WO2020118830A1 - 字典训练及图像超分辨重建方法、系统、设备及存储介质 - Google Patents

字典训练及图像超分辨重建方法、系统、设备及存储介质 Download PDF

Info

Publication number
WO2020118830A1
WO2020118830A1 PCT/CN2019/071173 CN2019071173W WO2020118830A1 WO 2020118830 A1 WO2020118830 A1 WO 2020118830A1 CN 2019071173 W CN2019071173 W CN 2019071173W WO 2020118830 A1 WO2020118830 A1 WO 2020118830A1
Authority
WO
WIPO (PCT)
Prior art keywords
resolution
low
sample image
dictionary
image block
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2019/071173
Other languages
English (en)
French (fr)
Inventor
胡战利
梁栋
谷佩键
杨永峰
刘新
郑海荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
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 Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2020118830A1 publication Critical patent/WO2020118830A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—Two-dimensional [2D] image generation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting

Definitions

  • the invention belongs to the field of computer technology, and particularly relates to a dictionary training and image super-resolution reconstruction method, system, equipment and storage medium.
  • Computerized tomography is a cross-section scan of a certain part of the human body with a highly sensitive detector using precise collimated X-ray beams, gamma rays, ultrasound, etc., which can be used for the inspection of various diseases , Is one of the largest medical imaging diagnostic equipment in China.
  • Images obtained by low-dose CT have low resolution, usually have high noise, blurry and obvious granularity, which is unreliable for observation in clinical medicine and is very likely to cause misdiagnosis.
  • High resolution using high-dose CT, but high-dose CT will produce high-intensity radiation and cause great harm to the patient's body, and because full sampling is required, the CT time is longer, and it is easy to appear due to patient movement And caused artifacts.
  • the object of the present invention is to provide a dictionary training and image super-resolution reconstruction method, system, equipment and storage medium, aiming to solve the problem in the prior art that it is impossible to reduce the radiation dose to the patient while ensuring the CT image quality .
  • the present invention provides a dictionary acquisition method.
  • the dictionary acquisition method includes the following steps:
  • the sample image block is defined in the data A mapping relationship between the high-resolution sample feature data and the low-resolution sample feature data, the mapping relationship depends on the low-resolution sample feature data;
  • Each sample image block pair data set obtained by grouping the sample image blocks into data is separately learned using a first machine learning algorithm to obtain a temporary mapping relationship corresponding to each of the first machine learning algorithm.
  • the final solution and the low-resolution sample feature data are calculated based on the mapping relationship to calculate high-resolution temporary sample feature data, and the high-resolution temporary sample feature data and the low-resolution sample feature data constitute a temporary sample image block Data
  • the first machine learning algorithm is specifically: a decision tree algorithm, in which a node splitting function is defined, and the node splitting function uses a response function that depends on the low-resolution sample feature data as Restrictions.
  • the decision tree algorithm also defines a quality function that depends on the low-resolution sample feature data and the high-resolution sample feature data.
  • the number and depth of decision trees of the decision tree algorithm are adaptively selected according to needs.
  • the present invention also provides an image super-resolution reconstruction method.
  • the image super-resolution reconstruction method is based on the dictionary acquisition method as described above.
  • the image super-resolution reconstruction method includes:
  • the dictionary acquisition method or the image super-resolution reconstruction method is applied to an electronic computed tomography CT image, a positron emission computed tomography PET image, or a single photon emission computed tomography SPECT image.
  • the present invention provides a computing system, the system including:
  • a preprocessing unit for obtaining a high-resolution sample image and a low-resolution sample image corresponding to the high-resolution sample image; selecting a high-resolution sample image block from the high-resolution sample image, Selecting a low-resolution sample image block corresponding to the high-resolution sample image block from the resolution sample image to form a sample image block pair; for the high-resolution sample image block in the sample image block pair and the low
  • the feature sample image block of the resolution is subjected to feature extraction to obtain high-resolution sample feature data and low-resolution sample feature data constituting the sample image block pair data; and,
  • the dictionary obtaining unit is used to define the sample based on the sparse expression of the high-resolution sample image block on the high-resolution dictionary and the sparse expression of the low-resolution sample image block on the corresponding low-resolution dictionary
  • system further includes:
  • the reconstruction unit is used to obtain a low-resolution to-be-processed image; using the high-resolution dictionary and the low-resolution dictionary, image super-resolution reconstruction is performed on the low-resolution to-be-processed image.
  • the present invention also provides a computing device, including a memory and a processor, where the processor executes the computer program stored in the memory to implement the steps in the foregoing method.
  • the present invention also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method are implemented.
  • the invention uses the acquired high- and low-resolution dictionaries to reconstruct the image to obtain a high-resolution imaging image, which involves: defining and machine learning the corresponding high and low The mapping relationship between the feature data of the resolution samples; performing machine learning classification screening on the obtained multiple mapping relationships to obtain a final mapping relationship; performing machine learning with the final mapping relationship to obtain corresponding high- and low-resolution dictionaries .
  • a low-resolution imaging image obtained by imaging with a low radiation dose and a short CT time can be used to reconstruct a high-resolution image, which can reduce the occurrence of artifacts caused by patient motion, thereby ensuring the quality of the CT image while achieving Scan quickly and reduce the radiation dose.
  • Embodiment 1 is an implementation flowchart of a dictionary acquisition method provided in Embodiment 1 of the present invention
  • Embodiment 4 is a flowchart of an image super-resolution reconstruction method provided in Embodiment 4 of the present invention.
  • FIG. 6 is an enlarged view of detail A in FIG. 2;
  • FIG. 7 is an enlarged view of detail B in FIG. 3;
  • Embodiment 9 is a schematic structural diagram of a computing system provided by Embodiment 5 of the present invention.
  • FIG. 10 is a schematic structural diagram of a computing system provided in Embodiment 6 of the present invention.
  • FIG. 11 is a schematic structural diagram of a computing device according to Embodiment 7 of the present invention.
  • FIG. 1 shows the implementation process of the dictionary acquisition method provided in Embodiment 1 of the present invention.
  • FIG. 1 shows the implementation process of the dictionary acquisition method provided in Embodiment 1 of the present invention.
  • the details are as follows:
  • step S101 a high-resolution sample image and a low-resolution sample image corresponding to the high-resolution sample image are obtained.
