WO2020124682A1 - 图像处理方法、装置、设备和存储介质 - Google Patents

图像处理方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2020124682A1
WO2020124682A1 PCT/CN2018/125629 CN2018125629W WO2020124682A1 WO 2020124682 A1 WO2020124682 A1 WO 2020124682A1 CN 2018125629 W CN2018125629 W CN 2018125629W WO 2020124682 A1 WO2020124682 A1 WO 2020124682A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
artifact
pure
sample data
natural
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/CN2018/125629
Other languages
English (en)
French (fr)
Inventor
葛永帅
陈剑威
朱炯滔
梁栋
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 WO2020124682A1 publication Critical patent/WO2020124682A1/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
    • G06T5/00—Image enhancement or restoration
    • G06T5/73—Deblurring; Sharpening
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/77—Retouching; Inpainting; Scratch removal
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • 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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging

Definitions

  • the present disclosure relates to the field of computer technology and the field of image processing technology, for example, to an image processing method, device, device, and storage medium.
  • X-ray grating phase contrast imaging technology is a grating imaging method based on Talbot effect and Lau effect. Through this technology, the object to be inspected can be effectively exposed to X-rays to obtain the absorption signal, the scattering signal and the refraction signal inside the object. Research shows that the refraction signal can effectively improve the contrast of soft tissue detection, while the scattered signal can greatly improve the detection sensitivity of particles or void structures inside the object. Therefore, X-ray grating phase-contrast imaging technology has received extensive attention from researchers and hopes to apply this technology to clinical testing.
  • this method can reduce the residual Moiré artifacts to a certain extent.
  • some studies have tried to reduce such artifacts based on wavelet transform technology.
  • the basic motivation of using wavelet transform technology is to solve the difficulty of different spatial frequency distribution of Moiré artifacts.
  • some people have proposed to use a certain mathematical method to estimate the small offset of each step, and optimize and correct it.
  • the main challenge lies in that it can only be applied to the case where the distribution of moiré artifacts is more regular and the orientation is more consistent.
  • these two methods have obvious limitations. Secondly, both of these methods may cause the loss of image resolution to varying degrees, causing difficulties in its practical application.
  • the optimization correction method based on the maximum likelihood method the shortcomings are: in principle, the X-ray grating phase contrast theory is based on the precise stepping, and the maximum likelihood method can only find the value of each step After the step correction, the offset can only alleviate some artifacts and cannot be completely removed. In addition, when the number of steps is small, the effect is not obvious.
  • the present disclosure provides an image processing method, device, device, and storage medium, which can effectively remove image artifacts.
  • the present disclosure provides an image processing method, including:
  • a pre-built artifact removal model is used to obtain an image after the artifact removal of the artifact image.
  • the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • the present disclosure also provides an image processing device including:
  • the image acquisition module is set to acquire the artifact image to be processed
  • the artifact removal module is set to use the pre-built artifact removal model to obtain the image after the artifact removal of the artifact image, and the sample data of the artifact removal model is to mathematically fuse the pure artifact image and the natural image get.
  • the present disclosure also provides a device including:
  • At least one processor At least one processor
  • a storage device configured to store at least one program
  • the at least one processor When the at least one program is executed by the at least one processor, the at least one processor implements the image processing method as described above.
  • the present disclosure also provides a computer-readable storage medium on which a computer program is stored, which when executed by a processor implements the image processing method described above.
  • FIG. 1 is a flowchart of the image processing method in Embodiment 1;
  • Embodiment 2 is a schematic diagram of the image processing method in Embodiment 1;
  • FIG. 3 is a schematic diagram of a pure artifact image in the first embodiment
  • Embodiment 5 is a schematic diagram of an artifact removal model in Embodiment 2.
  • FIG. 6 is a schematic diagram of an image processing process in Embodiment 2.
  • FIG. 7 is a schematic structural diagram of an image processing device in Embodiment 3.
  • Embodiment 8 is a schematic structural diagram of a device in Embodiment 4.
  • FIG. 1 is a flowchart of the image processing method in the first embodiment. This embodiment is applicable to the case of implementing image processing.
  • the method may be executed by an image processing device, which may use at least one of software and hardware.
  • the device may be configured in the device.
  • the artifact removal model in this embodiment is constructed based on the deep learning model.
  • FIG. 2 is a schematic diagram of the image processing method in Embodiment 1.
  • the sample data obtained by mathematical fusion is input into the artifact removal model for training to obtain a trained artifact removal model;
  • the real artifact image obtained in an experiment is input into the trained artifact removal model
  • the dotted line represents artifacts, and the result is an image with artifacts removed.
  • the method includes: S110-S120.
  • Artifacts refers to images of various forms that originally appear on the image when the scanned object does not exist, and Artifacts are images with different degrees of artifacts.
  • the artefacts in this embodiment are described using Moiré artifacts in X-ray grating phase contrast imaging as an example.
  • Moiré artifacts in X-ray grating phase contrast imaging systems, due to inaccurate phase stepping and unstable optical output intensity The result is an image with moire artifacts of varying degrees.
  • an image including moiré artifacts collected on a laboratory platform in real time may be obtained, or an image including moiré artifacts already on the Internet may be obtained.
  • a pre-built artifact removal model is used to obtain an image after the artifact removal of the artifact image, wherein the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image, including:
  • the pure artifact image and the natural image are fused using a preset fusion formula to obtain sample data.
  • the pure artifact image may be an image collected on a laboratory platform without objects, and the image includes only Moiré artifacts.
  • the phase stepped projection image is collected when there is no object on the laboratory platform, and a pure artifact image of the absorption image, the scattering image, and the refraction image can be obtained through the traditional signal extraction process.
  • FIG. 3 is a schematic diagram of a pure artifact image in Embodiment 1.
  • the figure is a pure artifact image of an absorption diagram, and the dotted line in the figure represents artifacts.
  • the natural image is an image without artifacts, and the source of the natural image may be, for example, an image obtained from the Internet without artifacts.
  • Normalize the pure artifact image and the natural image which may include: grayscale the collected natural image through Matlab (Matrix&Laboratory), and normalize it to between 0 and 1; by preset
  • the normalization formula of the image is normalized to the collected pure artifact image.
  • the preset normalization formula is not limited in this embodiment, and it may be normalized.
  • the pixel average value of the artifact stripes in the image is 0, and the amplitude is 0.5.
  • Image fusion is performed on the pure artifact image and the natural image after the normalization process using a preset fusion formula to obtain sample data, which may include: inputting the pixel values of the pure artifact image and the natural image after the normalization process into the pre-processing Set the calculation in the fusion formula to obtain the fusion image, which is the sample data.
  • the preset fusion formula may be a linear formula or a non-linear formula. In this embodiment, a linear formula is used as an example for description.
