WO2023109008A1 - 气动光学效应空变模糊图像复原方法及系统 - Google Patents

气动光学效应空变模糊图像复原方法及系统 Download PDF

Info

Publication number
WO2023109008A1
WO2023109008A1 PCT/CN2022/093806 CN2022093806W WO2023109008A1 WO 2023109008 A1 WO2023109008 A1 WO 2023109008A1 CN 2022093806 W CN2022093806 W CN 2022093806W WO 2023109008 A1 WO2023109008 A1 WO 2023109008A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
kernel
blur
space
aero
Prior art date
Application number
PCT/CN2022/093806
Other languages
English (en)
French (fr)
Inventor
洪汉玉
王博
张耀宗
张天序
李琼
Original Assignee
武汉工程大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 武汉工程大学 filed Critical 武汉工程大学
Publication of WO2023109008A1 publication Critical patent/WO2023109008A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/73—Deblurring; Sharpening
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows

Definitions

  • the invention relates to the field of aerospace image processing, in particular to a method and system for restoring aero-optical effect space-varying blurred images.
  • Image blurring caused by aero-optical effects is complex, especially for large-scale remote sensing images, where each point of the entire image has different degrees of blurring and different types of blurring. It is very difficult to establish a restoration model for such images. In the process of image restoration, the accurate solution of the blur kernel is very important.
  • the existing methods for solving the blur kernel are various, which can be roughly divided into two categories: one is that the blur caused by the image is empty and invariant, and the overall information of the image is used to estimate A single blur kernel, and then deconvolve the image and the blur kernel to obtain the restored image; the second is that the blur caused by the image is empty and variable, so the image is divided into blocks or part of the image information is used to solve multiple different sub-region points Diffusion function, and then deconvolute the two to obtain the restored image of each sub-region, and finally merge to obtain a complete and clear image.
  • the main purpose of the present invention is to provide a method and system capable of effectively restoring aero-optical effect space-variant blurred images.
  • an aero-optical effect space-varying blurred image restoration method comprising the following steps:
  • the connectivity coefficient term and the L2 regularization term are added to the degradation model to make each point of the blur kernel continuous and smooth.
  • the threshold value is set to 0.6 in step S4.
  • step S5 specifically store all blur kernels vertically in the matrix in the order in which they are obtained, then traverse the pixels within the entire image, and compare each pixel with the center points of all image blocks Carry out Manhattan distance comparison, take the fuzzy kernel matrix corresponding to the two nearest neighbors and the second nearest neighbor as the reference matrix, and finally perform inverse distance linear interpolation on these two reference matrices to obtain the fuzzy kernel matrix of the corresponding pixel.
  • the present invention also provides an aero-optical effect space-varying blurred image recovery system, including:
  • An image acquisition module configured to acquire a space-varying blurred image of the aero-optical effect
  • the partition module is used to partition the space-variable fuzzy image for the first time by adopting a binary tree structure, and divide the entire image vertically and then horizontally into four regions with equal areas and shapes;
  • the regression model establishment module is used to establish a degradation model, and a smoothing factor constraint item is added to the degradation model;
  • the fuzzy kernel calculation module is used to solve the fuzzy kernel for each area by using the alternate minimization iterative method, compare the similarity of the fuzzy kernels of adjacent image blocks, and if it is less than the set threshold, repass the corresponding image block through the partition module according to the binary tree The structure continues to partition until the similarity of the blur kernels of adjacent image blocks meets the set threshold;
  • the convolution kernel matrix construction module is used to linearly interpolate all the solved fuzzy kernels point by point to obtain the fuzzy kernel of each pixel, and construct the convolution kernel matrix so that each row of the convolution kernel matrix corresponds to the fuzzy kernel of each pixel;
  • the image restoration module is used to construct a space-varying convolution model according to the convolution kernel matrix, and obtain a completely deconvoluted restored image.
  • the regression model building module is also used to add a connectivity coefficient item and an L2 regularization item in the degradation model, so that each point of the fuzzy kernel is continuous and smooth.
  • the threshold value is set to 0.6 in the fuzzy kernel calculation module.