  • a sample image set needs to be prepared, and there are sample image pairs in the set.
  • the sample image pair includes corresponding high-resolution sample images and low-resolution sample images for the same scene.
  • the high-resolution sample image and the low-resolution sample image can be obtained by performing high- and low-dose CT on the same scene, or by double-three interpolation down-sampling of the high-resolution sample image to simulate image degradation to obtain low resolution Sample images, or, through super-resolution reconstruction of low-resolution sample images, obtain high-resolution sample images, etc.
  • the quality of the sample images must meet the expected requirements as much as possible, so that a dictionary that meets the expectations can be trained.
  • step S102 a high-resolution sample image block is selected from the high-resolution sample image, and a low-resolution sample image block corresponding to the high-resolution sample image block is selected from the low-resolution sample image to form a sample image block pair .
  • the high- and low-resolution sample image blocks are also corresponding, that is, the high- and low-resolution sample image blocks correspond to the same area in the above scene.
  • step S103 feature extraction is performed on the high-resolution sample image block and the low-resolution sample image block in the sample image block pair to obtain the high-resolution sample feature data XH and the low-resolution sample constituting the sample image block pair data Characteristic data X L.
  • the sample image block pair data can be ⁇ X H , X L ⁇ .
  • the corresponding high-frequency and/or intermediate-frequency components can be extracted from the sample image block to obtain corresponding sample feature data, or the first-gradient feature and/or the second-gradient feature of the sample image patch can be obtained Corresponding sample characteristic data, etc.
  • this step may also involve the need to add corresponding feature operator processing to the extracted data, for example: multiplying the extracted data with the corresponding feature operator to finally obtain the sample feature data.
  • step S104 the same expression dictionary D H on the high-resolution sparse sparse representation of low resolution corresponding to the sample image on a low-resolution block dictionary D L sample image based on the high-resolution block, a block is defined on the sample image data mapping relation W (X L) between the sample characteristic data of high resolution and low resolution X H wherein the sample data X L, a mapping relationship W (X L) depends on the low-resolution data sample characteristics.
  • the sparse expression of the high-resolution sample image block on the high-resolution dictionary D H is:
  • ⁇ is a sparse expression.
  • the sparse expression ⁇ is the same.
  • mapping relationship W(X L ) between the high-resolution sample feature data X H and the low-resolution sample feature data X L in the sample image block pair data can be defined, and the mapping relationship W(X L ) depends on the low resolution Sample characteristic data, namely:
  • step S105 each sample image block pair data set obtained by grouping the sample image block pair data ⁇ X H , X L ⁇ is separately learned using the first machine learning algorithm to obtain each first machine learning algorithm Correspondingly, the temporary optimal solution of the mapping relationship.
  • the first machine learning algorithm may be a decision tree algorithm, K Nearest Neighbors (KNN) algorithm or other machine learning algorithms, which may divide N sample image blocks into data ⁇ X H , X L ⁇ M groups (N, M are natural numbers) of the sample image block pair data group, and the first machine learning algorithm is used to learn the data group for each sample image block to obtain M corresponding mapping relationships W (X L ) temporary optimal solution .
  • KNN K Nearest Neighbors
  • step S106 the second machine learning algorithm, the optimal solution from the set of the temporary mapping between W (X L), determining a mapping relationship W (X L) a mapping between the temporary optimal solution W (X L) Final solution.
  • step S107 the final solution W best (X L ) and the low-resolution sample feature data X L are calculated based on the mapping relationship to calculate the high-resolution temporary sample feature data X H ′ and the high-resolution temporary sample feature data X H ′ and The low-resolution sample feature data X L constitutes temporary sample image block pair data ⁇ X H ', X L ⁇ .
  • the high-resolution temporary sample feature data X H ′ can be obtained by the following formula:
  • step S108 the temporary sample image block is used to learn the data ⁇ X H ', X L ⁇ using the corresponding third machine learning algorithm to solve the high-resolution dictionary D H and the low-resolution dictionary D L , the third machine The learning algorithm defines the constraint relationships of high-resolution temporary sample feature data X H ', low-resolution sample feature data X L , high-resolution dictionary D H and low-resolution dictionary D L.
  • a definition function of the above constraint relationship may be determined:
  • the matrix E is connected between the high resolution and low resolution dictionary dictionary D H D L, ⁇ (E) is a specification of the constraints E, To take l 2 norm.
  • the corresponding high-resolution dictionary D H and low-resolution dictionary D L can be determined.
  • obtaining high and low resolution dictionaries specifically involves: by defining and machine learning the mapping relationship between corresponding high and low resolution sample feature data; performing machine learning classification screening on the obtained multiple mapping relationships to obtain A final mapping relationship; use this final mapping relationship for machine learning to obtain the corresponding high and low resolution dictionaries, and then use the acquired high and low resolution dictionaries to reconstruct the image super-resolution to obtain high-resolution imaging image.
  • a low-resolution imaging image obtained by imaging with a low radiation dose and a short CT time can be used to reconstruct a high-resolution image, which can reduce the occurrence of artifacts caused by patient motion, thereby ensuring the quality of the CT image while achieving Scan quickly and reduce the radiation dose.
  • the dictionary acquisition method of this embodiment includes calculating the above mapping relationship, the calculation amount is effectively reduced, and the running time is relatively Shorter, can improve the overall calculation efficiency.
  • this embodiment further provides the following content:
  • the first machine learning algorithm is specifically: a decision tree algorithm in which a node splitting function ⁇ (X L , ⁇ ) is defined, and the node splitting function ⁇ (X L , ⁇ ) depends on low-resolution sample feature data X L response function r ⁇ (X L) as the constraint condition, as follows:
  • ⁇ th is a threshold value
  • [] is one dimension of XL
  • the first machine learning algorithm also defines a quality function Q( ⁇ , ⁇ , X L , X H ) that depends on the low-resolution sample feature data X L and the high-resolution sample feature data X H , and thus serves as a feature selection
  • Q quality function
  • Le and Ri as the left and right nodes of the decision tree split downward, are the choices made according to the relevant constraints and conditions to achieve the effect of classifier classification, namely:
  • ⁇ is a selection function, which can be a basis function.
  • ⁇ is a selection function, which can be a basis function.
  • the machine learning in this article belongs to the weak classifier, and its basis function cannot be too complicated otherwise the calculation amount Very large, just use a general linear function.