  • the preset fusion formula can be set according to the type of pure artifact image.
  • the fusion coefficient is affected by the current laboratory platform and environmental factors.
  • the relative intensity of the Moiré artifacts of the obtained fusion image is about 20%, which includes The images of Moiré artifacts are similar, so the fused image can be used as a sample image.
  • the artifact removal model is trained based on the sample data obtained by mathematical fusion to obtain a trained artifact removal model; inputting the artifact image to be processed into the trained artifact removal model can The artifact image is obtained after the artifact is removed.
  • the artifact removal model is constructed based on a convolutional neural network.
  • an artifact image to be processed is obtained, and an artifact removal model constructed in advance is used to obtain an image after the artifact image is removed.
  • the sample data of the artifact removal model is obtained by combining the pure artifact image and the natural image. Obtained through mathematical fusion.
  • a large amount of sample data is obtained through mathematical fusion, and an artifact removal model is trained based on the obtained sample data, so that the artifact removal model can more effectively achieve the removal of image artifacts, and compared with the related art.
  • Methods such as Fourier transform or wavelet transform improve the accuracy and processing efficiency of removing artifacts.
  • Embodiment 4 is a flowchart of an image processing method in Embodiment 2. This embodiment changes the above image processing method based on the above embodiment. Correspondingly, the method of this embodiment includes: S210-S250.
  • the phase stepped projection image is collected when there is no object on the laboratory platform, and a pure artifact image of the absorption image, the scattering image, and the refraction image can be obtained through the traditional signal extraction process.
  • the collected natural images are gray-scaled by Matlab (Matrix&Laboratory) and normalized to between 0 and 1; the pure artifact images collected are normalized by the preset normalization formula deal with.
  • the pre-normalized image and the natural image are fused using a preset fusion formula to obtain sample data.
  • the pixel values of the pure artifact image and the natural image after the normalization process are input into a preset fusion formula for calculation, and a fusion image can be obtained, and the fusion image is sample data.
  • the preset fusion formula may be a linear formula or a nonlinear formula. In this embodiment, a linear formula is used as an example for description.
  • the sample data is used as the input of the convolutional neural network, and the natural image corresponding to the sample data is used as the output of the convolutional neural network to train the convolutional neural network to obtain an artifact removal model.
  • the artifact removal model is constructed based on the convolutional neural network, which may include a down-sampling module and an up-sampling module, and the down-sampling module and the up-sampling module are connected through a residual connection layer.
  • FIG. 5 is a schematic diagram of an artifact removal model in Embodiment 2.
  • the downsampling module in the figure includes D convolutional layers, the size of the input image of convolutional layer 1 is recorded as M ⁇ N, and the mode of stride (stride) is 2 for downsampling convolution.
  • Convolutional layer 1 includes For multiple cascaded convolution units, the size of the output feature image is (M/2) ⁇ (N/2), and the size of the input image of the convolution layer 2 is (M/2) ⁇ (N/2), By analogy, the size of the feature image output by the convolution layer D is (M/2 ⁇ D) ⁇ (N/2 ⁇ D).
  • the downsampling module in the figure also includes D convolutional layers. The upsampling process is opposite to the downsampling process.
  • the input image size of the convolutional layer D+1 is (M/2 ⁇ D) ⁇ (N/2 ⁇ D ), the convolutional layer D+1 is a deconvolution layer, and the upsampling convolution is performed in a mode with a stride of 2, and the size of the output feature image is (M/2 ⁇ (D-1)) ⁇ ( N/2 ⁇ (D-1)), the following deconvolution layer and so on, until the last layer, the convolution layer D+D, no longer upsampling, the output image size and the first input image of the downsampling module
  • the size is the same, the output size is M ⁇ N.
  • the size of the convolution kernel used in multiple convolution layers in FIG. 5 can be 3 ⁇ 3, 5 ⁇ 5, or 7 ⁇ 7, etc., and the number of input and output feature images of multiple convolution layers can be 8, 16, 32, or 64.
  • the activation function of multiple convolutional layers can be a linear rectification function (Rectified Linear Unit, ReLU), a linear rectification function with leakage (Leaky ReLU), a hyperbolic tangent (Tanh) function, or a Sigmoid function.
  • the outputs of the multiple convolutional layers of the downsampling module are connected to the outputs of the multiple convolutional layers of the upsampling module, respectively.
  • the down-sampling module and the up-sampling module are connected through the residual connection layer, which can reduce the problem of gradient disappearance in the transmission of the convolutional neural network, and ensure the resolution of the image.
  • the sample data is used as the input of the convolutional neural network as shown in FIG. 5, and the natural image corresponding to the sample data is used as the supervision of the output image to train the convolutional neural network to obtain a trained artifact removal model. .
  • the artifacts in this embodiment are described by taking Moiré artifacts in X-ray grating phase contrast imaging as an example.
  • an image including moiré artifacts collected on a laboratory platform in real time may be obtained, or an image including moiré artifacts already on the Internet may be obtained.
  • a pre-built artifact removal model is used to obtain an image after the artifact image is removed.
  • the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • FIG. 6 is a schematic diagram of the image processing process in the second embodiment.
  • the downsampling module and the upsampling module of the artifact removal model in the figure are composed of 5 convolutional layers, and there are 5 residual connection layers that separate the downsampling module.
  • the output of is connected to the output of the upsampling module.
  • the convolution kernel used in the convolution layer in FIG. 6 is a 7 ⁇ 7 convolution kernel, and the activation function of the convolution layer is the ReLU function.
  • the size of the input artifact image to be processed is 224 ⁇ 224, the dotted line in the figure represents the artifact, and the output is the image with the artifact removed.
  • an artifact image to be processed is obtained, and an artifact removal model constructed in advance is used to obtain an image after the artifact image is removed.
  • the sample data of the artifact removal model is obtained by combining the pure artifact image and the natural image. Obtained through mathematical fusion.
  • a large amount of sample data is obtained through mathematical fusion, and an artifact removal model is trained based on the obtained sample data, so that the artifact removal model can more effectively achieve the removal of image artifacts, compared with Fourier in the related art.
  • Methods such as leaf transform or wavelet transform can be applied to many different kinds of artifacts, which improves the accuracy and processing efficiency of removing artifacts; and the artifact removal model in this embodiment uses a combination of downsampling and upsampling.
  • Convolutional neural network can remove artifacts without losing image resolution, and the structure of the neural network can be flexibly configured and expanded.
  • FIG. 7 is a schematic structural diagram of an image processing device in Embodiment 3, and this embodiment can be applied to the case of implementing image processing.
  • the image processing apparatus provided in this embodiment can execute the image processing method provided in any of the embodiments, and has function modules and beneficial effects corresponding to the execution method.
  • the device includes an image acquisition module 310 and an artifact removal module 320, where:
  • the image acquisition module 310 is configured to acquire an artifact image to be processed
  • the artifact removal module 320 is set to use the pre-built artifact removal model to obtain an image after the artifact image is removed.
