Functional imaging noise reduction and registration method and device, computer equipment and storage medium

Abstract

The invention relates to a functional imaging noise reduction and registration method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring an original video sequence, and preprocessing the original video sequence; extracting a set number of continuous video frames from the preprocessed video sequence, inputting the continuous video frames into a Vi deoSD network, and carrying out network training by taking an intermediate frame after outputting noise reduction and registering with front and rear frames as a training target to obtain a trained noise reduction registration network; and extracting a set number of continuous video frames from the video sequence to be processed, replacing an intermediate frame in the continuous video frames with a set template frame, inputting the intermediate frame into the trained noise reduction registration network, and outputting the intermediate frame which is subjected to noise reduction and registration with the fixed frame through the noise reduction registration network. According to the method, denoising and registration are used as two correlated tasks for training at the same time, so that the denoising and motion calibration effects can be improved, and meanwhile, the data processing efficiency is improved.

Description

Technical Field

[0001] This application belongs to the technical field of functional imaging data processing, and particularly relates to a method and device for functional imaging noise reduction and registration, a computer device, and a storage medium. Background Art

[0002] Neurons are the basic functional units in the brain, and the electrical activities and chemical signal transmissions between them are the basis of brain information processing. Functional imaging can record large-scale neuronal activities and help scientists explore the complex functions of the neuronal brain. However, such functional imaging techniques are easily interfered by noise and motion displacement, affecting the accuracy of data analysis results. Traditional methods perform motion calibration and data noise reduction on the data, and then perform operations such as signal extraction. In the field of functional imaging data processing, the currently commonly used motion calibration method is NormCorre, and NoRMCorre can operate in a rigid or piecewise rigid (pw-rigid) manner. For rigid registration, the displacement vector is calculated by computing the maximum value of the cross-correlation between each frame and the template. Under the piecewise rigid method, the FOV in any given frame is divided into a set of overlapping blocks, and then these small blocks are registered separately, and then a smooth deformation field is created by smooth interpolation and merging. In this way, a similar non-rigid registration effect can be obtained by taking denser blocks. This method requires a long calculation time and is severely interfered by noise. In the case of large noise, taking denser blocks will cause the results to be unstable.

[0003] Existing deep learning noise reduction methods include DeepCad, DeepCAD-RT, DeepInterpolation, and SUPPORT, etc. Among them, DeepCad, DeepCad-RT, and DeepInterpolation all assume that the displacements of consecutive frames in the video have been removed, so motion calibration needs to be performed first before use. Then, DeepCad and DeepCad-RT separate the odd frames and even frames of the video sequence, input the odd frames into a 3D-Unet, and use the even frames as the output targets of the network to train the network. DeepInterpolation inputs the first N frames and the last N frames of the current frame to be denoised into the VideoSD network, and uses the current frame as the output target of the network to train the network. SUPPORT uses hollow convolution to block the pixels to be denoised currently, only inputs the pixels around it into the denoising network, and then uses the pixels to be denoised currently as the output target of the network to train the network. Although SUPPORT does not require motion registration before use, if operations such as neuronal signal extraction are needed after denoising, motion calibration also needs to be performed. Performing motion calibration directly in the presence of noise will affect the calibration effect and the operation time will also be longer. Summary of the Invention

[0004] The present application provides a method, apparatus, computer device, and storage medium for functional imaging denoising and registration, aiming to solve at least one of the above technical problems in the prior art to a certain extent.

[0005] To solve the above problems, the present application provides the following technical solutions:

[0006] A method for functional imaging denoising and registration, comprising:

[0007] Obtain an original video sequence and preprocess the original video sequence;

[0008] Extract a set number of consecutive video frames from the preprocessed video sequence, input the consecutive video frames into the VideoSD network, and perform network training with the output of the denoised and registered intermediate frame with the previous and next frames as the training target to obtain a trained denoising and registration network;

[0009] Extract a set number of consecutive video frames from the video sequence to be processed, replace the intermediate frame in the consecutive video frames with a set template frame, and then input it into the trained denoising and registration network, and output the denoised and registered intermediate frame with the fixed frame through the denoising and registration network.