  • the convolution kernel matrix construction module is specifically used to store all blur kernels in the matrix vertically in the order of obtaining, and then traverse the pixels in the entire image, and combine each pixel with all image blocks
  • the Manhattan distance comparison is performed on the center point of the region, and the blur kernel matrix corresponding to the two regions of the nearest neighbor and the second nearest neighbor is taken as the reference matrix.
  • the inverse distance linear interpolation is performed on the two reference matrices to obtain the blur kernel of the corresponding pixel. matrix.
  • the present invention also provides a computer storage medium, which can be executed by a processor, and has a computer program stored therein, and the computer program executes the method for restoring a space-variant blurred image with aero-optical effect described in the above technical solution.
  • the beneficial effects produced by the present invention are: the present invention divides the aero-optical effect space-varying blur kernel image multiple times through the binary tree structure until the similarity of the blur kernels of adjacent image blocks meets certain requirements.
  • this fuzzy kernel similarity adaptive subregion method a space-varying image is segmented into multiple approximately space-invariant images to a certain extent, which improves the accuracy of the blur kernel estimation of the space-varying image on the one hand, and on the other hand
  • the similarity between the blur kernels in each area of the image is made, thereby effectively improving the quality of the subsequent point-by-point interpolation blur kernels, so that the rapid and effective restoration of the real spatially variable blurred image of the wide field of view can be realized.
  • Fig. 1 is a flowchart of a method for restoring a blurred image caused by aero-optical effect and space variation according to an embodiment of the present invention
  • Fig. 2 is the flow chart of the self-adaptive binary tree partition of the embodiment of the present invention.
  • Fig. 3 is a flow chart of fuzzy kernel estimation according to an embodiment of the present invention.
  • Fig. 4 is the space-varying blurred image of the embodiment of the present invention.
  • FIG. 5 is a schematic diagram of image partitioning according to an embodiment of the present invention.
  • Fig. 6 is the initial fuzzy kernel solved by the embodiment of the present invention.
  • Fig. 7 is the fuzzy kernel solved point by point in the embodiment of the present invention (the upper left corner of the image is the fuzzy kernel of each pixel in 5 rows and 5 columns);
  • Fig. 8 is the final point-by-point deconvolution result of the embodiment of the present invention.
  • the method for restoring a blurred image caused by aero-optical effect space variation in an embodiment of the present invention includes the following steps:
  • a point-by-point deconvolution algorithm for an aero-optical effect space-varying blurred image in the embodiment of the present invention is implemented by using the 2018b version of the MATLAB program.
  • the algorithm flow is shown in Figure 1, including the following steps:
  • the degraded image obtained in step S1 is denoted as f (as shown in FIG. 4 ), and the size of the image is 436*452.
  • f the degraded image obtained in step S1
  • the size of the image is 436*452.
  • the first item is the smooth factor constraint item, to ensure that the difference between the adjacent points of the latent image obtained each time is extremely small, so that it has the maximum spatial correlation;
  • the second item is the connectivity coefficient item, which ensures that the points of the fuzzy kernel k are continuous;
  • the third item is the L2 regularization item, which makes the obtained blur kernel smoother. The process is shown in Figure 3.
  • the alternate minimization iterative method can be used to obtain the blur kernel k of the corresponding block and the clear latent image f m .
  • the size of the blur kernel is set to 21 ⁇ 21.
  • the calculation formula of the blur kernel k is as follows:
  • f m represents the clear latent image obtained in the iterative process
  • k is the blur kernel
  • 2 is the two-norm, guarantee The difference between the adjacent points of the clear latent image obtained each time is extremely small, so that it has the largest spatial correlation.
  • the latent image f m is calculated as follows:
  • ⁇ (A) is a connectivity coefficient, which ensures that each point in the fuzzy kernel k is continuous, and does not affect other fuzzy kernels in step S4, and is defined as follows:
  • P r (i, j) is the area whose radius is r from the center of (i, j), and r is set to 1;
  • (i′, j′) is the index of each point of the fuzzy kernel;
  • step S4 specifically compare and calculate the similarity of blur kernels between two adjacent image blocks:
  • h 1 and h 2 are the vector forms of the estimated blur kernels respectively, and the value of ⁇ (h 1 , h 2 ) is within [0,1], the closer to 1, the higher the similarity between the two blur kernels .