  • the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • an artifact image to be processed is obtained, and an artifact removal model constructed in advance is used to obtain an image after the artifact image is removed.
  • the sample data of the artifact removal model is obtained by combining the pure artifact image and the natural image. Obtained through mathematical fusion.
  • a large amount of sample data is obtained through mathematical fusion, and an artifact removal model is trained based on the obtained sample data, so that the artifact removal model can more effectively achieve the removal of image artifacts, and compared with the related art.
  • Methods such as Fourier transform or wavelet transform improve the accuracy and processing efficiency of removing artifacts.
  • the artifact removal module 320 includes:
  • the processing unit is set to obtain pure artifact images and natural images, and normalize the pure artifact images and natural images;
  • the sample unit is set to perform image fusion on the pure artifact image and the natural image after the normalization process using a preset fusion formula to obtain sample data.
  • the device further includes a model module, and the model module is set to:
  • the sample data is used as the input of the convolutional neural network, and the natural image corresponding to the sample data is used as the output of the convolutional neural network to train the convolutional neural network to obtain an artifact removal model.
  • the artifact removal model includes a down-sampling module and an up-sampling module, and the down-sampling module and the up-sampling module are connected through a residual connection layer.
  • the image processing apparatus provided in this embodiment can execute the image processing method provided in any of the embodiments, and has function modules and beneficial effects corresponding to the execution method.
  • FIG. 8 is a schematic structural diagram of a device in Embodiment 4.
  • FIG. 8 shows a block diagram of an exemplary device 412 suitable for implementing the present embodiment.
  • the device 412 shown in FIG. 8 is only an example, and should not bring any limitation to the function and use range of this embodiment.
  • the device 412 is represented in the form of a general-purpose device.
  • the components of the device 412 may include: at least one processor 416, a storage device 428, and a bus 418 connecting different system components (including the storage device 428 and the processor 416).
  • Bus 418 represents one or more of several types of bus structures, including a storage device bus or storage device controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of a variety of bus structures.
  • these architectures include the industry standard architecture (Industry Subversive Alliance, ISA) bus, the micro channel architecture (Micro Channel Architecture, MAC) bus, the enhanced ISA bus, and the video electronics standards association (Video Electronics Electronics Standards Associations, VESA ) Local bus and Peripheral Component Interconnect (PCI) bus.
  • ISA Industry Subversive Alliance
  • MAC Micro Channel Architecture
  • VESA Video Electronics Electronics Standards Associations
  • PCI Peripheral Component Interconnect
  • the device 412 may be a readable medium including various computer systems. These media may be any available media that can be accessed by device 412, including volatile and non-volatile media, removable and non-removable media.
  • the storage device 428 may include a computer system readable medium in the form of volatile memory, such as at least one of a random access memory (Random Access Memory, RAM) 430 and a cache memory 432.
  • the device 412 may include other removable/non-removable, volatile/nonvolatile computer system storage media.
  • the storage system 434 can be used to read and write non-removable, non-volatile magnetic media (not shown in FIG. 8 and is commonly referred to as a "hard disk drive").
  • each drive may be connected to the bus 418 through at least one data medium interface.
  • the storage device 428 may include at least one program product having a set (eg, at least one) of program modules that are configured to perform the functions of various embodiments.
  • a program/utility tool 440 having a set of (at least one) program modules 442 may be stored in, for example, a storage device 428.
  • Such program modules 442 include an operating system, at least one application program, other program modules, and program data. In these examples Each or some combination of may include the realization of the network environment.
  • the program module 442 generally performs at least one of the functions and methods in the embodiments described in any embodiment.
  • the device 412 can also communicate with at least one of the following: at least one external device 414 (such as a keyboard, graphics processor (GPU), pointing terminal, and display 424, etc.), at least one that enables a user to interact with the device 412 Any terminal that enables the device 412 to communicate with at least one other computing terminal (such as a network card and modem, etc.). Such communication may be performed through an input/output (I/O) interface 422.
  • the device 412 may also communicate with at least one network (such as a local area network (Local Area Network, LAN), a wide area network (Wide Area Network, WAN), and a public network, such as the Internet) through the network adapter 420. As shown in FIG.
  • the network adapter 420 communicates with other modules of the device 412 through the bus 418. It should be understood that although not shown in the figure, at least one of other hardware and software modules may be used in conjunction with the device 412, including: microcode, terminal driver, redundant processor, external disk drive array, disk array (Redundant Arrays of Independent Disks) , RAID) systems, tape drives, and data backup storage systems.
  • the processor 416 runs a program stored in the storage device 428 to execute various functional applications and data processing, for example, to implement the image processing method provided in this embodiment, the method includes:
  • the pre-built artifact removal model is used to obtain an image after artifact removal.
  • the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • the fifth embodiment also provides a computer-readable storage medium on which a computer program is stored.
  • the image processing method as provided in this embodiment is implemented.
  • the method includes:
  • the pre-built artifact removal model is used to obtain an image after artifact removal.
  • the sample data of the artifact removal model is obtained by mathematically fusing the pure artifact image and the natural image.
  • the computer storage medium of this embodiment may use any combination of at least one computer-readable medium.
  • the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium.
  • the computer-readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above.
  • Computer-readable storage media include: electrical connections with at least one wire, portable computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only Memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.
  • the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.
  • the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which computer-readable program code is carried. This propagated data signal can take many forms, including electromagnetic signals, optical signals, or any suitable combination of the above.
  • the computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable medium may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. .
  • the program code contained on the computer-readable medium may be transmitted using any appropriate medium, including wireless, electric wire, optical cable, RF, etc., or any suitable combination of the foregoing.
  • the computer program code for performing the operations of the above embodiments may be written in one or more programming languages or a combination thereof, and the programming languages include object-oriented programming languages such as Java, Smalltalk, C++, Python, and A variety of open source deep learning toolkits developed by the Python framework also include conventional procedural programming languages-such as "C" language or similar programming languages.
  • the program code may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or terminal.
  • the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (for example, through an Internet service provider Internet connection).
  • LAN local area network
  • WAN wide area network
  • Internet service provider Internet connection for example, AT&T, MCI, Sprint, EarthLink, MSN, GTE, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Processing (AREA)

