基于非局部相似模型的图像恢复算法研究
发布时间：2019-01-22 18:27
【摘要】：图像恢复旨在尽可能的对原始图像进行高保真度的重建,如何提高图像的恢复性能,一直是图像处理领域的研究热点。图像恢复与图像采集、存储和传输过程息息相关,有效的图像信息获取框架对之后的图像恢复过程起着重要作用。压缩感知,作为一种新兴的信号采集理论,给图像处理领域带来了革命性突破。该理论能够以远低于香农-奈奎斯特采样定理要求的频率对稀疏或可压缩信号进行同步采样和压缩,并利用少量的随机测量值对信号进行重建。自提出以来,压缩感知理论受到了图像处理领域学者们的广泛关注。图像恢复,作为压缩感知理论的核心问题之一,一直是该领域的研究热点。目前大部分的压缩感知图像重建算法都是利用图像信号在某个特征空间下的稀疏性构建目标优化函数,但没有充分考虑图像信号的其他先验信息,影响了算法的重建性能和算法的适应性。对于图像信号,除了在特定特征空间下的稀疏性以外,还具备很多其他属性,如图像的局部特性和结构化属性等。如何有效利用图像的这些属性,进一步提高图像恢复性能,是本文的研究重点。本文基于非局部相似模型,以压缩感知中的图像恢复算法为对象,进行了研究。考虑图像的非局部自相似性,提出一种基于图像相似块低秩的压缩感知图像重建算法,将图像恢复问题转化为聚合的相似块矩阵秩最小问题。算法以最小压缩感知重建误差为约束构建优化模型,并采用加权核范数最小化算法求解低秩优化问题,很好地挖掘了图像自身的信息和结构化稀疏特征,保护了图像的结构和纹理细节。多个测试图像,不同采样率下的实验证明了算法的有效性,特别是在低采样率下对于纹理较为丰富的图像,提出的算法图像重建质量较明显的优于最新的同类算法。进一步,考虑在传统的基于非局部相似模型的图像恢复算法中,采用简单的矩形形状完成图像样本块的提取和相似块匹配,破坏了图像的结构信息,特别是图像边缘处的结构特征,提出一种基于非局部相似的形状自适应压缩感知图像恢复算法。算法采用超像素算法进行样本块提取,并采用与样本块相同的形状进行相似块匹配。由于有效利用了给定图像的结构信息,所提取到的样本块对图像的边界依附性更强。同时,由于块内像素冗余度更高,得到的相似块低秩矩阵的秩更小,对压缩感知图像的重建更有益。
[Abstract]:The purpose of image restoration is to reconstruct the original image with high fidelity as much as possible. How to improve the performance of image restoration has always been a hot topic in the field of image processing. Image recovery is closely related to the process of image acquisition, storage and transmission. Effective image information acquisition framework plays an important role in the process of image recovery. Compression sensing, as a new theory of signal acquisition, has brought a revolutionary breakthrough to the field of image processing. The theory can synchronously sample and compress sparse or compressible signals at frequencies far lower than those required by Shannon-Nyquist sampling theorem and reconstruct the signals with a small number of random measurements. Since it was put forward, the theory of compressed perception has been paid more and more attention by many researchers in the field of image processing. Image restoration, as one of the core issues in the theory of compression perception, has been a hot topic in this field. At present, most of the compressed perceptual image reconstruction algorithms use the sparsity of the image signal in a feature space to construct the objective optimization function, but the other prior information of the image signal is not fully considered. The reconstruction performance and adaptability of the algorithm are affected. For image signals, there are many other attributes besides sparsity in specific feature spaces, such as the local and structured properties of images. How to effectively utilize these attributes of images and further improve the performance of image recovery is the focus of this paper. Based on the nonlocal similarity model, the image restoration algorithm in compressed perception is studied in this paper. Considering the nonlocal self-similarity of images, a compressed perceptual image reconstruction algorithm based on the low rank of image similarity blocks is proposed, which transforms the image restoration problem into the rank minimization problem of aggregated similarity block matrix. The algorithm uses minimum compression perceptual reconstruction error as the constraint to construct the optimization model, and uses weighted kernel norm minimization algorithm to solve the low rank optimization problem. Protects the structure and texture details of the image. Experiments on multiple test images at different sampling rates demonstrate the effectiveness of the proposed algorithm, especially for images with rich texture at low sampling rate, and the proposed algorithm is superior to the latest similar algorithms. Furthermore, in the traditional image restoration algorithm based on the non-local similarity model, the simple rectangular shape is used to complete the image sample block extraction and similar block matching, which destroys the structure information of the image. Especially for the structural features of image edges, a shape adaptive compression perceptual image restoration algorithm based on non-local similarity is proposed. The superpixel algorithm is used to extract the sample block and the shape of the sample block is used to match the sample block. Because the structure information of a given image is used effectively, the extracted sample blocks are more dependent on the image boundaries. At the same time, because of the higher pixel redundancy in the block, the rank of the low rank matrix of the similar block is smaller, which is more beneficial to the reconstruction of compressed perceptual image.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