  • Set a correlation threshold T h which is taken as 0.6.
  • a space-varying image is divided into multiple approximately space-invariant images to a certain extent, which improves the accuracy of the estimation of the blur kernel of the space-varying image on the one hand, and makes the There is similarity between the blur kernels in each region of the image, thereby effectively improving the quality of the point-by-point interpolation blur kernels in step S4.
  • step S5 all fuzzy kernels k are longitudinally stored in the matrix Matrix_sum according to the obtained order, and then f(i, j) is traversed in the whole image, and f(i, j) is combined with all
  • the center point of the area of the image block is compared with the Manhattan distance, and the blur kernel matrix corresponding to the two areas of the nearest neighbor and the second nearest neighbor is taken as the reference matrix, and then the two reference matrices are used for inverse distance linear interpolation to obtain the blur kernel of the corresponding pixel.
  • Matrix k i, j as shown in Figure 7, the algorithm is as follows:
  • Matrix_sum is all the stored fuzzy kernel matrices
  • d mc , d nc are the center points of any two block areas
  • d m , d n are each pixel point and any two
  • the minimum distance after the Manhattan distance comparison of the block area k m and k n are the blur kernels corresponding to the two areas
  • k i, j are the blur kernel matrix corresponding to each pixel
  • sum is the summation operation.
  • step S6 a space-variation restoration model is established, and the formula is as follows:
  • K is the convolution kernel matrix, each row of the matrix corresponds to the point spread function of each f(i, j), ⁇ x and ⁇ y are derivative filters, which are respectively taken as [1,-1], [1,-1] T , the value of ⁇ is 2000, and the value of ⁇ is 0.8, and finally the complete deconvolution result is obtained As shown in Figure 8.
  • the present invention also provides an aero-optical effect space-variant blurred image recovery system, which is mainly used to realize the above-mentioned method embodiment, and the system includes:
  • An image acquisition module configured to acquire a space-varying blurred image of the aero-optical effect
  • the partition module is used to partition the space-variable fuzzy image for the first time by adopting a binary tree structure, and divide the entire image vertically and then horizontally into four regions with equal areas and shapes;
  • the regression model establishment module is used to establish a degradation model, and a smoothing factor constraint item is added to the degradation model;
  • the fuzzy kernel calculation module is used to solve the fuzzy kernel for each area by using the alternate minimization iterative method, compare the similarity of the fuzzy kernels of adjacent image blocks, and if it is less than the set threshold, repass the corresponding image block through the partition module according to the binary tree The structure continues to partition until the similarity of the blur kernels of adjacent image blocks meets the set threshold;
  • the convolution kernel matrix construction module is used to linearly interpolate all the solved fuzzy kernels point by point to obtain the fuzzy kernel of each pixel, and construct the convolution kernel matrix so that each row of the convolution kernel matrix corresponds to the fuzzy kernel of each pixel;
  • the image restoration module is used to construct a space-varying convolution model according to the convolution kernel matrix, and obtain a completely deconvoluted restored image.
  • the regression model building module is also used to add a connectivity coefficient item and an L2 regularization item in the degradation model, so that each point of the fuzzy kernel is continuous and smooth.
  • the threshold value is set to 0.6 in the fuzzy kernel calculation module.
  • the convolution kernel matrix construction module is specifically used to store all blur kernels in the matrix vertically in the order of obtaining, and then traverse the pixels in the entire image, and combine each pixel with all image blocks
  • the Manhattan distance comparison is performed on the center point of the region, and the blur kernel matrix corresponding to the two regions of the nearest neighbor and the second nearest neighbor is taken as the reference matrix.
  • the inverse distance linear interpolation is performed on the two reference matrices to obtain the blur kernel of the corresponding pixel. matrix.
  • the present application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), Magnetic Storage, Magnetic Disk, Optical Disk, Server, App Store, etc., on which computer programs, program When executed by the processor, corresponding functions are realized.
  • the computer-readable storage medium in this embodiment is used to implement the method for restoring a blurred image caused by aero-optical effect and space variation in the method embodiment when executed by a processor.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

