复数轮廓波单幅图像超分辨率算法设计
发布时间：2018-09-07 20:50
【摘要】：数字图像的超分辨率分析可以广泛应用于航空航天、精确制导与精确打击、农作物疾病诊断、自然灾害预报、身份识别与认证等诸多领域。根据低分辨率图像的来源类型,可以将数字图像的超分辨率分析分为两类:一类是多幅图像(或者称为序列图像)的超分辨率算法,另一类是单幅图像的超分辨率算法。本文研究单幅图像的超分辨率算法。从数字图像超分辨率的基本概念出发,将低分辨率需要内插的图像分为光滑部分和非光滑部分,针对两个部分研究了单幅数字图像的超分辨率算法。第一个部分针对光滑连续函数模型,探讨了空域内插的方法,主要包括:最近邻内插法、双线性内插法、双三次内插法、双三次样条法和二维直接内插法。实验结果表明,二维直接内插法明显优于前面四种内插算法,在相同复杂度的情况下,可以得到最优的超分辨率结果;而最近邻法在所有情况下表现出最差的性能,出现最严重的马赛克现象;双三次内插法和双三次样条法则优于最近邻法和双线性内插法。第二个部分的内插算法,采用了复数轮廓波变换。到目前为止,复数轮廓波变换主要有两种类型,即固定冗余度复数轮廓波变换和可变冗余度复数轮廓波变换。固定冗余度复数轮廓波变换是双树复数轮廓波变换,可以认为是基本轮廓波变换的实数部分和虚数部分;而冗余度可变的复数轮廓波变换则是基于映射的架构,可以根据情况选取不同的冗余度。由于后者具有较高的灵活性和重组性质,论文中选取其作为变换工具。文章阐述了这种复数轮廓波变换的实现方法,探讨了该种变换的基本性质,特别重点讨论了该变换的移不变水平,在该变换的基础上,设计了一种性能良好的单幅图像内插算法。该算法充分应用了复数轮廓波变换的移不变性质,将低分辨率需要内插图像的非光滑部分,在复数轮廓波域,对变换域进行内插,然后进行反变换,并在此基础上,将本算法和二维直接内插算法融合,得到比较满意的实验结果。实验结果表明,所提出的算法能够更好的实现单幅图像的超分辨率重建,在客观参数(例如峰值信噪比等)和主观效果两个方面,都有比较良好的表现。
[Abstract]:Super-resolution analysis of digital images can be widely used in aerospace, precision guidance and precision strike, crop disease diagnosis, natural disaster prediction, identification and authentication, and so on. According to the source type of low-resolution image, the super-resolution analysis of digital image can be divided into two categories: one is super-resolution algorithm for multiple images (or sequence image), the other is super-resolution algorithm for single image. In this paper, the super-resolution algorithm of single image is studied. Based on the basic concept of super-resolution of digital image, the image needed to be interpolated with low resolution is divided into smooth part and non-smooth part. The super-resolution algorithm of single digital image is studied for the two parts. In the first part, for the smooth continuous function model, the spatial interpolation methods are discussed, including nearest neighbor interpolation, bilinear interpolation, bicubic spline and two-dimensional direct interpolation. The experimental results show that the two-dimensional direct interpolation algorithm is superior to the previous four interpolation algorithms, and the optimal super-resolution results can be obtained under the same complexity, while the nearest neighbor method has the worst performance in all cases. The bicubic interpolation method and the bicubic spline rule are superior to the nearest neighbor method and the bilinear interpolation method. The second part of the interpolation algorithm, using the complex contour wave transform. Up to now, there are two main types of complex contour wave transform, that is, constant redundancy complex contour wave transform and variable redundancy complex contour wave transform. The constant redundancy complex contour wave transform is a double tree complex contour wave transform, which can be considered as the real and imaginary parts of the basic contour wave transformation, while the complex contour wave transformation with variable redundancy is based on the mapping framework. Different redundancy can be selected according to the situation. Because of its high flexibility and recombination, the latter is chosen as a transformation tool in this paper. In this paper, the realization method of the complex contour wave transform is expounded, the basic properties of the transformation are discussed, especially the invariant level of the transformation is discussed, on the basis of the transformation, A single image interpolation algorithm with good performance is designed. The algorithm makes full use of the shift invariant property of complex contour wave transform. The low resolution needs to interpolate the non-smooth part of the image. In the complex contour wave domain, the transform domain is interpolated, then the transform domain is inversely transformed. The proposed algorithm is fused with the two-dimensional direct interpolation algorithm and the experimental results are satisfactory. The experimental results show that the proposed algorithm can better realize the super-resolution reconstruction of a single image, and has good performance in both objective parameters (such as peak signal-to-noise ratio) and subjective effect.
【学位授予单位】：信阳师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2229378
[Abstract]:Super-resolution analysis of digital images can be widely used in aerospace, precision guidance and precision strike, crop disease diagnosis, natural disaster prediction, identification and authentication, and so on. According to the source type of low-resolution image, the super-resolution analysis of digital image can be divided into two categories: one is super-resolution algorithm for multiple images (or sequence image), the other is super-resolution algorithm for single image. In this paper, the super-resolution algorithm of single image is studied. Based on the basic concept of super-resolution of digital image, the image needed to be interpolated with low resolution is divided into smooth part and non-smooth part. The super-resolution algorithm of single digital image is studied for the two parts. In the first part, for the smooth continuous function model, the spatial interpolation methods are discussed, including nearest neighbor interpolation, bilinear interpolation, bicubic spline and two-dimensional direct interpolation. The experimental results show that the two-dimensional direct interpolation algorithm is superior to the previous four interpolation algorithms, and the optimal super-resolution results can be obtained under the same complexity, while the nearest neighbor method has the worst performance in all cases. The bicubic interpolation method and the bicubic spline rule are superior to the nearest neighbor method and the bilinear interpolation method. The second part of the interpolation algorithm, using the complex contour wave transform. Up to now, there are two main types of complex contour wave transform, that is, constant redundancy complex contour wave transform and variable redundancy complex contour wave transform. The constant redundancy complex contour wave transform is a double tree complex contour wave transform, which can be considered as the real and imaginary parts of the basic contour wave transformation, while the complex contour wave transformation with variable redundancy is based on the mapping framework. Different redundancy can be selected according to the situation. Because of its high flexibility and recombination, the latter is chosen as a transformation tool in this paper. In this paper, the realization method of the complex contour wave transform is expounded, the basic properties of the transformation are discussed, especially the invariant level of the transformation is discussed, on the basis of the transformation, A single image interpolation algorithm with good performance is designed. The algorithm makes full use of the shift invariant property of complex contour wave transform. The low resolution needs to interpolate the non-smooth part of the image. In the complex contour wave domain, the transform domain is interpolated, then the transform domain is inversely transformed. The proposed algorithm is fused with the two-dimensional direct interpolation algorithm and the experimental results are satisfactory. The experimental results show that the proposed algorithm can better realize the super-resolution reconstruction of a single image, and has good performance in both objective parameters (such as peak signal-to-noise ratio) and subjective effect.
【学位授予单位】：信阳师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前3条
1 罗敏;李辉;;低照度条件下超分辨率人脸图像采集系统设计与实现[J];计算机测量与控制;2016年11期
2 陈新武;刘真;马文娟;薛静;范春凤;;三特征Contourlet纹理图像检索算法[J];信阳师范学院学报(自然科学版);2015年03期
3 董江,叶梧,冯穗力;内插法在提高电视图像清晰度中的应用[J];电视技术;2004年02期
，本文编号：2229378
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2229378.html