Abstract

一种图像处理方法、装置、设备和存储介质，其中该方法包括：获取待处理的伪影图像；利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，其中，伪影去除模型的样本数据是通过将纯伪影图像和自然图像进行数学融合得到。

Description

图像处理方法、装置、设备和存储介质
本申请要求在2018年12月17日提交中国专利局、申请号为201811541513.8的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本公开涉及计算机技术领域以及图像处理技术领域，例如涉及一种图像处理方法、装置、设备和存储介质。
背景技术
X射线光栅相衬成像技术是一种基于泰伯(Talbot)效应和Lau效应的光栅成像方法。通过该技术，能够有效对待检物体进行X射线曝光，从而得到物体内部的吸收信号、散射信号和折射信号。研究表明，折射信号可以有效的提高软组织探测的对比度，而散射信号可以极大的提高对物体内部颗粒或者空隙结构的探测灵敏度。因此，X射线光栅相衬成像技术得到了研究人员的广泛关注，并希望将该技术应用到临床检测当中。
上述吸收、散射以及折射信号的提取是基于多个相位步进投影图得到的。但是，由于受光栅步进位移精确度和光机输出稳定性的制约，传统信号提取方法会导致得到的吸收图、散射图和折射图会有残存的莫尔伪影。这些残存的莫尔伪影会极大的降低图像的可读性，也就是间接降低X射线光栅相衬成像技术的辐射剂量利用效率。为了减少残存的莫尔伪影，同时提高辐射剂量利用效率，有研究曾尝试使用基于傅里叶变换技术的去伪影方法。通过压低频率域空间与莫尔伪影相对应的频率信息，该方法可以在一定程度上减少残存的莫尔伪影。 同时，也有研究曾尝试基于小波变换技术来减少该类伪影。使用小波变换技术的基本动机是为了解决莫尔伪影不同空间频率分布的困难。另外，也有人提出用一定的数学方法估计出每个步进的微小偏移量，并进行优化和修正。
对于基于傅里叶变换或者小波变换的方法来说，主要的挑战在于只能应用于莫尔条纹伪影分布比较有规律，取向比较一致的情况。对于具有任意形状的莫尔伪影分布来说，这两种方法都有明显的局限性。其次，这两种方法都可能造成图像分辨率在不同程度上有损失，给其实际应用带来困难。对基于最大似然法的优化修正法而言，存在的不足是：原则上X射线光栅相衬理论是建立在精准步进的情况上的，最大似然法只能求出每个步进的偏移量，对于步进修正之后，只能缓解一部分伪影并不能完全去除。此外，对于步进数较少的时候，效果并不明显。但是，该方法的另外一个缺点是没有考虑光机输出不稳定带来的影响。我们知道，实验过程中对于长时间曝光，光机的光斑会漂移，光斑的漂移量往往大于机械移动光栅的偏移量，光斑漂移间接导致步进不精准。虽然对光强不一致可以进行后期校正，但是对于有些实验，特别是很难对数据进行校正的情况时，这种光强校正的图像后处理算法就变得比较困难了。综上，由于残余莫尔伪影的分布形态各异，给以上方法的应用带来了很多困难，导致这些方法均有一定的局限性。
发明内容
本公开提供了一种图像处理方法、装置、设备和存储介质，可以有效去除图像的伪影。
本公开提供了一种图像处理方法，包括：
获取待处理的伪影图像；以及
利用预先构建的伪影去除模型，得到所述伪影图像去除伪影之后的图像，所述伪影去除模型的样本数据是通过将纯伪影图像和自然图像进行数学融合得到。
本公开还提供了一种图像处理装置，该装置包括：
图像获取模块，设置为获取待处理的伪影图像；
伪影去除模块，设置为利用预先构建的伪影去除模型，得到所述伪影图像去除伪影之后的图像，所述伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。
本公开还提供了一种设备，所述设备包括：
至少一个处理器；
存储装置，设置为存储至少一个程序；
当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如上所述的图像处理方法。
本公开还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如上所述的图像处理方法。
附图说明
图1为实施例一中的图像处理方法的流程图；
图2为实施例一中的图像处理方法的示意图；
图3为实施例一中的纯伪影图像示意图；
图4为实施例二中的图像处理方法的流程图；
图5为实施例二中的伪影去除模型的示意图；
图6为实施例二中的图像处理过程的示意图；
图7为实施例三中的图像处理装置的结构示意图；
图8为实施例四中的设备的结构示意图。
具体实施方式
实施例一
图1为实施例一中的图像处理方法的流程图，本实施例可适用于实现图像处理的情况，该方法可以由图像处理装置执行，该装置可以采用软件的方式和硬件的方式中至少之一实现，例如，该装置可配置于设备中。
本实施例中的伪影去除模型是基于深度学习模型构建的，对伪影图像的处理参见图2，图2为实施例一中的图像处理方法的示意图。图中将通过数学融合得到的样本数据输入伪影去除模型中进行训练，得到训练好的伪影去除模型；图中将一个实验中得到的真实的伪影图像输入该训练好的伪影去除模型中，虚线代表伪影，得到的结果为去除伪影的图像。
如图1所示，该方法包括：S110-S120。
S110中，获取待处理的伪影图像。
其中，伪影(Artifacts)是指原本被扫描物体并不存在而在图像上却出现的多种形态的影像，伪影图像为带有不同程度的伪影的图像。本实施例中的伪影以X射线光栅相衬成像中的莫尔伪影为例进行说明，对于X射线光栅相衬成像系统而言，由于相位步进不精准和光机输出光强不稳定会导致最终获得的图像带有不同程度的莫尔伪影。
本实施例中可以获取实验室平台上实时采集的包括莫尔伪影的图像，也可以获取互联网中已有的包括莫尔伪影的图像。
S120中，利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图 像，其中，伪影去除模型的样本数据是通过将纯伪影图像和自然图像进行数学融合得到。
其中，伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到，包括：
获取纯伪影图像和自然图像，并对纯伪影图像和自然图像进行归一化处理；以及
对归一化处理之后的纯伪影图像和自然图像采用预设融合公式进行图像融合，得到样本数据。
纯伪影图像可以为在实验室平台上采集的不带物体的图像，该图像仅包括莫尔伪影。本实施例中，在实验室平台上无物体的情况下采集到相位步进投影图，通过传统信号提取过程可以得到吸收图、散射图和折射图的纯伪影图像。示例性的，参见图3，图3为实施例一中的纯伪影图像示意图，图中为吸收图的纯伪影图像，图中的虚线代表伪影。自然图像为不带伪影的图像，自然图像的来源例如可以为从互联网上获取的无伪影的图像。
对纯伪影图像和自然图像进行归一化处理，可以包括：将采集到的自然图像通过Matlab(Matrix&Laboratory)进行灰度化处理，并将其归一化到0到1之间；通过预设的归一化公式对采集到的纯伪影图像进行归一化处理。该预设的归一化公式本实施例中不作限定，可以实现归一化即可。例如预设的归一化公式可以为Mn＝[Yn-mean(Yn)]/[max(Yn)-min(Yn)]，其中Mn代表纯伪影图像归一化之后的像素值，Yn代表纯伪影图像的初始像素值，mean(Yn)代表对Yn取平均值，max(Yn)代表对Yn取最大值，min(Yn)代表对Yn取最小值，归一化之后的纯伪影图像中的伪影条纹的像素均值为0，振幅为0.5。