本发明公开了一种气动光学效应空变模糊图像复原方法，包括以下步骤：采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；建立退化模型，且在退化模型中添加平滑因子约束项；使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。

Description

气动光学效应空变模糊图像复原方法及系统 【技术领域】
本发明涉及航天图像处理领域，尤其涉及一种气动光学效应空变模糊图像复原方法及系统。
【背景技术】
近年来，我国航天技术发展迅速，但遥感应用的发展相对滞后，在大气湍流的影响下，配备在航天飞行器上的光学成像系统往往难以获取到清晰、可利用性高的地面图像，复杂目标的识别难度加大，有效、快速地对气动光学效应空变模糊图像进行复原显得至为关键。
由气动光学效应引起的图像模糊是复杂的，尤其针对幅面大的遥感图像，整幅图像各点模糊程度不同，且模糊类型也不一样，对于此类图像建立复原模型存在很大的难度。在复原图像过程中，模糊核的准确求解十分重要，现有的求解模糊核方法多种多样，大致可以分为两类：一是认为造成图像的模糊是空不变的，利用图像整体信息估计单一模糊核，然后将图像与模糊核作去卷积运算得到复原图像；二是认为造成图像的模糊是空可变的，因此将图像进行分块或者使用部分图像信息求解多个不同子区域点扩散函数，再分别将二者作去卷积运算得到各子区域的复原图像，最后合并得到完整清晰图像。
但是气动光学效应遥感退化图像具有幅面大、退化因素多等特点，不能认为整幅图像模糊是单一的，这些方法往往需要进行非常复杂的计算，得到的模糊核也不精确，复原结果不理想(存在振铃效应)。因此，有必要针对上述问题设计专用的空变复原算法。
【发明内容】
本发明主要目的在于提供一种能够对气动光学效应空变模糊图像进行有效复原的方法及系统。
本发明所采用的技术方案是：
提供一种气动光学效应空变模糊图像复原方法，包括以下步骤：
S1、获取气动光学效应的空变模糊图像；
S2、采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
S3、建立退化模型，且在退化模型中添加平滑因子约束项；
S4、使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；
S5、对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
S6、根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
接上述技术方案，在退化模型中再增加连通系数项和L2正则化项，使得模糊核各点连续且平滑。
接上述技术方案，步骤S4中设定阈值为0.6。
接上述技术方案，步骤S5中，具体将所有模糊核按求得顺序纵向保存到矩阵中，然后在全图范围内对像素点进行遍历，并将每个像素点与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块的区域对应的模糊核矩阵作为基准矩阵，最后对这两个基准矩阵作反距离线性插值，得到对应像素点的模糊核矩阵。
本发明还提供一种气动光学效应空变模糊图像复原系统，包括：
图像获取模块，用于获取气动光学效应的空变模糊图像；
分区模块，用于采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
退换模型建立模块，用于建立退化模型，且在退化模型中添加平滑因子约束项；
模糊核计算模块，用于使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新通过分区模块按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；
卷积核矩阵构造模块，用于对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
图像复原模块，用于根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
接上述技术方案，退换模型建立模块还用于在退化模型中增加连通系数项和L2正则化项，使得模糊核各点连续且平滑。
接上述技术方案，模糊核计算模块中设定阈值为0.6。