[0010] The technical solution adopted in the embodiment of the present application further includes: the preprocessing of the original video sequence is specifically:

[0011] Perform rigid registration on the original video sequence using the NormCorre method;

[0012] Perform mean-variance normalization and data augmentation operations on the rigidly registered video sequence to obtain the preprocessed video sequence.

[0013] The technical solution adopted in the embodiment of the present application further includes: the VideoSD network includes a feature extraction module, a registration module, and a denoising module. The feature extraction module is used to extract a feature map from the preprocessed video sequence, and the feature map contains the position information of the intermediate frame to be registered, where the intermediate frame is the original intermediate frame in the original video sequence; the input of the registration module is the feature map extracted by the feature extraction module and other frames to be registered, and the output is the previous and next frames registered with the intermediate frame; the denoising module is used to reconstruct the denoised intermediate frame using the registered previous and next frames.

[0014] The technical solution adopted in the embodiment of the present application further includes: the training process of the VideoSD network includes:

[0015] Take 2n + 1 consecutive video frames from the preprocessed video sequence, input the consecutive video frames into the VideoSD network. The VideoSD network extracts feature maps through a feature extraction module and outputs a feature map with the number of channels a. Combine all the feature maps with the front and back frames respectively to obtain 2n input frames with a size of batchsize*(a + 1)*x*y, where a is a user-defined value. Input the 2n input frames into a registration module for registration to obtain 2n sets of deformation fields for the middle frame, and apply the 2n sets of deformation fields to the 2n front and back frames to obtain the front and back frames registered with the middle frame. Input the front and back frames registered with the middle frame into a noise reduction module for noise reduction processing to obtain a noise-reduced middle frame.

[0016] The technical solution adopted in the embodiment of the present application further includes: The training process of the VideoSD network further includes:

[0017] Compare the noise-reduced middle frame and the middle frame to be reconstructed output by the feature extraction module with the original middle frame, calculate the L1 loss and the MSE loss, and calculate the smoothness loss of the deformation field of the front and back frames with respect to the middle frame. Optimize the VideoSD network through the decrease of the L1 loss, the MSE loss and the smoothness loss to obtain a trained noise reduction and registration network.

[0018] The technical solution adopted in the embodiment of the present application further includes: The feature extraction module is a two-dimensional convolutional deep learning network, and the registration module and the noise reduction module are respectively a two-dimensional convolutional network.

[0019] Another technical solution adopted in the embodiment of the present application is: A functional imaging noise reduction and registration device, including:

[0020] A preprocessing module: used to obtain an original video sequence and preprocess the original video sequence;

[0021] A model training module: used to extract a set number of consecutive video frames from the preprocessed video sequence, input the consecutive video frames into the VideoSD network, and perform network training with the output of a noise-reduced and registered middle frame with the front and back frames as the training target to obtain a trained noise reduction and registration network;

[0022] A noise reduction and registration module: used to extract a set number of consecutive video frames from a video sequence to be processed, replace the middle frame in the consecutive video frames with a set template frame, and then input it into the trained noise reduction and registration network, and output a noise-reduced and registered middle frame with a fixed frame through the noise reduction and registration network.

[0023] The technical solution adopted in the embodiment of the present application further includes: The VideoSD network includes a feature extraction module, a registration module, and a noise reduction module. The feature extraction module is used to extract a feature map from the preprocessed video sequence. The feature map contains the position information of the intermediate frame to be registered, where the intermediate frame is the original intermediate frame in the original video sequence. The input of the registration module is the feature map extracted by the feature extraction module and other frames to be registered, and the output is the front and back frames registered with the intermediate frame. The noise reduction module is used to reconstruct the denoised intermediate frame by using the registered front and back frames.