对归一化处理之后的纯伪影图像和自然图像采用预设融合公式进行图像融 合，得到样本数据，可以包括：将归一化处理之后的纯伪影图像和自然图像的像素值均输入预设融合公式中进行计算，可以得到融合图像，该融合图像为样本数据。其中，预设融合公式可以为线性公式，也可以为非线性公式，本实施例中以线性公式为例进行说明。预设融合公式可以根据纯伪影图像的类型进行设置，示例性的，对于吸收图和散射图的纯伪影图像，预设融合公式均可以为An＝(3-a)×In+a×Mn，其中An代表融合图像的像素值，Mn代表纯伪影图像归一化之后的像素值，In代表归一化处理之后的自然图像的像素值，a代表耦合系数；对于折射图的纯伪影图像，预设融合公式可以为Sn＝(6.28-a)×Pn+a×Mn，其中Sn代表融合图像的像素值，Mn代表纯伪影图像归一化之后的像素值，Pn代表归一化处理之后的自然图像的相邻像素点的像素差值，a代表耦合系数。融合系数受当前的实验室平台以及环境因素影响的，本实施例中耦合系数的取值为0到0.3时，得到的融合图像的莫尔伪影的相对强度为20％左右，与真实的包括莫尔伪影的图像类似，因此，该融合图像可以作为样本图像。
示例性的，随机从多种类型的256张纯伪影图像和十万张自然图像中分别挑选两张，再通过归一化处理以及预设融合公式生成十万张的融合图像，即得到十万个样本数据。
由于实验室平台以及客观条件的局限，实际中很难采集到数量巨大的不包括伪影的图像，因此本实施例中通过数学融合得到符合要求的大量的样本数据，可以更简单便利地实现样本数据的获取。
在一实施例中，基于数学融合获取到的样本数据对伪影去除模型进行训练，得到训练好的伪影去除模型；将待处理的伪影图像输入该训练好的伪影去除模型中，可以得到该伪影图像去除伪影之后的图像。在一实施例中，所述伪影去除模型是基于卷积神经网络构建的。
本实施例通过获取待处理的伪影图像，并利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，其中伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。本实施例通过数学融合得到大量的样本数据，并基于得到的样本数据去训练伪影去除模型，使得该伪影去除模型可以更加有效地实现图像伪影的去除，并且相对于相关技术中的傅里叶变换或者小波变换等方法，提高了去除伪影的精确度和处理效率。
实施例二
图4为实施例二中的图像处理方法的流程图。本实施例在上述实施例的基础上，对上述图像处理方法进行了改变。相应的，本实施例的方法包括：S210-S250。
S210中，获取纯伪影图像和自然图像，并对纯伪影图像和自然图像进行归一化处理。
本实施例中，在实验室平台上无物体的情况下采集到相位步进投影图，通过传统信号提取过程可以得到吸收图、散射图和折射图的纯伪影图像。将采集到的自然图像通过Matlab(Matrix&Laboratory)进行灰度化处理，并将其归一化到0到1之间；通过预设的归一化公式对采集到的纯伪影图像进行归一化处理。
S220中，对归一化处理之后的纯伪影图像和自然图像采用预设融合公式进行图像融合，得到样本数据。
将归一化处理之后的纯伪影图像和自然图像的像素值均输入预设融合公式中进行计算，可以得到融合图像，该融合图像为样本数据。其中，预设融合公式可以为线性公式，也可以为非线性公式，本实施例中以线性公式为例进行说 明。
示例性的，随机从多种类型的256张纯伪影图像和十万张自然图像中分别挑选两张，再通过归一化处理以及预设融合公式生成十万张的融合图像，即得到十万个样本数据。
S230中，将样本数据作为卷积神经网络的输入，样本数据对应的自然图像作为卷积神经网络的输出对卷积神经网络进行训练，得到伪影去除模型。
其中，伪影去除模型是基于卷积神经网络构建的，可以包括降采样模块和升采样模块，降采样模块和升采样模块通过残差连接层连接。参见图5，图5为实施例二中的伪影去除模型的示意图。图中的降采样模块包括D个卷积层，卷积层1的输入图像的尺寸记为M×N，采用步长(stride)为2的模式进行降采样卷积，卷积层1内部包括多个级联卷积单元，输出的特征图像的尺寸为(M/2)×(N/2)，卷积层2的输入图像的尺寸即为(M/2)×(N/2)，以此类推，卷积层D输出的特征图像的尺寸为(M/2^D)×(N/2^D)。图中的降采样模块中也包括D个卷积层，升采样过程与降采样过程相反，卷积层D+1的输入图像的尺寸为(M/2^D)×(N/2^D)，卷积层D+1为反卷积层，采用步长(stride)为2的模式进行升采样卷积，输出的特征图像的尺寸为(M/2^(D-1))×(N/2^(D-1))，后面的反卷积层以此类推，直到最后一层即卷积层D+D不再升采样，输出的图像尺寸与降采样模块最开始输入的图像尺寸一致，输出的尺寸为M×N。
图5中多个卷积层采用的卷积核的大小可以为3×3、5×5或者7×7等，多个卷积层的输入输出特征图像数目可以为8、16、32或者64等，多个卷积层的激活函数可以为线性整流函数(Rectified Linear Unit，ReLU)、带泄露线性整流函数(Leaky ReLU)、双曲正切(Tanh)函数或者Sigmoid函数等。图5中的降采样模块和升采样模块的每一卷积层通过残差连接层连接，分别将降采 样模块多个卷积层的输出连接到升采样模块多个卷积层的输出上。本实施例中通过残差连接层将降采样模块和升采样模块连接，可以减少卷积神经网络传输中的梯度消失的问题，并且保证了图像的分辨率。
在一实施例中，将样本数据作为如图5所示的卷积神经网络的输入，样本数据对应的自然图像作为输出图像的监督对卷积神经网络进行训练，得到训练好的伪影去除模型。
S240中，获取待处理的伪影图像。
本实施例中的伪影以X射线光栅相衬成像中的莫尔伪影为例进行说明。本实施例中可以获取实验室平台上实时采集的包括莫尔伪影的图像，也可以获取互联网中已有的包括莫尔伪影的图像。
S250中，利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像。
其中，伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。
图6为实施例二中的图像处理过程的示意图，图中的伪影去除模型的降采样模块和升采样模块均由5个卷积层组成，有5个残差连接层分别把降采样模块的输出连接到升采样模块的输出上，图6中的卷积层采用的卷积核为7×7的卷积核，卷积层的激活函数为ReLU函数。输入的待处理的伪影图像的尺寸大小为224×224，图中的虚线代表伪影，输出的是去除伪影的图像。
本实施例通过获取待处理的伪影图像，并利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，其中伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。本实施例通过数学融合得到大量的样本数据，并基于得到的样本数据去训练伪影去除模型，使得该伪影去除模型可以 更加有效地实现图像伪影的去除，相对于相关技术中的傅里叶变换或者小波变换等方法，可以适用于多种不同种类的伪影，提高了去除伪影的精确度和处理效率；并且本实施例中的伪影去除模型采用降采样与升采样结合的深度卷积神经网络，可以保证图像分辨率不丢失的情况下去除伪影，且神经网络的结构可以灵活地配置扩展。
实施例三
图7为实施例三中的图像处理装置的结构示意图，本实施例可适用于实现图像处理的情况。本实施例所提供的图像处理装置可执行任意实施例所提供的图像处理方法，具备执行方法相应的功能模块和有益效果。该装置包括图像获取模块310和伪影去除模块320，其中：
图像获取模块310，设置为获取待处理的伪影图像；
伪影去除模块320，设置为利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。
