High dynamic range imaging method based on progressive generative network

Abstract

The invention discloses a high dynamic range imaging method based on a progressive generative network. The method comprises the steps that a progressive generative network is constructed, and the progressive generative network comprises a pseudo-static low-dynamic-range generative network and a detail guiding high-dynamic-range generative network; a plurality of low dynamic range images are acquired, and the plurality of low dynamic range images are respectively reference frame images and non-reference frame images; processing the plurality of low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to the non-reference frame image; and inputting the initial pseudo-static low-dynamic-range image and the reference frame image into a detail-guided high-dynamic-range generation network, and outputting a high-dynamic-range image. According to the invention, the technical problems of insufficient accuracy and detail loss when the ghosting problem caused by large-amplitude motion in HDR imaging is processed in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a high dynamic range imaging method based on a progressive generation network. Background Art

[0002] High Dynamic Range (HDR) imaging technology aims to address the limitations of traditional digital photographic sensors in capturing a wide range of real-world luminance. Since most digital photographic sensors can only capture a small part of the luminance range in the environment, the resulting Low Dynamic Range (LDR) images are often overexposed or underexposed, resulting in the loss of important image details. To address this problem, HDR imaging technology was developed to fully represent scene details under various lighting conditions in a single image. With the advancement of technology, HDR imaging has been widely used in fields such as saliency detection and video compression.

[0003] In dynamic scenes, it is a common method to generate HDR images from LDR images with different exposure levels, namely Multi-exposure Fusion (MEF). Although MEF can produce satisfactory results in static scenes, it often produces ghosting artifacts in dynamic scenes, especially when shooting with a handheld camera.

[0004] In order to solve the ghosting problem in HDR imaging, researchers have developed a variety of methods, including alignment-based techniques, exclusion-based strategies, and patch-based solutions. Although these methods have their own advantages in reducing ghosting, they still face the problem of insufficient accuracy or loss of details when dealing with large motion or occlusion scenes. With the development of deep neural networks (DNN), DNN-based methods have been widely used in HDR imaging to solve the ghosting problem. However, due to the limitation of the receptive field, traditional methods based on convolutional neural networks (CNN) are difficult to effectively deal with ghosting problems caused by large motion. Although the Transformer-based method can capture global information, it is insufficient in generating edge details. Summary of the invention

[0005] The embodiment of the present invention provides a high dynamic range imaging method based on a progressive generation network, so as to at least solve the technical problems of insufficient accuracy and loss of details when the prior art processes the ghosting problem caused by large motion in HDR imaging.

[0006] According to one aspect of an embodiment of the present invention, a high dynamic range imaging method based on a progressive generation network is provided. The method may include: constructing a progressive generation network, wherein the progressive generation network includes: a pseudo-static low dynamic range generation network and a detail-guided high dynamic range generation network; acquiring multiple low dynamic range images, wherein the multiple low dynamic range images are reference frame images and non-reference frame images respectively; processing the multiple low dynamic range images through the pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to the non-reference frame image; inputting the initial pseudo-static low dynamic range image and the reference frame image into the detail-guided high dynamic range generation network, and outputting a high dynamic range image.

[0007] Optionally, the method processes multiple low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to a non-reference frame image, including: using a semantic segmentation model to segment each low dynamic range image separately to generate an initial mask of multiple objects moving in each low dynamic range image; combining the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image to generate a target mask of the target object corresponding to the non-reference frame image, wherein the background initial mask of the reference frame image is determined by the initial masks of multiple objects moving in the reference frame image; and inputting the target mask of the target object corresponding to the non-reference frame image into a diffusion model to obtain a pseudo-static low dynamic range image corresponding to the non-reference frame image.

[0008] Optionally, the expression for segmenting each low dynamic range image respectively by using the semantic segmentation model to generate initial masks of multiple moving objects in each low dynamic range image is:

[0009] M i =SAM(L i ,text),i=1,2,3

[0010] Where i represents the number of low dynamic range images. When i=1,3, L i Represents a non-reference frame image, when i=2, L i represents the reference frame image, text represents multiple objects moving in the low dynamic range image, M i An initial mask for each object representing motion in each low dynamic range image.