[0024] Another technical solution adopted in the embodiment of the present application is: A computer device, the computer device includes a processor and a memory coupled to the processor, where,

[0025] The memory stores program instructions for implementing the functional imaging noise reduction and registration method;

[0026] The processor is used to execute the program instructions stored in the memory to control the functional imaging noise reduction and registration method.

[0027] Another technical solution adopted in the embodiment of the present application is: A storage medium stores program instructions that can be run by a processor, and the program instructions are used to execute the functional imaging noise reduction and registration method.

[0028] Compared with the prior art, the beneficial effects produced by the embodiment of the present application are as follows: The functional imaging noise reduction and registration method, device, computer device, and storage medium in the embodiment of the present application regard denoising and registration as two interrelated tasks and use a common loss for training simultaneously, which is beneficial to improving the overall effect and making the process more concise and fast; The feature extraction module can better extract information related to registration and suppress interference information such as noise, achieving a faster and more accurate effect. By adopting a self-supervised deep learning method and optimizing the processing process by simultaneously performing noise reduction and motion calibration, the noise reduction and motion calibration effects can be improved, and at the same time, the data processing efficiency can be enhanced, providing a more accurate, simple, and efficient data processing method for bioscientists. Description of the Drawings

[0029] Figure 1 is a flowchart of the functional imaging noise reduction and registration method in the embodiment of the present application;

[0030] Figure 2 is a schematic structural diagram of the functional imaging noise reduction and registration device in the embodiment of the present application;

[0031] Figure 3 is a schematic structural diagram of the computer device in the embodiment of the present application;

[0032] Figure 4Schematic diagram of the storage medium according to an embodiment of the present application. Detailed implementation manners

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0034] The terms "first", "second", and "third" in the present application are only used for descriptive purposes, and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first", "second", and "third" may explicitly or implicitly include at least one of such features. In the description of the present application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or computer device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or computer devices.

[0035] Referring to "embodiment" herein means that a specific feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0036] Specifically, please refer to Figure 1 , which is a flowchart of the functional imaging noise reduction and registration method according to an embodiment of the present application. The functional imaging noise reduction and registration method according to an embodiment of the present application includes the following steps:

[0037] S100: Obtain an original video sequence and preprocess the original video sequence;

[0038] In this step, in the original video sequence of functional imaging, the deformation field between each video frame and the template frame should be continuous and smooth. In the embodiment of the present application, first, preprocessing such as rigid registration is performed on the obtained original video sequence with noise and displacement, and then the rigid registration result is input into the constructed VideoSD network for training. Specifically, in the embodiment of the present application, the existing method NormCorre is used for rigid registration, and then mean-variance normalization and data augmentation operations are performed on the original video sequence after rigid registration to obtain the preprocessed video sequence.

[0039] S110: Extract a set number of consecutive video frames from the preprocessed video sequence, input the extracted consecutive video frames into the VideoSD network, and use the output of the denoised and registered intermediate frame with the previous and next frames as the training target for network training to obtain the trained denoising and registration network;

[0040] In this step, the embodiment of the present application designs the VideoSD network structure based on two assumptions: First, the signal of the intermediate frame can be obtained by interpolating the previous and next N frames; Second, in microscopic imaging, the deformation field between each frame and the intermediate frame should be continuous and smooth. Based on this, the VideoSD network of the embodiment of the present application includes three parts: a feature extraction module, a registration module, and a denoising module. The feature extraction module is used to fuse the information of the intermediate frame and its previous and next frames, filter out unnecessary information such as random noise, and then extract the feature map suitable for registration. The feature map contains the position information of the intermediate frame to be registered; Among them, in order to make the network training faster, an additional output is added to the feature extraction module, and the reconstruction loss is calculated by combining this output with the intermediate frame, so that the feature extraction module can learn faster the information that needs to output the position of the intermediate frame as the main information; The input of the registration module is the feature map extracted by the feature extraction module and the other frames to be registered, and the output is the previous and next frames registered with the intermediate frame; The denoising module is used to reconstruct the denoised intermediate frame by using the registered previous and next frames. Among them, during network training, the intermediate frame is the original intermediate frame in the original video sequence. Using the original intermediate frame as the template frame and the original intermediate frame as the output target of the network, since the noise is independent and cannot be reconstructed by the previous and next frames, the network output is the denoised intermediate frame. Since the output is reconstructed from the previous and next frames, the feature extraction module only provides position information for the registration module, so that the brightness information of the template frame will not affect the final output result.