接上述技术方案，卷积核矩阵构造模块具体用于将所有模糊核按求得顺序纵向保存到矩阵中，然后在全图范围内对像素点进行遍历，并将每个像素点与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块的区域对应的模糊核矩阵作为基准矩阵，最后对这两个基准矩阵作反距离线性插值，得到对应像素点的模糊核矩阵。
本发明还提供一种计算机存储介质，其可被处理器执行，且其内存储有计算机程序，该计算机程序执行上述技术方案所述的气动光学效应空变模糊图像复原方法。
本发明产生的有益效果是：本发明通过二叉树结构对气动光学效应空变模糊核图像进行多次分区，直到相邻图像块的模糊核的相似性满足一定的要求。通过这种模糊核相似性自适应分区域方法在一定程度上将一幅空变图像分割成了多幅近似空不变图像，一方面提升了对于空变图像模糊核估计的精确性，另一方面使得图像各区域模糊核之间具有相似性，从而有效提升后续逐点插值模糊核的质量，从而可以实现对宽视场真实空变模糊图像进行快速有效复原。
【附图说明】
下面将结合附图及实施例对本发明作进一步说明，附图中：
图1是本发明实施例气动光学效应空变模糊图像复原方法的流程图；
图2是本发明实施例自适应二叉树分区的流程图；
图3是本发明实施例模糊核估计的流程图；
图4是本发明实施例空变模糊图像；
图5是本发明实施例图像分区示意图；
图6是本发明实施例求解的初始模糊核；
图7是本发明实施例逐点求解的模糊核(图像左上角为5行5列每个像素点的模糊核)；
图8是本发明实施例最终逐点去卷积结果。
【具体实施方式】
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。
如图1所示，本发明实施例气动光学效应空变模糊图像复原方法，包括以下步骤：
S1、获取气动光学效应的空变模糊图像；
S2、采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
S3、建立退化模型，且在退化模型中添加平滑因子约束项；
S4、使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新按照二叉树结构继续分区，如图5所示，直到相邻图像块模糊核的相似性满足设定阈值；
S5、对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
S6、根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
本发明实施例的一种气动光学效应空变模糊图像逐点去卷积算法，采用2018b版本的MATLAB程序实现，算法流程如图1所示，包括以下步骤：
如图2所示，本发明的实施例中，步骤S1中获取的退化图像记为f(如图4)，图像大小为436*452。采用二叉树结构对图像进行第一次横纵方向的分区，得到面积形状相等的四块区域f 1，f 2，f 2，f 4，如图5所示。
f→(f 1，f 2，f 3，f 4)    (1)
步骤S3中建立退化模型时，添加3个约束项，第一项是平滑因子约束项，保证每一次求得的潜像相邻点之间的差异为极小，使其具有最大空间相关性；第二项是连通系数项，保证模糊核k各点连续；第三项是L2正则化项，使得到的模糊核较平滑，流程如图3所示。
具体可使用交替极小化迭代方法求解得到相应块区的模糊核k以及清晰潜像f m，模糊核尺寸设定为21×21，模糊核k的计算公式如下：
Figure PCTCN2022093806-appb-000001
f m表示迭代过程中求得的清晰潜像，k为模糊核，||·|| 2为二范数，保证
Figure PCTCN2022093806-appb-000002
每一次求得的清晰潜像相邻点之间的差异为极小，使其具有最大的空间相关性。
潜像f m通过如下计算得到：
Figure PCTCN2022093806-appb-000003
λ(A)为连通系数，保证模糊核k内各点连续，在步骤S4中不对其他模糊核造成影响，定义如下：
Figure PCTCN2022093806-appb-000004
P r(i，j)为以(i，j)中心的半径为r的领域，r取1；(i′，j′)为模糊核各点索引；
Figure PCTCN2022093806-appb-000005
为模糊核的L2正则化项，使得到的模糊核较平滑；
步骤S4中，具体比较计算两两相邻图像块的模糊核相似性：