本实施例通过获取待处理的伪影图像，并利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，其中伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。本实施例通过数学融合得到大量的样本数据，并基于得到的样本数据去训练伪影去除模型，使得该伪影去除模型可以更加有效地实现图像伪影的去除，并且相对于相关技术中的傅里叶变换或者小波变换等方法，提高了去除伪影的精确度和处理效率。
在一实施例中，伪影去除模块320包括：
处理单元，设置为获取纯伪影图像和自然图像，并对纯伪影图像和自然图 像进行归一化处理；
样本单元，设置为对归一化处理之后的纯伪影图像和自然图像采用预设融合公式进行图像融合，得到样本数据。
在一实施例中，该装置还包括模型模块，模型模块是设置为：
将样本数据作为卷积神经网络的输入，样本数据对应的自然图像作为卷积神经网络的输出对卷积神经网络进行训练，得到伪影去除模型。
在一实施例中，伪影去除模型包括降采样模块和升采样模块，降采样模块和升采样模块通过残差连接层连接。
本实施例所提供的图像处理装置可执行任意实施例所提供的图像处理方法，具备执行方法相应的功能模块和有益效果。
实施例四
图8为实施例四中的设备的结构示意图。图8示出了适于用来实现本实施方式的示例性设备412的框图。图8显示的设备412仅仅是一个示例，不应对本实施例的功能和使用范围带来任何限制。
如图8所示，设备412以通用设备的形式表现。设备412的组件可以包括：至少一个处理器416，存储装置428，连接不同系统组件(包括存储装置428和处理器416)的总线418。
总线418表示几类总线结构中的一种或多种，包括存储装置总线或者存储装置控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构包括工业标准体系结构(Industry Subversive Alliance，ISA)总线，微通道体系结构(Micro Channel Architecture，MAC)总线，增强型ISA总线、视频电子标准协会(Video  Electronics Standards Association，VESA)局域总线以及外围组件互连(Peripheral Component Interconnect，PCI)总线。
设备412可以是包括多种计算机系统可读介质。这些介质可以是任何能够被设备412访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。
存储装置428可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(Random Access Memory，RAM)430和高速缓存存储器432中至少一个。设备412可以包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储系统434可以用于读写不可移动的、非易失性磁介质(图8未显示，通常称为“硬盘驱动器”)。尽管图8中未示出，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘，例如只读光盘(Compact Disc Read-Only Memory，CD-ROM)，数字视盘(Digital Video Disc-Read Only Memory，DVD-ROM)或者其它光介质)读写的光盘驱动器。在这些情况下，每个驱动器可以通过至少一个数据介质接口与总线418相连。存储装置428可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程序模块被配置以执行多个实施例的功能。
具有一组(至少一个)程序模块442的程序/实用工具440，可以存储在例如存储装置428中，这样的程序模块442包括操作系统、至少一个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块442通常执行任一实施例所描述的实施例中的功能和方法中的至少一个。
设备412也可以与下述至少一个进行通信：至少一个外部设备414(例如键 盘、图形处理器(Graphics Processing Uni，GPU)、指向终端以及显示器424等)，至少一个使得用户能与该设备412交互的终端，使得该设备412能与至少一个其它计算终端进行通信的任何终端(例如网卡和调制解调器等等)。这种通信可以通过输入/输出(I/O)接口422进行。并且，设备412还可以通过网络适配器420与至少一个网络(例如局域网(Local Area Network，LAN)、广域网(Wide Area Network，WAN)和公共网络中至少一个，例如因特网)通信。如图8所示，网络适配器420通过总线418与设备412的其它模块通信。应当明白，尽管图中未示出，可以结合设备412使用其它硬件和软件模块中至少一个，包括：微代码、终端驱动器、冗余处理器、外部磁盘驱动阵列、磁盘阵列(Redundant Arrays of Independent Disks，RAID)系统、磁带驱动器以及数据备份存储系统等。
处理器416通过运行存储在存储装置428中的程序，从而执行多种功能应用以及数据处理，例如实现本实施例所提供的图像处理方法，该方法包括：
获取待处理的伪影图像；
利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。
实施例五
本实施例五还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本实施例所提供的图像处理方法，该方法包括：
获取待处理的伪影图像；
利用预先构建的伪影去除模型，得到伪影图像去除伪影之后的图像，伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到。
本实施例的计算机存储介质，可以采用至少一个计算机可读的介质的任意组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质(非穷举的列表)包括：具有至少一个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。
计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。
计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括无线、电线、光缆以及RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言或其组合来编写用于执行上述实施例操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言-诸如Java、Smalltalk、C++、Python以及基于Python框架开发的多类开源深度学习工具包，还包括常规的过程式程序设计语言-诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执 行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或终端上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)-连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。