[0041] Further, the training process of the VideoSD network is specifically as follows: Take 2n + 1 consecutive video frames from a video sequence with noise and displacement, and input the 2n + 1 consecutive video frames into the VideoSD network. The VideoSD network first extracts feature maps through a feature extraction module and outputs feature maps (template features) with the number of channels being a. The feature maps contain the position information of the intermediate frame (the (n + 1)-th frame) to be reconstructed. Then, combine all the feature maps with the 2n frames before and after respectively to obtain 2n input frames with a size of batchsize*(a + 1)*x*y, where a is a user-defined value; input the 2n input frames into the registration module for registration to obtain 2n sets of deformation fields for the intermediate frame, and apply the 2n sets of deformation fields to the 2n frames before and after to obtain the frames before and after that are registered with the intermediate frame. Finally, input the frames before and after that are registered with the intermediate frame into the noise reduction module for noise reduction processing to obtain the denoised intermediate frame. Among them, the feature extraction module is a two-dimensional convolutional deep learning network, the size of its input data is batchsize*(2n + 1)*x*y, and the output includes template features with a data size of batchsize*a*x*y and the intermediate frame to be reconstructed with a size of batchsize*1*x*y. The registration module and the noise reduction module are respectively a two-dimensional convolutional network. In this embodiment of the application, the Unet network is taken as an example. It can be understood that the feature extraction module, the registration module, and the noise reduction module can also be replaced with other deep learning network architectures.

[0042] Further, during network training, by comparing the denoised intermediate frame and the intermediate frame to be reconstructed output by the feature extraction module with the original intermediate frame, calculate the L1 loss and the MSE (Mean Square Error) loss, and calculate the smoothness loss for the deformation field of the 2n frames before and after with respect to the intermediate frame, that is, calculate the difference between adjacent pixels in the x and y directions of the input, and then take the absolute value and average. Optimize the VideoSD network by using the three loss functions of the L1 loss, the MSE loss, and the smoothness loss to decrease, so that the network pays more attention to global information, thereby reducing the interference of noise on registration and outputting a more practical deformation field, and optimizing the registration effect through the noise reduction loss, thereby improving the registration effect and the noise reduction effect.

[0043] S120: Extract a set number of consecutive video frames from the video sequence to be processed, replace the intermediate frame in the consecutive video frames with a set template frame, and then input it into the trained noise reduction and registration network, and output the denoised intermediate frame that is registered with the fixed frame through the noise reduction and registration network;

[0044] In this step, during inference, first perform preprocessing such as rigid registration on the extracted 2n + 1 frames, and then input the preprocessing result into the trained noise reduction and registration network. The template frame can be replaced with the required template, which can be a certain fixed frame.

[0045] Based on the above, the functional imaging denoising and registration method of the embodiments of the present application regards denoising and registration as two interrelated tasks, and uses a common loss for simultaneous training, which is beneficial to improving the overall effect and makes the process simpler and faster; the feature extraction module can better extract information related to registration and suppress interference information such as noise, achieving a faster and more accurate effect. By adopting a self-supervised deep learning method and optimizing the processing flow by simultaneously performing denoising and motion calibration, the denoising and motion calibration effects can be improved, and the data processing efficiency can be enhanced at the same time, providing a more accurate, simple and efficient data processing method for bioscientists.

