鬼成像像质增强方法研究及超分辨实现
发布时间：2019-01-23 21:15
【摘要】：近年来,鬼成像这种新型的计算成像技术引起了越来越广泛的关注,与传统光学成像相比具有某些独特的成像优势。但是其成像分辨率一直是制约该技术发展的一个主要问题。为了促进鬼成像能够早日从实验阶段走向实用化,本文针对如何改善鬼成像的成像质量以及实现超分辨率成像做出研究。本文的工作主要是从改善鬼成像质量的算法出发,分析鬼成像计算模型的特点,寻求提高鬼成像质量的有效途径。主要内容如下:探究压缩感知五种经典重构算法,验证其可以在少量测量条件下恢复图像的可实行性。构建基于压缩感知的计算鬼成像模型,进行仿真实验。分别选用离散余弦变换矩阵和高斯随机矩阵作为稀疏基底和观测矩阵。合理选择压缩采样比和光场自由传播距离,并用傅里叶变换对光场传播建立数学模型。从而建立完整的基于压缩感知的计算式鬼成像系统。针对该系统鬼成像重构图像中存在的噪声问题,提出了BM3D算法与鬼成像相结合的计算方法。BM3D算法通过对图像自身相似块组合滤波,可有效去除原始算法带来的某些误差。设置相同实验参数进行对比实验,结果表明该方法有效提高了重构图像的峰值信噪比和视觉效果。最后分析了BM3D算法在应用到鬼成像问题中的适用范围。基于多幅鬼成像图像的超分辨重构。对同一物体从6个不同角度分别进行采集。通过SIFT算法对6幅图像配准,从图像中提取出对应同一场景的信息并匹配到同一幅图像中,然后通过自适应插值的方法进行超分辨重构。结果表明该方法得到的图像大部分区域有良好的视觉改善效果,原本不够清晰的信息得以分辨。但是鬼成像自身恢复图像时模糊严重的像素无法在关键点索引时被提取,一定程度上影响了超分辨重构的结果。最后分析了两种改善成像质量方法各自的优缺点,提出了改进意见,对鬼成像质量提高的研究与实际应用发展做出展望。
[Abstract]:In recent years, ghost imaging, a new computational imaging technology, has attracted more and more attention. Compared with traditional optical imaging, it has some unique imaging advantages. However, the imaging resolution has been a major problem restricting the development of the technology. In order to promote the application of ghost imaging from experimental stage to practical stage, this paper studies how to improve the imaging quality of ghost imaging and how to realize super-resolution imaging. The main work of this paper is to find an effective way to improve the quality of ghost imaging by analyzing the characteristics of the calculation model of ghost imaging from the point of view of the algorithm to improve the quality of ghost imaging. The main contents are as follows: five classical reconstruction algorithms of compression perception are explored to verify the practicability of image restoration under a few measurement conditions. A computational ghost imaging model based on compression perception is constructed and simulated. Discrete cosine transform matrix and Gao Si random matrix are used as sparse base and observation matrix respectively. The compression sampling ratio and the free propagation distance of light field are reasonably selected, and the mathematical model of light field propagation is established by Fourier transform. Thus, a complete computational ghost imaging system based on compression perception is established. In order to solve the noise problem in the reconstructed image of ghost imaging system, a method of combining BM3D algorithm with ghost image is proposed. By filtering the image with its own similar blocks, BM3D algorithm can effectively remove some errors caused by the original algorithm. The experimental results show that the proposed method can effectively improve the PSNR and visual effect of reconstructed images. Finally, the application range of BM3D algorithm in ghost imaging problem is analyzed. Superresolution reconstruction based on multiple ghost images. The same object was collected from six different angles. Six images are registered by SIFT algorithm. The information corresponding to the same scene is extracted from the image and matched to the same image. Then the super-resolution reconstruction is carried out by adaptive interpolation. The results show that most of the images obtained by this method have a good visual improvement effect, and the original information is not clear enough to distinguish. However, the pixels with serious blur can not be extracted when the key points are indexed, which affects the result of super-resolution reconstruction to some extent. Finally, the advantages and disadvantages of the two methods to improve the imaging quality are analyzed, and the improvement suggestions are put forward, and the prospects for the research and practical application of improving the quality of ghost imaging are made.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;O431.2
本文编号：2414198
[Abstract]:In recent years, ghost imaging, a new computational imaging technology, has attracted more and more attention. Compared with traditional optical imaging, it has some unique imaging advantages. However, the imaging resolution has been a major problem restricting the development of the technology. In order to promote the application of ghost imaging from experimental stage to practical stage, this paper studies how to improve the imaging quality of ghost imaging and how to realize super-resolution imaging. The main work of this paper is to find an effective way to improve the quality of ghost imaging by analyzing the characteristics of the calculation model of ghost imaging from the point of view of the algorithm to improve the quality of ghost imaging. The main contents are as follows: five classical reconstruction algorithms of compression perception are explored to verify the practicability of image restoration under a few measurement conditions. A computational ghost imaging model based on compression perception is constructed and simulated. Discrete cosine transform matrix and Gao Si random matrix are used as sparse base and observation matrix respectively. The compression sampling ratio and the free propagation distance of light field are reasonably selected, and the mathematical model of light field propagation is established by Fourier transform. Thus, a complete computational ghost imaging system based on compression perception is established. In order to solve the noise problem in the reconstructed image of ghost imaging system, a method of combining BM3D algorithm with ghost image is proposed. By filtering the image with its own similar blocks, BM3D algorithm can effectively remove some errors caused by the original algorithm. The experimental results show that the proposed method can effectively improve the PSNR and visual effect of reconstructed images. Finally, the application range of BM3D algorithm in ghost imaging problem is analyzed. Superresolution reconstruction based on multiple ghost images. The same object was collected from six different angles. Six images are registered by SIFT algorithm. The information corresponding to the same scene is extracted from the image and matched to the same image. Then the super-resolution reconstruction is carried out by adaptive interpolation. The results show that most of the images obtained by this method have a good visual improvement effect, and the original information is not clear enough to distinguish. However, the pixels with serious blur can not be extracted when the key points are indexed, which affects the result of super-resolution reconstruction to some extent. Finally, the advantages and disadvantages of the two methods to improve the imaging quality are analyzed, and the improvement suggestions are put forward, and the prospects for the research and practical application of improving the quality of ghost imaging are made.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;O431.2
【参考文献】
相关期刊论文 前2条
1 赵生妹;庄鹏;;Correspondence normalized ghost imaging on compressive sensing[J];Chinese Physics B;2014年05期
2 白旭;李永强;赵生妹;;基于压缩感知的差分关联成像方案研究[J];物理学报;2013年04期
相关博士学位论文 前2条
1 李虎;关联光学成像新机制与新方法研究[D];上海交通大学;2014年
2 丁南南;基于特征点的图像配准技术研究[D];中国科学院研究生院（长春光学精密机械与物理研究所）;2012年
相关硕士学位论文 前2条
1 李超;基于小波变换与BM3D的多图去噪方法研究[D];华北电力大学;2015年
2 华翌辰;基于压缩感知的鬼成像研究[D];南京理工大学;2014年
，本文编号：2414198
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2414198.html