Figure PCTCN2022093806-appb-000006
其中，h 1和h 2分别为估计出的模糊核的向量形式，ε(h 1，h 2)的取值在[0,1]内，越接近1表示两个模糊核的相似度越高。设定一个相关性阈值T h，取为0.6，当相邻区域的模糊核相似度存在ε(h 1，h 2)≥T h时，视为满足条件不再进行分区，当相邻区域的两个模糊核相似度存在ε(h 1，h 2)＜T h时，视为不满足条件，再次对相应区域块进行二叉树结构分区，然后计算这部分区域的两两模糊核相似度，直至满足设定条件，分区结束，如图6所示。依据模糊核相似性自适应分区域方法在一定程度上将一幅空变图像分割成了多幅近似空不变图像，一方面提升了对于空变图像模糊核估计的精确性，另一方面使得图像各区域模糊核之间具有相似性，从而有效提升步骤S4中逐点插值模糊核的质量。
步骤S5中，将所有模糊核k按求得顺序纵向保存到矩阵Matrix_sum当中，然后在全图范围内对f(i，j)进行遍历，并在这个过程中将f(i，j)与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块区域对应的模糊核矩阵作为基准矩阵，再将这两个基准矩阵作反距离线性插值得到对应像素点的模糊核矩阵k i，j，如图7所示，算法如下：
Figure PCTCN2022093806-appb-000007
Figure PCTCN2022093806-appb-000008
其中k为所有求得块区域的模糊核，Matrix_sum为储存的所有模糊核矩阵，d mc、d nc为任意两个块区域的中心点，d m、d n为每个像素点与任意两个块区域进行曼哈顿距离比较后的最小距离，k m、k n为对应两块区域的模糊核，k i，j为每个像素点对应的模糊核矩阵，sum为求和操作。
步骤S6中建立空变复原模型，公式如下：
Figure PCTCN2022093806-appb-000009
K为卷积核矩阵，矩阵每一行对应每个f(i，j)的点扩散函数,Δ x和Δ y是导数滤波器，分别取为[1,-1]、[1,-1] T,λ取值为2000，α取值0.8,最终得到完整的去卷积结果
Figure PCTCN2022093806-appb-000010
如图8所示。
本发明还提供一种气动光学效应空变模糊图像复原系统，主要用于实现上述方法实施例，该系统包括：
图像获取模块，用于获取气动光学效应的空变模糊图像；
分区模块，用于采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
退换模型建立模块，用于建立退化模型，且在退化模型中添加平滑因子约束项；
模糊核计算模块，用于使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新通过分区模块按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；
卷积核矩阵构造模块，用于对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
图像复原模块，用于根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
接上述技术方案，退换模型建立模块还用于在退化模型中增加连通系数项和L2正则化项，使得模糊核各点连续且平滑。
接上述技术方案，模糊核计算模块中设定阈值为0.6。
接上述技术方案，卷积核矩阵构造模块具体用于将所有模糊核按求得顺序纵向保存到矩阵中，然后在全图范围内对像素点进行遍历，并将每个像素点与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块的区域对应的模糊核矩阵作为基准矩阵，最后对这两个基准矩阵作反距离线性插值，得到对应像素点的模糊核矩阵。
本申请还提供一种计算机可读存储介质，如闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘、服务器、App应用商城等等，其上存储有计算机程序，程序被处理器执行时实现相应功能。本实施例的计算机可读存储介质用于被处理器执行时实现方法实施例的气动光学效应空变模糊图像复原方法。
应当理解的是，对本领域普通技术人员来说，可以根据上述说明加以改进或变换，而所有这些改进和变换都应属于本发明所附权利要求的保护范围。