本文编号：2413452
[Abstract]:The purpose of image restoration is to reconstruct the original image with high fidelity as much as possible. How to improve the performance of image restoration has always been a hot topic in the field of image processing. Image recovery is closely related to the process of image acquisition, storage and transmission. Effective image information acquisition framework plays an important role in the process of image recovery. Compression sensing, as a new theory of signal acquisition, has brought a revolutionary breakthrough to the field of image processing. The theory can synchronously sample and compress sparse or compressible signals at frequencies far lower than those required by Shannon-Nyquist sampling theorem and reconstruct the signals with a small number of random measurements. Since it was put forward, the theory of compressed perception has been paid more and more attention by many researchers in the field of image processing. Image restoration, as one of the core issues in the theory of compression perception, has been a hot topic in this field. At present, most of the compressed perceptual image reconstruction algorithms use the sparsity of the image signal in a feature space to construct the objective optimization function, but the other prior information of the image signal is not fully considered. The reconstruction performance and adaptability of the algorithm are affected. For image signals, there are many other attributes besides sparsity in specific feature spaces, such as the local and structured properties of images. How to effectively utilize these attributes of images and further improve the performance of image recovery is the focus of this paper. Based on the nonlocal similarity model, the image restoration algorithm in compressed perception is studied in this paper. Considering the nonlocal self-similarity of images, a compressed perceptual image reconstruction algorithm based on the low rank of image similarity blocks is proposed, which transforms the image restoration problem into the rank minimization problem of aggregated similarity block matrix. The algorithm uses minimum compression perceptual reconstruction error as the constraint to construct the optimization model, and uses weighted kernel norm minimization algorithm to solve the low rank optimization problem. Protects the structure and texture details of the image. Experiments on multiple test images at different sampling rates demonstrate the effectiveness of the proposed algorithm, especially for images with rich texture at low sampling rate, and the proposed algorithm is superior to the latest similar algorithms. Furthermore, in the traditional image restoration algorithm based on the non-local similarity model, the simple rectangular shape is used to complete the image sample block extraction and similar block matching, which destroys the structure information of the image. Especially for the structural features of image edges, a shape adaptive compression perceptual image restoration algorithm based on non-local similarity is proposed. The superpixel algorithm is used to extract the sample block and the shape of the sample block is used to match the sample block. Because the structure information of a given image is used effectively, the extracted sample blocks are more dependent on the image boundaries. At the same time, because of the higher pixel redundancy in the block, the rank of the low rank matrix of the similar block is smaller, which is more beneficial to the reconstruction of compressed perceptual image.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【参考文献】
相关期刊论文 前6条
1 王良君;石光明;李甫;谢雪梅;林耀海;;多稀疏空间下的压缩感知图像重构[J];西安电子科技大学学报;2013年03期
2 潘宗序;黄慧娟;禹晶;胡少兴;张爱武;马洪兵;孙卫东;;基于压缩感知与结构自相似性的遥感图像超分辨率方法[J];信号处理;2012年06期
3 刘记红;黎湘;徐少坤;庄钊文;;基于改进正交匹配追踪算法的压缩感知雷达成像方法[J];电子与信息学报;2012年06期
4 李志林;陈后金;李居朋;姚畅;杨娜;;一种有效的压缩感知图像重建算法[J];电子学报;2011年12期
5 焦李成;杨淑媛;刘芳;侯彪;;压缩感知回顾与展望[J];电子学报;2011年07期
6 练秋生;肖莹;;基于小波树结构和迭代收缩的图像压缩感知算法研究[J];电子与信息学报;2011年04期
，本文编号：2413452
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2413452.html