Claims (10)

  1. 一种图像处理方法，包括：
    获取待处理的伪影图像；以及
    利用预先构建的伪影去除模型，得到所述伪影图像去除伪影之后的图像，其中，所述伪影去除模型的样本数据是通过将纯伪影图像和自然图像进行数学融合得到。
  2. 根据权利要求1所述的方法，其中，所述伪影去除模型的样本数据通过将纯伪影图像和自然图像进行数学融合得到，包括：
    获取纯伪影图像和自然图像，并对所述纯伪影图像和所述自然图像进行归一化处理；以及
    对归一化处理之后的所述纯伪影图像和所述自然图像采用预设融合公式进行图像融合，得到所述样本数据。
  3. 根据权利要求1所述的方法，在获取待处理的伪影图像之前，还包括：
    将所述样本数据作为卷积神经网络的输入，所述样本数据对应的自然图像作为所述卷积神经网络的输出对所述卷积神经网络进行训练，得到所述伪影去除模型。
  4. 根据权利要求1所述的方法，其中，所述伪影去除模型包括降采样模块和升采样模块，所述降采样模块和所述升采样模块通过残差连接层连接。
  5. 一种图像处理装置，包括：
    图像获取模块，设置为获取待处理的伪影图像；
    伪影去除模块，设置为利用预先构建的伪影去除模型，得到所述伪影图像去除伪影之后的图像，其中，所述伪影去除模型的样本数据是通过将纯伪影图像和自然图像进行数学融合得到。
  6. 根据权利要求5所述的装置，其中，所述伪影去除模块包括：
    处理单元，设置为获取纯伪影图像和自然图像，并对所述纯伪影图像和所述自然图像进行归一化处理；
    样本单元，设置为对归一化处理之后的所述纯伪影图像和所述自然图像采用预设融合公式进行图像融合，得到样本数据。
  7. 根据权利要求5所述的装置，其中，还包括模型模块，所述模型模块是设置为：
    将样本数据作为卷积神经网络的输入，所述样本数据对应的自然图像作为所述卷积神经网络的输出对所述卷积神经网络进行训练，得到所述伪影去除模型。
  8. 根据权利要求5所述的装置，其中，所述伪影去除模型包括降采样模块和升采样模块，所述降采样模块和所述升采样模块通过残差连接层连接。
  9. 一种设备，包括：
    至少一个处理器；
    存储装置，设置为存储至少一个程序；
    当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如权利要求1-4中任一所述的图像处理方法。
  10. 一种计算机可读存储介质，存储有计算机程序，其中，该程序被处理器执行时实现如权利要求1-4中任一所述的图像处理方法。
PCT/CN2018/125629 2018-12-17 2018-12-29 图像处理方法、装置、设备和存储介质 Ceased WO2020124682A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811541513.8A CN111325675B (zh) 2018-12-17 2018-12-17 图像处理方法、装置、设备和存储介质
CN201811541513.8 2018-12-17

Publications (1)

Publication Number Publication Date
WO2020124682A1 true WO2020124682A1 (zh) 2020-06-25

Family

ID=71102448

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/125629 Ceased WO2020124682A1 (zh) 2018-12-17 2018-12-29 图像处理方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN111325675B (zh)
WO (1) WO2020124682A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815692A (zh) * 2020-07-15 2020-10-23 大连东软教育科技集团有限公司 无伪影数据及有伪影数据的生成方法、系统及存储介质
CN112419372A (zh) * 2020-11-11 2021-02-26 广东拓斯达科技股份有限公司 图像处理方法、装置、电子设备及存储介质
CN113962870A (zh) * 2020-07-20 2022-01-21 浙江宇视科技有限公司 一种图像锅盖效应抑制方法、装置、电子设备和存储介质
CN116432727A (zh) * 2021-12-29 2023-07-14 北京字节跳动网络技术有限公司 一种图像处理方法、装置、电子设备及存储介质

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114004903B (zh) * 2020-07-28 2026-03-03 上海联影医疗科技股份有限公司 伪影校正方法、装置、计算机设备和存储介质
CN113255756B (zh) * 2021-05-20 2024-05-24 联仁健康医疗大数据科技股份有限公司 图像融合方法、装置、电子设备及存储介质
CN114219777B (zh) * 2021-12-06 2025-08-15 上海交通大学 内窥镜图像鲁棒腔道检测方法、系统、终端及介质
CN114529695B (zh) * 2021-12-30 2025-03-25 北京城市网邻信息技术有限公司 全景图像处理与生成方法、装置、电子设备及存储介质
CN116309901B (zh) * 2022-12-29 2026-04-24 上海联影智能科技股份有限公司 伪影去除方法、装置、电子设备及存储介质
CN119791719A (zh) * 2023-10-10 2025-04-11 上海博动医疗科技股份有限公司 一种血管内图像生成方法、装置、设备及介质
CN117689980B (zh) * 2024-02-04 2024-05-24 青岛海尔科技有限公司 构建环境识别模型的方法、识别环境的方法及装置、设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107403446A (zh) * 2016-05-18 2017-11-28 西门子保健有限责任公司 用于使用智能人工代理的图像配准的方法和系统
CN107610193A (zh) * 2016-06-23 2018-01-19 西门子保健有限责任公司 使用深度生成式机器学习模型的图像校正
CN107958472A (zh) * 2017-10-30 2018-04-24 深圳先进技术研究院 基于稀疏投影数据的pet成像方法、装置、设备及存储介质
US20180177461A1 (en) * 2016-12-22 2018-06-28 The Johns Hopkins University Machine learning approach to beamforming

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017012642A1 (en) * 2015-07-17 2017-01-26 Hewlett-Packard Indigo B.V. Methods of detecting moire artifacts
CN107730479B (zh) * 2017-08-30 2021-04-20 中山大学 基于压缩感知的高动态范围图像去伪影融合方法
CN107945132B (zh) * 2017-11-29 2022-10-04 深圳安科高技术股份有限公司 一种基于神经网络的ct图像的伪影校正方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107403446A (zh) * 2016-05-18 2017-11-28 西门子保健有限责任公司 用于使用智能人工代理的图像配准的方法和系统
CN107610193A (zh) * 2016-06-23 2018-01-19 西门子保健有限责任公司 使用深度生成式机器学习模型的图像校正
US20180177461A1 (en) * 2016-12-22 2018-06-28 The Johns Hopkins University Machine learning approach to beamforming
CN107958472A (zh) * 2017-10-30 2018-04-24 深圳先进技术研究院 基于稀疏投影数据的pet成像方法、装置、设备及存储介质

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815692A (zh) * 2020-07-15 2020-10-23 大连东软教育科技集团有限公司 无伪影数据及有伪影数据的生成方法、系统及存储介质
CN111815692B (zh) * 2020-07-15 2023-12-01 东软教育科技集团有限公司 无伪影数据及有伪影数据的生成方法、系统及存储介质
CN113962870A (zh) * 2020-07-20 2022-01-21 浙江宇视科技有限公司 一种图像锅盖效应抑制方法、装置、电子设备和存储介质
CN112419372A (zh) * 2020-11-11 2021-02-26 广东拓斯达科技股份有限公司 图像处理方法、装置、电子设备及存储介质
CN112419372B (zh) * 2020-11-11 2024-05-17 广东拓斯达科技股份有限公司 图像处理方法、装置、电子设备及存储介质
CN116432727A (zh) * 2021-12-29 2023-07-14 北京字节跳动网络技术有限公司 一种图像处理方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN111325675B (zh) 2023-12-26
CN111325675A (zh) 2020-06-23

Similar Documents

Publication Publication Date Title
WO2020124682A1 (zh) 图像处理方法、装置、设备和存储介质
Whiteley et al. DirectPET: full-size neural network PET reconstruction from sinogram data
CN110956126B (zh) 一种联合超分辨率重建的小目标检测方法
Wang et al. Smartphone-based wound assessment system for patients with diabetes
CN113393514B (zh) 一种三维无序点云数据处理方法、系统及设备
Mulleti et al. Ellipse fitting using the finite rate of innovation sampling principle
CN110796108B (zh) 一种人脸质量检测的方法、装置、设备及存储介质
US11222228B2 (en) Techniques for example-based affine registration
CN110046116A (zh) 一种张量填充方法、装置、设备及存储介质
US12450690B2 (en) Method, device, and computer program product for image processing
CN115601299A (zh) 基于图像的肝硬化状态智能评估系统及其方法
CN110310347A (zh) 一种x射线光栅相衬成像滤波迭代重建方法和系统
CN114241204B (zh) 图像识别方法、装置、设备、介质及计算机产品
WO2024087359A1 (zh) 用于内窥镜的病灶检测方法、装置、电子设备及存储介质
CN110020597A (zh) 一种用于头晕/眩晕辅诊的眼部视频处理方法及系统
Li et al. Unsupervised data fidelity enhancement network for spectral CT reconstruction
CN120318072A (zh) 图像超分辨率重建方法、装置、设备、介质及程序产品
Wang et al. FSTU-Net: A sparse-view CT reconstruction framework for enhanced imaging efficiency and quality in neutron tomography
CN114972118B (zh) 检查图像的降噪方法、装置、可读介质和电子设备
CN109785406B (zh) 光栅成像系统的信息提取方法及提取装置
CN116486467A (zh) 眼睛检测框的确定方法、装置、设备及存储介质
CN113239978A (zh) 医学图像预处理模型与分析模型的相关方法和装置
CN112577976B (zh) 光栅相衬成像信号提取方法、装置、存储介质及电子设备
CN118485813B (zh) 基于深度学习的全景效果评估系统及方法
CN116721143B (zh) 3d医学图像的深度信息处理装置及方法

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: 18944128

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 10.11.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18944128

Country of ref document: EP

Kind code of ref document: A1