[0046] Please refer to Figure 2 for the structural schematic diagram of the functional imaging denoising and registration device of the embodiments of the present application. The functional imaging denoising and registration method device 40 of the embodiments of the present application includes:

[0047] A preprocessing module 41: configured to obtain an original video sequence and perform preprocessing on the original video sequence;

[0048] A model training module 42: configured to extract a set number of consecutive video frames from the preprocessed video sequence, input the consecutive video frames into the VideoSD network, and perform network training with the output of a denoised and registered intermediate frame with the previous and next frames as the training target to obtain a trained denoising and registration network;

[0049] A denoising and registration module 43: configured to extract a set number of consecutive video frames from the video sequence to be processed, replace the intermediate frame in the consecutive video frames with a set template frame, and then input the frames into the trained denoising and registration network, and output a denoised and registered intermediate frame with a fixed frame through the denoising and registration network.

[0050] It should be noted that for the information interaction, execution process, etc. between the above-mentioned devices / units, since they are based on the same concept as the method embodiments of the present application, their specific functions and the technical effects brought about can be specifically referred to in the method embodiment part, and will not be elaborated here.

[0051] The device provided by the embodiments of the present application can be applied in the foregoing method embodiments. For details, please refer to the description of the above method embodiments and will not be elaborated here.

[0052] Please refer to Figure 3 for the structural schematic diagram of the computer device of the embodiments of the present application. The computer device 50 includes:

[0053] A memory 51 storing executable program instructions;

[0054] A processor 52 connected to the memory 51;

[0055] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: obtain the original video sequence and preprocess the original video sequence; extract a set number of consecutive video frames from the preprocessed video sequence, input the consecutive video frames into the VideoSD network, and perform network training with the output denoised and registered intermediate frame with the previous and next frames as the training target to obtain a trained denoising and registration network; extract a set number of consecutive video frames from the video sequence to be processed, and replace the intermediate frame in the consecutive video frames with a set template frame and then input it into the trained denoising and registration network, and output the denoised and registered intermediate frame with the fixed frame through the denoising and registration network.

[0056] Among them, the processor 52 can also be called a CPU (Central Processing Unit, central processing unit). The processor 52 may be an integrated circuit chip with signal processing capabilities. The processor 52 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0057] Please refer to Figure 4, which is a schematic structural diagram of the storage medium according to an embodiment of the present application. The storage medium according to the embodiment of the present application stores program instructions 61 that can implement the following steps: obtaining an original video sequence and preprocessing the original video sequence; extracting a set number of consecutive video frames from the preprocessed video sequence, inputting the consecutive video frames into the VideoSD network, and training the network with the output of a denoised and registered intermediate frame with the previous and next frames as the training target to obtain a trained denoising and registration network; extracting a set number of consecutive video frames from the video sequence to be processed, replacing the intermediate frame in the consecutive video frames with a set template frame, and then inputting it into the trained denoising and registration network, and outputting a denoised and registered intermediate frame with a fixed frame through the denoising and registration network. Among them, the program instructions 61 can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network computer device, etc.) or a processor to execute all or part of the steps of the methods according to various embodiments of the present application. The foregoing storage medium includes: various media that can store program instructions such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, or terminal computer devices such as computers, servers, mobile phones, and tablets. Among them, the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0058] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0059] In addition, each functional unit in various embodiments of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. The above is only the implementation manner of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A functional imaging denoising and registration method, characterized in that: include: Obtaining an original video sequence and preprocessing the original video sequence; Extract a set number of continuous video frames from the preprocessed video sequence, input the continuous video frames into the VideoSD network, and perform network training with the intermediate frames after outputting noise reduction and registration with the previous and next frames as the training target, so as to obtain a trained noise reduction and registration network; A set number of continuous video frames are extracted from the video sequence to be processed, and the intermediate frames in the continuous video frames are replaced with the set template frames and then input into the trained denoising and registration network, and the denoising and registration network outputs the intermediate frames after denoising and registration with the fixed frame.

2. The functional imaging denoising and registration method according to claim 1, characterized in that: The preprocessing of the original video sequence is specifically as follows: Using the NormCorre method to perform rigid registration on the original video sequence; The video sequence after rigid registration is subjected to mean variance normalization and data augmentation operations to obtain a preprocessed video sequence.