Claims (9)

  1. 一种气动光学效应空变模糊图像复原方法，其特征在于，包括以下步骤：
    S1、获取气动光学效应的空变模糊图像；
    S2、采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
    S3、建立退化模型，且在退化模型中添加平滑因子约束项；
    S4、使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；
    S5、对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
    S6、根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
  2. 根据权利要求1所述的气动光学效应空变模糊图像复原方法，其特征在于，在退化模型中再增加连通系数项和L2正则化项，使得模糊核各点连续且平滑。
  3. 根据权利要求1所述的气动光学效应空变模糊图像复原方法，其特征在于，步骤S4中设定阈值为0.6。
  4. 根据权利要求1所述的气动光学效应空变模糊图像复原方法，其特征在于，步骤S5中，具体将所有模糊核按求得顺序纵向保存到矩阵中，然后在全图范围内对像素点进行遍历，并将每个像素点与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块的区域对应的模糊核矩阵作为基准矩阵，最后对这两个基准矩阵作反距离线性插值，得到对应像素点的模糊核矩阵。
  5. 一种气动光学效应空变模糊图像复原系统，其特征在于，包括：
    图像获取模块，用于获取气动光学效应的空变模糊图像；
    分区模块，用于采用二叉树结构对空变模糊图像进行第一次分区，将整幅图像先纵向再横向划分为四块面积形状相等的区域；
    退换模型建立模块，用于建立退化模型，且在退化模型中添加平滑因子约束项；
    模糊核计算模块，用于使用交替极小化迭代方法对每一块区域求解模糊核，分别比较相邻图像块模糊核的相似性，若小于设定阈值则将相应图像块重新通过分区模块按照二叉树结构继续分区，直到相邻图像块模糊核的相似性满足设定阈值；
    卷积核矩阵构造模块，用于对所有求解的模糊核逐点线性插值得到每一个像素点的模糊核，构造卷积核矩阵，使得卷积核矩阵每一行对应每一个像素点的模糊核；
    图像复原模块，用于根据卷积核矩阵构建空变卷积模型，求解得到完整去卷积的复原图像。
  6. 根据权利要求5所述的气动光学效应空变模糊图像复原系统，其特征在于，退换模型建立模块还用于在退化模型中增加连通系数项和L2正则化项，使得模糊核各点连续且平滑。
  7. 根据权利要求5所述的气动光学效应空变模糊图像复原系统，其特征在于，模糊核计算模块中设定阈值为0.6。
  8. 根据权利要求5所述的气动光学效应空变模糊图像复原系统，其特征在于，卷积核矩阵构造模块具体用于将所有模糊核按求得顺序纵向保存到矩阵中，然后在全图范围内对像素点进行遍历，并将每个像素点与所有图像块的区域中心点进行曼哈顿距离比较，取当中最近邻和次近邻的两块的区域对应的模糊核矩阵作为基准矩阵，最后对这两个基准矩阵作反距离线性插值，得到对应像素点的模糊核矩阵。
  9. 一种计算机存储介质，其特征在于，其可被处理器执行，且其内存储有计算机程序，该计算机程序执行权利要求1-4中任一项所述的气动光学效应空变模糊图像复原方法。
PCT/CN2022/093806 2021-12-17 2022-05-19 气动光学效应空变模糊图像复原方法及系统 WO2023109008A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111548604.6A CN113935925B (zh) 2021-12-17 2021-12-17 气动光学效应空变模糊图像复原方法及系统
CN202111548604.6 2021-12-17

Publications (1)

Publication Number Publication Date
WO2023109008A1 true WO2023109008A1 (zh) 2023-06-22

Family

ID=79289316

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/093806 WO2023109008A1 (zh) 2021-12-17 2022-05-19 气动光学效应空变模糊图像复原方法及系统

Country Status (2)