  • this embodiment further provides the following content:
  • the number n and depth ⁇ max of the decision tree algorithm can be adaptively selected according to needs.
  • n the number of decision trees n is different, and the effect of obtaining the final result is also different.
  • the larger the n, the better the effect, but it will also reach a saturation state, which is about n 10. Is the same, and its node splitting function ⁇ (X L , ⁇ ) is the same.
  • This embodiment provides an image super-resolution reconstruction method.
  • the image super-resolution reconstruction method is based on the dictionary acquisition method as described above.
  • the image super-resolution reconstruction method mainly includes the process shown in FIG. 4:
  • step S401 a low-resolution image to be processed is obtained.
  • step S402 the high-resolution dictionary D H and the low-resolution dictionary D L acquired in the dictionary acquisition method are used to perform image super-resolution reconstruction on the low-resolution to-be-processed image.
  • the image super-resolution reconstruction of the low-resolution to-be-processed image can be directly processed by using the high-resolution dictionary D H and the low-resolution dictionary D L acquired in the above dictionary acquisition method, which may include:
  • the image to be processed with resolution is divided into blocks, feature extraction, the sparse expression ⁇ is obtained when the cost function meets the minimum, and then the sparse expression ⁇ and the high-resolution dictionary D H are used to obtain high-resolution imaging image feature data, and finally the high For a resolution imaged image, the cost function can be as follows:
  • the above dictionary acquisition method or image super-resolution reconstruction method can be applied not only to CT images, but also to positron emission computed tomography (Positron Emission Computerized Tomography, PET) images or single photon emission computed tomography (Single Photon Emission) Computed Tomography (SPECT) image processing.
  • PET positron emission computed tomography
  • SPECT Single Photon emission computed tomography
  • the low-resolution to-be-processed image shown in FIG. 3 can be reconstructed into the high-resolution imaging shown in FIG. 5 by the image super-resolution reconstruction method of this embodiment.
  • the image, the high-resolution imaging image shown in FIG. 5 and the high-resolution sample image shown in FIG. 2 are of comparable quality.
  • FIG. 6, FIG. 7 and FIG. 8 for detailed enlarged views of corresponding positions in FIG. 2, FIG. 3 and FIG. 5 respectively.
  • the image super-resolution reconstruction of the low-resolution to-be-processed image may further include the processes of the high-resolution dictionary D H and the low-resolution dictionary D L acquired in the above dictionary acquisition method.
  • FIG. 9 shows the structure of the computing system provided by Embodiment 5 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown, including:
  • the pre-processing unit 901 is used to obtain a high-resolution sample image and a low-resolution sample image corresponding to the high-resolution sample image; select a high-resolution sample image block from the high-resolution sample image and select a low-resolution sample image from the low-resolution sample image Select the low-resolution sample image block corresponding to the high-resolution sample image block to form a sample image block pair; perform feature extraction on the high-resolution sample image block and the low-resolution sample image block in the sample image block pair to obtain the composition High-resolution sample feature data and low-resolution sample feature data of the sample image block pair data; and,
  • the dictionary acquisition unit 902 is used to define that the sparse expression of the high-resolution sample image block on the high-resolution dictionary is consistent with the sparse expression of the low-resolution sample image block on the corresponding low-resolution dictionary, and define that the sample image block has a high The mapping relationship between the resolution sample feature data and the low resolution sample feature data.
  • the mapping relationship depends on the low resolution sample feature data; each sample image block pair data group obtained by grouping the sample image block pair data, respectively.
  • the first machine learning algorithm is used for learning to obtain the temporary optimal solution of the mapping relationship corresponding to each of the first machine learning algorithms; through the second machine learning algorithm, from the set of temporary optimal solutions of the mapping relationship, a The temporary optimal solution of the mapping relationship is used as the final solution of the mapping relationship; the final solution of the mapping relationship and the low-resolution sample feature data are used to calculate the high-resolution temporary sample feature data, the high-resolution temporary sample feature data and the low-resolution sample feature data.
  • Temporary sample image block pair data; the temporary sample image block pair data is learned using the corresponding third machine learning algorithm to solve the high-resolution dictionary and the low-resolution dictionary.
  • a high-resolution temporary sample is established in the third machine learning algorithm Constraint relationship of feature data, low-resolution sample feature data, high-resolution dictionary and low-resolution dictionary.
  • each unit may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit, which is not used to limit the present invention.
  • Each unit implements the steps in the above corresponding method embodiments, such as steps S101 to S108 shown in FIG. 1, and details are not repeated here.
  • this embodiment further provides the following content:
  • the computing system further includes:
  • the reconstruction unit 1001 is used to obtain a low-resolution to-be-processed image; the high-resolution dictionary and the low-resolution dictionary are used to perform image super-resolution reconstruction on the low-resolution to-be-processed image.
  • each unit may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit, which is not used to limit the present invention.
  • Each unit implements the steps in the above corresponding method embodiments, for example, steps S401 to S402 shown in FIG. 4, and details are not repeated here.
  • FIG. 11 shows the structure of the computing device provided in Embodiment 7 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown.
  • the computing device in the embodiment of the present invention includes a processor 1101 and a memory 1102.
  • the processor 1101 executes the computer program 1103 stored in the memory 1102
  • the steps in the foregoing method embodiments are implemented, for example, steps S101 to S108 shown in FIG.
  • the processor 1101 executes the computer program 1103
  • the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 901 to 902 shown in FIG. 9.
  • the computing device in this embodiment of the present invention may be a computer or a computer network system.
  • the steps implemented when the processor 1101 in the computing device executes the computer program 1103 to implement the above methods reference may be made to the description of the foregoing method embodiments, and details are not described herein again.
  • a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method embodiments are implemented, for example, FIG. 1 Steps S101 to S108 shown.
  • the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 901 to 902 shown in FIG. 9.
  • the computer-readable storage medium in the embodiments of the present invention may include any entity or device capable of carrying computer program code, and a recording medium, such as ROM/RAM, magnetic disk, optical disk, flash memory, and other memories.