3. The functional imaging denoising and registration method according to claim 2, characterized in that: The VideoSD network includes a feature extraction module, a registration module and a noise reduction module. The feature extraction module is used to extract a feature map from the preprocessed video sequence, and the feature map contains the position information of the intermediate frame that needs to be registered, wherein the intermediate frame is the original intermediate frame in the original video sequence; the input of the registration module is the feature map extracted by the feature extraction module and other frames that need to be registered, and the output is the front and back frames registered with the intermediate frame; the noise reduction module is used to reconstruct the denoised intermediate frame using the registered front and back frames.

4. The functional imaging denoising and registration method according to claim 3, characterized in that: The training process of the VideoSD network includes: 2n+1 continuous video frames are taken out from the preprocessed video sequence, and the continuous video frames are input into the VideoSD network. The VideoSD network extracts feature maps through a feature extraction module and outputs a feature map with a channel number a; all feature maps are combined with the front and back frames respectively to obtain 2n input frames with a size of batchsize*(a+1)*x*y, wherein a is a user-defined value; the 2n input frames are input into the registration module for registration to obtain 2n groups of deformation fields for the intermediate frames, and the 2n groups of deformation fields are applied to the front and back frames of the 2n frames to obtain the front and back frames registered with the intermediate frames; the front and back frames registered with the intermediate frames are input into the denoising module for denoising to obtain the denoised intermediate frames.

5. The functional imaging denoising and registration method according to claim 4, characterized in that: The training process of the VideoSD network also includes: The denoised intermediate frame and the intermediate frame to be reconstructed output by the feature extraction module are compared with the original intermediate frame, and the L1 loss and MSE loss are calculated, which is beneficial for calculating the smoothness loss of the deformation field of the intermediate frame for the previous and next frames. The VideoSD network is optimized by reducing the L1 loss, MSE loss and smoothness loss to obtain a trained denoising and registration network.

6. The functional imaging denoising and registration method according to any one of claims 1 to 5, characterized in that: The feature extraction module is a two-dimensional convolutional deep learning network, and the registration module and the noise reduction module are respectively a two-dimensional convolutional network.

7. A functional imaging noise reduction and registration device, characterized in that: include: Preprocessing module: used to obtain the original video sequence and preprocess the original video sequence; Model training module: used to extract a set number of continuous video frames from the preprocessed video sequence, input the continuous video frames into the VideoSD network, and perform network training with the intermediate frames after outputting noise reduction and registration with the previous and next frames as the training target, so as to obtain a trained noise reduction and registration network; Noise reduction and registration module: used to extract a set number of continuous video frames from the video sequence to be processed, and replace the intermediate frames in the continuous video frames with the set template frames and input them into the trained noise reduction and registration network, and output the intermediate frames after noise reduction and registration with the fixed frames through the noise reduction and registration network.

8. The functional imaging noise reduction and registration device according to claim 7, characterized in that: The VideoSD network includes a feature extraction module, a registration module and a noise reduction module. The feature extraction module is used to extract a feature map from the preprocessed video sequence, and the feature map contains the position information of the intermediate frame that needs to be registered, wherein the intermediate frame is the original intermediate frame in the original video sequence; the input of the registration module is the feature map extracted by the feature extraction module and other frames that need to be registered, and the output is the front and back frames registered with the intermediate frame; the noise reduction module is used to reconstruct the denoised intermediate frame using the registered front and back frames.

9. A computer device, characterized in that: The computer device includes a processor and a memory coupled to the processor, wherein: The memory stores program instructions for implementing the functional imaging denoising and registration method according to any one of claims 1 to 6; The processor is used to execute the program instructions stored in the memory to control a functional imaging noise reduction and registration method.

10. A storage medium, characterized in that: The device stores program instructions executable by a processor, wherein the program instructions are used to execute the functional imaging denoising and registration method according to any one of claims 1 to 6.

Images