基于边缘引导的红外图像超分辨率算法研究毕业论文
2021-03-12 23:59:53
摘 要
目前,无论是军事、运输、自动化、执法、消防、科研、建筑还是安全等领域,红外图像成像技术都受到了广泛的应用。但是受限于目前的硬件成像水平,所得到的红外图像的分辨率往往不能满足人肉眼的需求使得红外图像的应用不能得到更进一步的扩展。
基于此现状,在本文中我们提出了一种新的算法用于单幅红外图像超分辨处理。即将红外图像的超分辨问题转化为边缘图像的超分辨率问题。首先提取红外图像的边缘图像,然后将边缘图像分割成多个小的像素块,然后利用马尔科夫随机场模型为每个低分辨率的边缘像素块找到对应的高分辨边缘像素块,接着将这些高分辨率边缘像素块整合起来,得到一张完整的超分辨边缘图像。最后,在超分辨率边缘图像的引导下,利用一个联合双边滤波器进行插值,这样我们得到的超分辨率红外图像在有较好的边缘保留的情况下,边缘也不会有明显的锯齿。
最后,将本文的算法同最近邻插值算法、双线性插值算法和双三次插值算法进行了超分辨率两倍的比较,本文算法的结果无论从视觉上还是评价指标(均方根误差(RMSE),结构相似性(SSIM))上都要优于其他三个算法,这也证明了本文算法的优越性。
关键词:红外图像;超分辨率;边缘提取;马尔科夫随机场;联合双边滤波插值
Abstract
At present, infrared image imaging technology has been widely used in the fields of military, transportation, automation, law enforcement, fire protection, scientific research, architecture and safety. However, the resolution of infrared images can not meet the needs of the human eye limited by the current level of hardware, which makes the application of infrared images can not be further expanded.
Based on this situation, we propose a new algorithm for super resolution processing of single infrared image. More specifically, the super-resolution problem of infrared image is transformed into the super-resolution problem of edge image. Firstly, We extracted the edge of the infrared image and then the edge image is divided into many small pixel blocks. Every small pixel blocks corresponded to a high resolution pixel blocks which can be find in dataset by using a Markov random field. and then a super-resolution edge image can be obtained by gathering the high resolution pixel blocks Finally, under the guidance of the super-resolution edge image, we use a Joint bilateral interpolation to get our super-resolution infrared image which has a good edge preserving, and the edges do not have significant jagged edges.
Finally, the algorithm of this paper is compared with the nearest neighbor interpolation algorithm, bilinear interpolation algorithm and bicubic interpolation algorithm. The results is superior to the other three algorithms in both visual and the evaluation such as root mean square error and structural similarity index, which also proves the superiority of this algorithm.
Keywords:IR image;Super resolution;Edge extraction;Markov random field;Joint bilateral interpolation
目录
摘要 1
Abstract II
第1章 绪论 1
1.1研究目的及意义 1
1.2图像超分辨技术的研究现状 2
1.3本论文的主要工作 3
第2章 红外图像超分辨率算法设计原理 4
2.1 滤波算法 4
2.1.1高斯滤波原理 5
2.1.2双边滤波原理 6
2.2边缘检测算法 7
2.2.1边缘检测算法综述 7
2.2.2Canny算子边缘检测 7
2.3边缘图像超分辨率算法 12
2.3.1马尔科夫随机场(MRF)基本理论 13
2.3.2马尔科夫随机场(MRF)在图像分割中的应用 15
2.3.3马尔科夫随机场(MRF)模型建立 17
2.4插值算法 18
2.4.1双三次插值算法 19
2.4.2联合双边滤波插值算法 20
2.5超分辨率图像质量评价指标 21
第3章 实验过程及结果比较 23
3.1实验过程 23
3.2实验结果比较 26
第4章 工作总结和展望 29
参考文献 30
附录A 31
致谢 37
第1章 绪论
1.1研究目的及意义
自然界一切物体的温度均比绝对零度要高,这样就不可避免地会产生红外辐射[1]。红外成像技术就是基于此而产生的。红外成像技术主要工作在7um-14um这个波段。这是由于用于红外成像的红外照相机对该波段非常敏感,可以很好的感应到环境中物体散发、传播和反射的红外辐射,这样外部光线就很难影响获取红外图像。
与可见光成像相比,红外图像具有如下优点: