];重庆大学;2013年

2 杨波;基于小波的像素级图像融合算法研究[D];上海交通大学;2008年

，

本文编号：1671696