Landscapes

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

Abstract

一种字典训练及图像超分辨重建方法、系统、设备及存储介质，该方法包括：获得低分辨率待处理图像后，利用所获取的高、低分辨率字典对其进行图像超分辨重建，得到高分辨率成像图像，其中涉及：通过定义并机器学习相应的高、低分辨率样本特征数据之间的映射关系；对得到的多个映射关系进行机器学习分类筛选，得到一个最终的映射关系；以该最终的映射关系进行机器学习，得到相应的高、低分辨率字典。这样，可采用低辐射剂量、短CT时间成像得到的低分辨率成像图像，重建得到高分辨率图像，可减少因病人运动而引起的伪影现象的发生，从而在保证CT图像质量的同时实现快速扫描，且降低辐射剂量。

Description

字典训练及图像超分辨重建方法、系统、设备及存储介质 技术领域
本发明属于计算机技术领域，尤其涉及一种字典训练及图像超分辨重建方法、系统、设备及存储介质。
背景技术
电子计算机断层扫描(Computed Tomography，CT)是利用精确准直的X线束、γ射线、超声波等，与灵敏度极高的探测器一同围绕人体的某一部位作断面扫描，可用于多种疾病的检查，是我国装机量最大的医疗影像诊断设备之一。
通过低剂量的CT得到的图像分辨率低，通常会具有高噪音、较为模糊且颗粒化明显，这对于临床医学上的观察是不可靠的，非常有可能造成误诊的情况，目前为保证CT图像高分辨率，采用的是高剂量的CT，但高剂量CT会产生高强度的辐射而对病人身体产生极大的伤害，并且由于需要全采样，因此CT时间较长，同时容易出现因病人运动而引起的伪影现象。
发明内容
本发明的目的在于提供一种字典训练及图像超分辨重建方法、系统、设备及存储介质，旨在解决现有技术所存在的、无法在保证CT图像质量的同时降低对病人的辐射剂量的问题。
一方面，本发明提供了一种字典获取方法，所述字典获取方法包括下述步骤：
获得高分辨率样本图像以及与所述高分辨率样本图像对应的低分辨率样本图像；
从所述高分辨率样本图像中选取高分辨率样本图像块，从所述低分辨率样本图像中选取与所述高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对；
对所述样本图像块对中的高分辨率样本图像块及所述低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据及低分辨率样本特征数据；
基于所述高分辨率样本图像块在高分辨率字典上的稀疏表达与所述低分辨率样本图像块在对应的低分辨率字典上的稀疏表达一致，定义所述样本图像块对数据中所述高分辨率样本特征数据与所述低分辨率样本特征数据之间的映射关系，所述映射关系依赖于所述低分辨率样本特征数据；
将所述样本图像块对数据分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一所述第一机器学习算法所对应的、映射关系临时最优解；
通过第二机器学习算法，从所述映射关系临时最优解的集合中，确定一所述映射关系临时最优解作为映射关系最终解；
以所述映射关系最终解及所述低分辨率样本特征数据，计算得到高分辨率临时样本特征数据，所述高分辨率临时样本特征数据以及所述低分辨率样本特征数据组成临时样本图像块对数据；
将所述临时样本图像块对数据采用相应的第三机器学习算法进行学习，解出所述高分辨率字典及所述低分辨率字典，所述第三机器学习算法中定义有所述高分辨率临时样本特征数据、所述低分辨率样本特征数据、所述高分辨率字典及所述低分辨率字典的约束关系。
进一步的，所述第一机器学习算法具体为：决策树算法，所述决策树算法中定义了节点分裂函数，所述节点分裂函数以一依赖于所述低分辨率样本特征数据的响应函数作为约束条件。
进一步的，所述决策树算法中还定义了一依赖于所述低分辨率样本特征数 据及所述高分辨率样本特征数据的质量函数。
进一步的，所述决策树算法的决策树数量及深度因需适应性选择。
另一方面，本发明还提供了一种图像超分辨重建方法，所述图像超分辨重建方法基于如上述的字典获取方法，所述图像超分辨重建方法包括：
获得低分辨率待处理图像；
利用所述高分辨率字典及所述低分辨率字典，对所述低分辨率待处理图像进行图像超分辨重建。
进一步的，所述字典获取方法或所述图像超分辨重建方法应用于电子计算机断层扫描CT图像、正电子发射型计算机断层显像PET图像或单光子发射型计算机断层成像SPECT图像。
另一方面，本发明提供了一种计算系统，所述系统包括：
预处理单元，用于获得高分辨率样本图像以及与所述高分辨率样本图像对应的低分辨率样本图像；从所述高分辨率样本图像中选取高分辨率样本图像块，从所述低分辨率样本图像中选取与所述高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对；对所述样本图像块对中的高分辨率样本图像块及所述低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据及低分辨率样本特征数据；以及，
字典获取单元，用于基于所述高分辨率样本图像块在高分辨率字典上的稀疏表达与所述低分辨率样本图像块在对应的低分辨率字典上的稀疏表达一致，定义所述样本图像块对数据中所述高分辨率样本特征数据与所述低分辨率样本特征数据之间的映射关系，所述映射关系依赖于所述低分辨率样本特征数据；将所述样本图像块对数据分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一所述第一机器学习算法所对应的、映射关系临时最优解；通过第二机器学习算法，从所述映射关系临时最优解的集合中，确定一所述映射关系临时最优解作为映射关系最终解；以所述映射关系最终解及所述低分辨率样本特征数据，计算得到高分辨率临时样本特征数据，所 述高分辨率临时样本特征数据以及所述低分辨率样本特征数据组成临时样本图像块对数据；将所述临时样本图像块对数据采用相应的第三机器学习算法进行学习，解出所述高分辨率字典及所述低分辨率字典，所述第三机器学习算法中建立了所述高分辨率临时样本特征数据、所述低分辨率样本特征数据、所述高分辨率字典及所述低分辨率字典的约束关系。
进一步的，所述系统还包括：
重建单元，用于获得低分辨率待处理图像；利用所述高分辨率字典及所述低分辨率字典，对所述低分辨率待处理图像进行图像超分辨重建。
另一方面，本发明还提供了一种计算设备，包括存储器及处理器，所述处理器执行所述存储器中存储的计算机程序时实现如上述方法中的步骤。
另一方面，本发明还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如上述方法中的步骤。
本发明在获得低分辨率待处理图像后，利用所获取的高、低分辨率字典对其进行图像超分辨重建，得到高分辨率成像图像，其中涉及：通过定义并机器学习相应的高、低分辨率样本特征数据之间的映射关系；对得到的多个映射关系进行机器学习分类筛选，得到一个最终的映射关系；以该最终的映射关系进行机器学习，得到相应的高、低分辨率字典。这样，可采用低辐射剂量、短CT时间成像得到的低分辨率成像图像，重建得到高分辨率图像，可减少因病人运动而引起的伪影现象的发生，从而在保证CT图像质量的同时实现快速扫描，且降低辐射剂量。
附图说明
图1是本发明实施例一提供的字典获取方法的实现流程图；
图2是本发明实施例所涉及的高分辨率样本图像；
图3是本发明实施例所涉及的低分辨率样本图像；
图4是本发明实施例四提供的图像超分辨重建方法的实现流程图；
图5是本发明实施例所涉及的高分辨率成像图像；
图6是图2中的细节A放大图；
图7是图3中的细节B放大图；
图8是图5中的细节C放大图；
图9是本发明实施例五提供的计算系统的结构示意图；
图10是本发明实施例六提供的计算系统的结构示意图；
图11是本发明实施例七提供的计算设备的结构示意图。
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仅用以解释本发明，并不用于限定本发明。
以下结合具体实施例对本发明的具体实现进行详细描述：
实施例一：
图1示出了本发明实施例一提供的字典获取方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S101中，获得高分辨率样本图像以及与高分辨率样本图像对应的低分辨率样本图像。
本实施例中，在进行字典获取处理前，需要准备样本图像集合，该集合中存在样本图像对，样本图像对包含针对于同一场景的相对应的高分辨率样本图像以及低分辨率样本图像，如图2及图3所示。高分辨率样本图像以及低分辨率样本图像的获得，可以通过对同一场景进行高、低剂量CT获得，或者，通过对高分辨率样本图像进行双三次插值下采样模拟图像退化，得到低分辨率样本图像，或者，通过低分辨率样本图像的超分辨重建，得到高分辨率样本图像等。不论采用何种途径得到高、低分辨率样本图像，样本图像的质量要尽可能 满足期望要求，这样才能训练出符合期望的字典。
在步骤S102中，从高分辨率样本图像中选取高分辨率样本图像块，从低分辨率样本图像中选取与高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对。
本实施例中，高、低分辨率样本图像块也是相对应的，也就是说，高、低分辨率样本图像块所对应的是上述场景中的同一区域。
在步骤S103中，对样本图像块对中的高分辨率样本图像块及低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据X H及低分辨率样本特征数据X L。样本图像块对数据可为{X H，X L}。
本实施例中，可对样本图像块进行相应的高频和/或中频分量的提取，得到相应的样本特征数据，或者，对样本图像块的一阶梯度特征和/或二阶梯度特征，得到相应的样本特征数据等。
相应的，本步骤中可能还涉及需要对提取得到的数据加入相应的特征算子处理，例如：将提取得到的数据与相应的特征算子相乘，最终才得到样本特征数据。
在步骤S104中，基于高分辨率样本图像块在高分辨率字典D H上的稀疏表达与低分辨率样本图像块在对应的低分辨率字典D L上的稀疏表达一致，定义样本图像块对数据中高分辨率样本特征数据X H与低分辨率样本特征数据X L之间的映射关系W(X L)，映射关系W(X L)依赖于低分辨率样本特征数据。
本实施例中，高分辨率样本图像块在高分辨率字典D H上的稀疏表达为：
X H＝D H×α
其中，α为稀疏表达式。
而低分辨率样本图像块在低分辨率字典D L上的稀疏表达为：
X L＝D L×α
上述两个稀疏表达中，稀疏表达式α一致。
基于此，可定义样本图像块对数据中高分辨率样本特征数据X H与低分辨率样本特征数据X L之间的映射关系W(X L)，映射关系W(X L)依赖于低分辨率样本特征数据，即：
X H＝X L×W(X L)
在步骤S105中，将样本图像块对数据{X H，X L}分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一第一机器学习算法所对应的、映射关系临时最优解。
本实施例中，第一机器学习算法可以为决策树算法、K最临近(K Nearest Neighbors，KNN)算法或其他机器学习算法，可将N个样本图像块对数据{X H，X L}分成M组(N、M为自然数)样本图像块对数据组，利用第一机器学习算法对每一样本图像块对数据组进行学习，得到M个相应的映射关系W(X L)临时最优解。
在步骤S106中，通过第二机器学习算法，从映射关系W(X L)临时最优解的集合中，确定一映射关系W(X L)临时最优解作为映射关系W(X L)最终解。
本实施例中，当得到M个相应的映射关系W(X L)临时最优解时，需要从这些临时最优解中，选择一个作为映射关系W(X L)最终解，具体可通过投票选择算法，以大部分临时最优解所最接近的一临时最优解，作为映射关系最终解W best(X L)。
在步骤S107中，以映射关系最终解W best(X L)及低分辨率样本特征数据X L，计算得到高分辨率临时样本特征数据X H’，高分辨率临时样本特征数据X H’以及低分辨率样本特征数据X L组成临时样本图像块对数据{X H’，X L}。
本实施例中，可通过下式，得到高分辨率临时样本特征数据X H’：
X H’＝W best(X L)×X L
在步骤S108中，将临时样本图像块对数据{X H’，X L}采用相应的第三机 器学习算法进行学习，解出高分辨率字典D H及低分辨率字典D L，第三机器学习算法中定义有高分辨率临时样本特征数据X H’、低分辨率样本特征数据X L、高分辨率字典D H及低分辨率字典D L的约束关系。
本实施例中，可确定一上述约束关系的定义函数：
Figure PCTCN2019071173-appb-000001
其中，E为高分辨率字典D H及低分辨率字典D L之间的连接矩阵，Γ(E)为E的规范约束，
Figure PCTCN2019071173-appb-000002
为取l 2范数。
当上述定义函数满足最小值时，即可确定出相应的高分辨率字典D H及低分辨率字典D L。
实施本实施例，获取高、低分辨率字典具体涉及：通过定义并机器学习相应的高、低分辨率样本特征数据之间的映射关系；对得到的多个映射关系进行机器学习分类筛选，得到一个最终的映射关系；以该最终的映射关系进行机器学习，得到相应的高、低分辨率字典，进而利用所获取的高、低分辨率字典对其进行图像超分辨重建，得到高分辨率成像图像。这样，可采用低辐射剂量、短CT时间成像得到的低分辨率成像图像，重建得到高分辨率图像，可减少因病人运动而引起的伪影现象的发生，从而在保证CT图像质量的同时实现快速扫描，且降低辐射剂量。另外，由于上述字典的获取与传统的、计算量偏大且运行时间过长的字典训练方法不同，本实施例的字典获取方法包含的是计算上述映射关系，计算量得到有效减少，运行时间相对偏短，可提高整体计算效率。
实施例二：
本实施例在实施例一基础上，进一步提供了如下内容：
第一机器学习算法具体为：决策树算法，该决策树算法中定义了节点分裂函数δ(X L,θ)，节点分裂函数δ(X L,θ)以一依赖于低分辨率样本特征数据X L的响应函数r θ(X L)作为约束条件，具体如下：
Figure PCTCN2019071173-appb-000003
r θ(X L)＝X L[θ 1]-θ th，
其中，Θth为一门限值，[]为取XL的一维，
Figure PCTCN2019071173-appb-000004
第一机器学习算法中还定义了一依赖于低分辨率样本特征数据X L及高分辨率样本特征数据X H的质量函数Q(δ,θ,X L,X H)，由此作为特征选择的依据，具体如下：
Figure PCTCN2019071173-appb-000005
其中，Le、Ri作为决策树向下分裂的左、右节点，即为根据相关的约束和条件作出的选择以达到分类器分类的效果，即：
Figure PCTCN2019071173-appb-000006
Figure PCTCN2019071173-appb-000007
其中，
Figure PCTCN2019071173-appb-000008
为正则表达式，σ是个选择函数，可为一个基函数，一般机器学习中需要通过定义基函数来完成分类器工作，本文中的机器学习属于弱分类器，其基函数不能过于复杂否则计算量极大，采用一般的线性函数即可。
实施例三：
本实施例在实施例二基础上，进一步提供了如下内容：
本实施例中，决策树算法的决策树数量n及深度ξ max可因需适应性选择。
在具体应用中，决策树数量n不同，得到最后结果的效果也不同，n越大 效果越好，但是也会达到一个饱和状态，大概在n＝10左右即达到饱和，每个决策树的算法是相同的，其节点分裂函数δ(X L,θ)是同一个。
实施例四：
本实施例提供了一种图像超分辨重建方法，该图像超分辨重建方法基于如上述的字典获取方法，图像超分辨重建方法主要包括如图4所示的流程：
在步骤S401中，获得低分辨率待处理图像。
在步骤S402中，利用字典获取方法中获取的高分辨率字典D H及低分辨率字典D L，对低分辨率待处理图像进行图像超分辨重建。
本实施例中，对低分辨率待处理图像进行图像超分辨重建可直接利用上述字典获取方法中获取的高分辨率字典D H及低分辨率字典D L来进行处理，具体可包括：对低分辨率待处理图像进行分块、特征提取、在代价函数满足最小时得到稀疏表达式α、再由稀疏表达式α及高分辨率字典D H得到高分辨率成像图像特征数据，最终重建出高分辨率成像图像，代价函数可如下式：
Figure PCTCN2019071173-appb-000009
其中，||·|| 1为取l 1范数，λ为常量参数。
上述字典获取方法或图像超分辨重建方法不仅可应用于CT图像，还可以应用于正电子发射型计算机断层显像(Positron Emission Computed Tomography，PET)图像或单光子发射型计算机断层成像(Single Photon Emission Computed Tomography，SPECT)图像等的处理。
结合图2、图3及图5，可以发现：通过本实施例的图像超分辨重建方法处理，如图3所示的低分辨率待处理图像可重建为如图5所示的高分辨率成像图像，如图5所示的高分辨率成像图像与如图2所示的高分辨率样本图像质量相当。图2、图3及图5中对应位置的细节放大图可分别参考图6、图7及图8。
对低分辨率待处理图像进行图像超分辨重建还可包括上述字典获取方法中获取的高分辨率字典D H及低分辨率字典D L的过程。
实施例五：
图9示出了本发明实施例五提供的计算系统的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
预处理单元901，用于获得高分辨率样本图像以及与高分辨率样本图像对应的低分辨率样本图像；从高分辨率样本图像中选取高分辨率样本图像块，从低分辨率样本图像中选取与高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对；对样本图像块对中的高分辨率样本图像块及低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据及低分辨率样本特征数据；以及，
字典获取单元902，用于基于高分辨率样本图像块在高分辨率字典上的稀疏表达与低分辨率样本图像块在对应的低分辨率字典上的稀疏表达一致，定义样本图像块对数据中高分辨率样本特征数据与低分辨率样本特征数据之间的映射关系，映射关系依赖于低分辨率样本特征数据；将样本图像块对数据分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一所述第一机器学习算法所对应的、映射关系临时最优解；通过第二机器学习算法，从映射关系临时最优解的集合中，确定一映射关系临时最优解作为映射关系最终解；以映射关系最终解及低分辨率样本特征数据，计算得到高分辨率临时样本特征数据，高分辨率临时样本特征数据以及低分辨率样本特征数据组成临时样本图像块对数据；将临时样本图像块对数据采用相应的第三机器学习算法进行学习，解出高分辨率字典及低分辨率字典，第三机器学习算法中建立了高分辨率临时样本特征数据、低分辨率样本特征数据、高分辨率字典及低分辨率字典的约束关系。
在本发明实施例中，各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。各单元实现上述相应方法实施例中的步骤，例如图1所示的步骤S101至S108，具体此处不再赘述。
实施例六：
本实施例在实施例五基础上，进一步提供了如下内容：
如图10所示，本实施例中，计算系统还包括：
重建单元1001，用于获得低分辨率待处理图像；利用高分辨率字典及低分辨率字典，对低分辨率待处理图像进行图像超分辨重建。
同样，在本发明实施例中，各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。各单元实现上述相应方法实施例中的步骤，例如图4所示的步骤S401至S402，具体此处不再赘述。
实施例七：
图11示出了本发明实施例七提供的计算设备的结构，为了便于说明，仅示出了与本发明实施例相关的部分。
本发明实施例的计算设备包括处理器1101及存储器1102，处理器1101执行存储器1102中存储的计算机程序1103时实现上述各个方法实施例中的步骤，例如图1所示的步骤S101至S108。或者，处理器1101执行计算机程序1103时实现上述各装置实施例中各单元的功能，例如图9所示单元901至902的功能。
本发明实施例的计算设备可以为计算机或计算机网络系统。该计算设备中处理器1101执行计算机程序1103时实现上述各方法时实现的步骤可参考前述方法实施例的描述，在此不再赘述。
实施例八：
在本发明实施例中，提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序被处理器执行时实现上述各方法实施例中的步骤，例如，图1所示的步骤S101至S108。或者，该计算机程序被处理器执行时实现上述各装置实施例中各单元的功能，例如图9所示单元901至902的功能。
本发明实施例的计算机可读存储介质可以包括能够携带计算机程序代码的 任何实体或装置、记录介质，例如，ROM/RAM、磁盘、光盘、闪存等存储器。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种字典获取方法，其特征在于，所述字典获取方法包括下述步骤：
    获得高分辨率样本图像以及与所述高分辨率样本图像对应的低分辨率样本图像；
    从所述高分辨率样本图像中选取高分辨率样本图像块，从所述低分辨率样本图像中选取与所述高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对；
    对所述样本图像块对中的高分辨率样本图像块及所述低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据及低分辨率样本特征数据；
    基于所述高分辨率样本图像块在高分辨率字典上的稀疏表达与所述低分辨率样本图像块在对应的低分辨率字典上的稀疏表达一致，定义所述样本图像块对数据中所述高分辨率样本特征数据与所述低分辨率样本特征数据之间的映射关系，所述映射关系依赖于所述低分辨率样本特征数据；
    将所述样本图像块对数据分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一所述第一机器学习算法所对应的、映射关系临时最优解；
    通过第二机器学习算法，从所述映射关系临时最优解的集合中，确定一所述映射关系临时最优解作为映射关系最终解；
    以所述映射关系最终解及所述低分辨率样本特征数据，计算得到高分辨率临时样本特征数据，所述高分辨率临时样本特征数据以及所述低分辨率样本特征数据组成临时样本图像块对数据；
    将所述临时样本图像块对数据采用相应的第三机器学习算法进行学习，解出所述高分辨率字典及所述低分辨率字典，所述第三机器学习算法中定义有所述高分辨率临时样本特征数据、所述低分辨率样本特征数据、所述高分辨率字典及所述低分辨率字典的约束关系。
  2. 如权利要求1所述的字典获取方法，其特征在于，所述第一机器学习算法具体为：决策树算法，所述决策树算法中定义了节点分裂函数，所述节点分裂函数以一依赖于所述低分辨率样本特征数据的响应函数作为约束条件。
  3. 如权利要求2所述的字典获取方法，其特征在于，所述决策树算法中还定义了一依赖于所述低分辨率样本特征数据及所述高分辨率样本特征数据的质量函数。
  4. 如权利要求2所述的字典获取方法，其特征在于，所述决策树算法的决策树数量及深度因需适应性选择。
  5. 一种图像超分辨重建方法，其特征在于，所述图像超分辨重建方法基于如权利要求1至4中任一项所述的字典获取方法，所述图像超分辨重建方法包括：
    获得低分辨率待处理图像；
    利用所述高分辨率字典及所述低分辨率字典，对所述低分辨率待处理图像进行图像超分辨重建。
  6. 如权利要求1至5任一项所述的字典获取方法或图像超分辨重建方法，其特征在于，所述字典获取方法或所述图像超分辨重建方法应用于电子计算机断层扫描CT图像、正电子发射型计算机断层显像PET图像或单光子发射型计算机断层成像SPECT图像。
  7. 一种计算系统，其特征在于，所述系统包括：
    预处理单元，用于获得高分辨率样本图像以及与所述高分辨率样本图像对应的低分辨率样本图像；从所述高分辨率样本图像中选取高分辨率样本图像块，从所述低分辨率样本图像中选取与所述高分辨率样本图像块对应的低分辨率样本图像块，以组成样本图像块对；对所述样本图像块对中的高分辨率样本图像块及所述低分辨率样本图像块进行特征提取，得到组成样本图像块对数据的高分辨率样本特征数据及低分辨率样本特征数据；以及，
    字典获取单元，用于基于所述高分辨率样本图像块在高分辨率字典上的稀 疏表达与所述低分辨率样本图像块在对应的低分辨率字典上的稀疏表达一致，定义所述样本图像块对数据中所述高分辨率样本特征数据与所述低分辨率样本特征数据之间的映射关系，所述映射关系依赖于所述低分辨率样本特征数据；将所述样本图像块对数据分组后所得到的每个样本图像块对数据组，分别采用第一机器学习算法进行学习，得到每一所述第一机器学习算法所对应的、映射关系临时最优解；通过第二机器学习算法，从所述映射关系临时最优解的集合中，确定一所述映射关系临时最优解作为映射关系最终解；以所述映射关系最终解及所述低分辨率样本特征数据，计算得到高分辨率临时样本特征数据，所述高分辨率临时样本特征数据以及所述低分辨率样本特征数据组成临时样本图像块对数据；将所述临时样本图像块对数据采用相应的第三机器学习算法进行学习，解出所述高分辨率字典及所述低分辨率字典，所述第三机器学习算法中建立了所述高分辨率临时样本特征数据、所述低分辨率样本特征数据、所述高分辨率字典及所述低分辨率字典的约束关系。
  8. 如权利要求7所述的系统，其特征在于，所述系统还包括：
    重建单元，用于获得低分辨率待处理图像；利用所述高分辨率字典及所述低分辨率字典，对所述低分辨率待处理图像进行图像超分辨重建。
  9. 一种计算设备，包括存储器及处理器，其特征在于，所述处理器执行所述存储器中存储的计算机程序时实现如权利要求1至6任一项所述方法中的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至6任一项所述方法中的步骤。
PCT/CN2019/071173 2018-12-12 2019-01-10 字典训练及图像超分辨重建方法、系统、设备及存储介质 Ceased WO2020118830A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811519177.7 2018-12-12
CN201811519177.7A CN109741254B (zh) 2018-12-12 2018-12-12 字典训练及图像超分辨重建方法、系统、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020118830A1 true WO2020118830A1 (zh) 2020-06-18

Family

ID=66358881

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/071173 Ceased WO2020118830A1 (zh) 2018-12-12 2019-01-10 字典训练及图像超分辨重建方法、系统、设备及存储介质

Country Status (2)

Country Link
CN (1) CN109741254B (zh)
WO (1) WO2020118830A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113628109A (zh) * 2021-07-16 2021-11-09 上海交通大学 基于可学习字典的人脸五官超分辨率方法和系统、介质

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325695B (zh) * 2020-02-29 2023-04-07 深圳先进技术研究院 基于多剂量等级的低剂量图像增强方法、系统及存储介质
CN111985430B (zh) * 2020-08-27 2025-01-28 深圳前海微众银行股份有限公司 图像数据标注方法、装置、设备及可读存储介质
CN113450427B (zh) * 2021-06-29 2023-09-01 深圳高性能医疗器械国家研究院有限公司 基于联合字典学习和深度网络的pet图像重建方法
CN113920215A (zh) * 2021-11-12 2022-01-11 中国人民解放军战略支援部队信息工程大学 基于稀疏表示的双能ct超分辨图像重建方法及系统
CN114926559B (zh) * 2022-05-09 2025-06-03 浙江大学 一种基于字典学习思想无衰减校正的pet重建方法
CN115841866B (zh) * 2023-02-20 2023-05-23 北京唯迈医疗设备有限公司 一种低辐射剂量图像的生成方法、终端设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106296583A (zh) * 2016-08-01 2017-01-04 西北工业大学 基于图像块组稀疏编码与成对映射的含噪高光谱图像超分辨率重构方法
CN107133915A (zh) * 2017-04-21 2017-09-05 西安科技大学 一种基于学习的图像超分辨率重构方法
CN107341765A (zh) * 2017-05-05 2017-11-10 西安邮电大学 一种基于卡通纹理分解的图像超分辨率重建方法
CN107492070A (zh) * 2017-07-10 2017-12-19 华北电力大学 一种双通道卷积神经网络的单图像超分辨率计算方法
WO2018120329A1 (zh) * 2016-12-28 2018-07-05 深圳市华星光电技术有限公司 基于稀疏域重构的单帧图像超分辨重建方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5769241B2 (ja) * 2011-07-15 2015-08-26 国立大学法人 筑波大学 超解像画像処理装置及び超解像画像処理用辞書作成装置
EP2927864B1 (en) * 2012-11-29 2018-06-13 NEC Corporation Image processing device and image processing method
US10339633B2 (en) * 2015-11-04 2019-07-02 Peking University Shenzhen Graduate School Method and device for super-resolution image reconstruction based on dictionary matching
CN107767337A (zh) * 2017-09-29 2018-03-06 深圳先进技术研究院 一种图像超分辨率重建方法、装置、移动终端及存储介质
CN108550111B (zh) * 2018-03-30 2021-12-03 西安工程大学 一种基于多级字典学习的残差实例回归超分辨重建方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106296583A (zh) * 2016-08-01 2017-01-04 西北工业大学 基于图像块组稀疏编码与成对映射的含噪高光谱图像超分辨率重构方法
WO2018120329A1 (zh) * 2016-12-28 2018-07-05 深圳市华星光电技术有限公司 基于稀疏域重构的单帧图像超分辨重建方法及装置
CN107133915A (zh) * 2017-04-21 2017-09-05 西安科技大学 一种基于学习的图像超分辨率重构方法
CN107341765A (zh) * 2017-05-05 2017-11-10 西安邮电大学 一种基于卡通纹理分解的图像超分辨率重建方法
CN107492070A (zh) * 2017-07-10 2017-12-19 华北电力大学 一种双通道卷积神经网络的单图像超分辨率计算方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113628109A (zh) * 2021-07-16 2021-11-09 上海交通大学 基于可学习字典的人脸五官超分辨率方法和系统、介质
CN113628109B (zh) * 2021-07-16 2023-11-28 上海交通大学 基于可学习字典的人脸五官超分辨率方法和系统、介质

Also Published As

Publication number Publication date
CN109741254A (zh) 2019-05-10
CN109741254B (zh) 2022-09-27

Similar Documents

Publication Publication Date Title
CN109741254B (zh) 字典训练及图像超分辨重建方法、系统、设备及存储介质
CN109146988B (zh) 基于vaegan的非完全投影ct图像重建方法
CN112348936B (zh) 一种基于深度学习的低剂量锥束ct图像重建方法
CN111325686B (zh) 一种基于深度学习的低剂量pet三维重建方法
CN109493328B (zh) 医学图像显示方法、查看设备以及计算机设备
CN107481297B (zh) 一种基于卷积神经网络的ct图像重建方法
CN113516586B (zh) 一种低剂量ct图像超分辨率去噪方法和装置
He et al. Downsampled imaging geometric modeling for accurate CT reconstruction via deep learning
CN110097611A (zh) 图像重建方法、装置、设备及存储介质
Li et al. Incorporation of residual attention modules into two neural networks for low‐dose CT denoising
CN112435164A (zh) 基于多尺度生成对抗网络的低剂量ct肺部图像的同时超分辨率和去噪方法
CN109934887A (zh) 一种基于改进的脉冲耦合神经网络的医学图像融合方法
CN111325695B (zh) 基于多剂量等级的低剂量图像增强方法、系统及存储介质
Marhamati et al. LAIU-Net: A learning-to-augment incorporated robust U-Net for depressed humans’ tongue segmentation
US20220044454A1 (en) Deep reinforcement learning for computer assisted reading and analysis
CN116309806A (zh) 一种基于CSAI-Grid RCNN的甲状腺超声图像感兴趣区域定位方法
CN111242947A (zh) 一种ct扫描图像质量评价方法、计算机可读存储介质、ct扫描装置
US12597189B2 (en) Methods and apparatus for synthetic computed tomography image generation
US20210074034A1 (en) Methods and apparatus for neural network based image reconstruction
CN109658464B (zh) 基于加权核范数极小的稀疏角ct图像重建方法
Chan et al. An attention-based deep convolutional neural network for ultra-sparse-view CT reconstruction
CN111489406A (zh) 生成高能ct图像模型的训练及生成方法、设备、存储介质
CN111724452B (zh) 一种低剂量ct图像重建方法
CN115861470A (zh) 图像伪影校正方法、装置、设备、存储介质和程序产品
CN116342726A (zh) 基于结构增强与伪影估计的稀疏角度ct图像伪影抑制方法

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: 19896424

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.11.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19896424

Country of ref document: EP

Kind code of ref document: A1