Country Link
CN (1) CN113935925B (zh)
WO (1) WO2023109008A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113935925B (zh) * 2021-12-17 2022-03-15 武汉工程大学 气动光学效应空变模糊图像复原方法及系统
CN117173058B (zh) * 2023-11-03 2024-02-02 武汉工程大学 空变模糊图像统一复原方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180137606A1 (en) * 2015-05-15 2018-05-17 Peking University Shenzhen Graduate School Method and system for image de-blurring
CN111815537A (zh) * 2020-07-16 2020-10-23 西北工业大学 一种新型图像盲解去模糊方法
CN113793284A (zh) * 2021-11-17 2021-12-14 武汉工程大学 气动光学效应空变模糊图像不均匀分块的图像复原方法
CN113793285A (zh) * 2021-11-17 2021-12-14 武汉工程大学 气动光学效应目标孪生图像的超快复原方法及系统
CN113935925A (zh) * 2021-12-17 2022-01-14 武汉工程大学 气动光学效应空变模糊图像复原方法及系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9392173B2 (en) * 2013-12-13 2016-07-12 Adobe Systems Incorporated Image deblurring based on light streaks

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180137606A1 (en) * 2015-05-15 2018-05-17 Peking University Shenzhen Graduate School Method and system for image de-blurring
CN111815537A (zh) * 2020-07-16 2020-10-23 西北工业大学 一种新型图像盲解去模糊方法
CN113793284A (zh) * 2021-11-17 2021-12-14 武汉工程大学 气动光学效应空变模糊图像不均匀分块的图像复原方法
CN113793285A (zh) * 2021-11-17 2021-12-14 武汉工程大学 气动光学效应目标孪生图像的超快复原方法及系统
CN113935925A (zh) * 2021-12-17 2022-01-14 武汉工程大学 气动光学效应空变模糊图像复原方法及系统

Also Published As

Publication number Publication date
CN113935925B (zh) 2022-03-15
CN113935925A (zh) 2022-01-14

Similar Documents

Publication Publication Date Title
Luo et al. Attention-aware multi-view stereo
WO2023109008A1 (zh) 气动光学效应空变模糊图像复原方法及系统
US9196021B2 (en) Video enhancement using related content
WO2022141178A1 (zh) 图像处理方法及装置
WO2023115801A1 (zh) 气动光学效应大视场退化图像逐点校正复原方法及系统
US20220277514A1 (en) Reconstructing three-dimensional scenes portrayed in digital images utilizing point cloud machine-learning models
CN113793284B (zh) 气动光学效应空变模糊图像不均匀分块的图像复原方法
CN111681165A (zh) 图像处理方法、装置、计算机设备和计算机可读存储介质
Matsuo et al. Efficient edge-awareness propagation via single-map filtering for edge-preserving stereo matching
Li et al. Learning local Gaussian process regression for image super-resolution
CN115564639A (zh) 背景虚化方法、装置、计算机设备和存储介质
WO2022233252A1 (zh) 图像处理方法、装置、计算机设备和存储介质
Simoes et al. A framework for fast image deconvolution with incomplete observations
US11256949B2 (en) Guided sparse feature matching via coarsely defined dense matches
CN111652800A (zh) 一种单张图像超分辨的方法及计算机可读存储介质
Xu et al. Robust resistance to noise and outliers: Screened Poisson Surface Reconstruction using adaptive kernel density estimation
Buades et al. Motion-compensated spatio-temporal filtering for multi-image and multimodal super-resolution
CN115797194A (zh) 图像降噪方法、装置、电子设备、存储介质和程序产品
CN115222591A (zh) 相机设备参数无关的快速多目鱼眼图像及视频拼接方法
Xie et al. Bidirectionally aligned sparse representation for single image super-resolution
CN112150532A (zh) 图像处理的方法、装置、电子设备和计算机可读介质
CN111598806A (zh) 图像滤波方法和装置、电子设备及存储介质
CN112580475A (zh) 一种人脸检测方法及系统
Sharma An integrated optimization approach for depth map enhancement on special Riemannian manifold
Chen et al. A Self-supervised SAR Image Despeckling Strategy Based on Parameter-sharing Convolutional Neural Networks

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22905776

Country of ref document: EP

Kind code of ref document: A1