[0011] Optionally, the expression for combining the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image to generate the target mask of the target object corresponding to the non-reference frame image is:

[0012] M′ i =M i ×(1-M2),i=1,3

[0013] Among them, M′ i It represents the target mask of the target object corresponding to the non-reference frame image, 1 represents the mask is 1, M2 represents the initial mask of multiple objects moving in the reference frame image, the background initial mask of the reference frame image, and 1-M2 represents the background initial mask of the reference frame image.

[0014] Optionally, the target mask of the target object corresponding to the non-reference frame image is input into the diffusion model to obtain an expression of a pseudo-static low dynamic range image corresponding to the non-reference frame image:

[0015] L′ i =SDB(M′ i ,L i ),i=1,3

[0016] Among them, L′ i is a pseudo-static low dynamic range image corresponding to the non-reference frame image, and SDB represents a stable diffusion model.

[0017] Optionally, the step of inputting the initial pseudo-static low dynamic range image and the reference frame image into a detail-guided high dynamic range generation network and outputting a high dynamic range image comprises: performing Gamma correction on the initial pseudo-static low dynamic range image and the reference frame image to obtain the corrected initial pseudo-static low dynamic range image and the reference frame image; connecting the initial pseudo-static low dynamic range image and the reference frame image as well as the corrected initial pseudo-static low dynamic range image and the reference frame image in the channel dimension to obtain a plurality of initial features of the target pseudo-static low dynamic range image, wherein the target pseudo-static low dynamic range image is a collection of the initial pseudo-static low dynamic range image and the reference frame image; and using a feature extractor to extract features from the plurality of initial features respectively to obtain each initial feature. each target feature corresponding to the target; each target feature is connected to obtain an initial feature map; the initial feature map is processed by the channel attention mechanism to obtain the feature guide corresponding to the initial feature map; the target pseudo-static low dynamic range image is feature extracted by Laplace transform to obtain the initial high-frequency feature corresponding to the target pseudo-static low dynamic range image; the initial high-frequency feature and the feature guide are feature fused, and the fused feature is convolved to obtain the final high-frequency feature; the initial feature map is feature extracted by the feature extractor to obtain the feature of the initial feature map; the final high-frequency feature and the feature of the initial feature map are processed by multiple detail enhancement modules, and the features processed by each detail enhancement module are feature fused to output a high dynamic range image.

[0018] Optionally, the initial pseudo-static low dynamic range image and the reference frame image are subjected to Gamma correction to obtain expressions of the corrected initial pseudo-static low dynamic range image and the reference frame image:

[0019]

[0020] Among them, when i=1,3, H i represents the corrected non-pseudo-static low dynamic range image, γ represents the Gamma correction parameter, γ = 2.2, H2 represents the corrected reference frame image, t i Represents image L i ′, and t2 represents the exposure time of image L2.

[0021] Optionally, the initial pseudo-static low dynamic range image and the reference frame image as well as the corrected initial pseudo-static low dynamic range image and the reference frame image are connected in the channel dimension to obtain the expressions of multiple initial features of the target pseudo-static low dynamic range image:

[0022] X i =Concat(L′ i ,H i ),i=1,2,3

[0023] Among them, when i=1,3, X i represents the initial features of the pseudo-static low dynamic range image. When i=2, X i Represents the initial features of the reference frame image.

[0024] Beneficial effects of the present invention:

[0025] The present invention proposes a high dynamic range imaging method based on a progressive generative network, constructs a progressive generative network, and the progressive generative network is composed of a pseudo-static low dynamic range generation network and a detail-guided high dynamic range generation network. In the pseudo-static LDR generation stage, a Segment Anything Model (SAM) with strong generalization ability is introduced for segmentation at the pixel level, so that the motion area can be identified even in overexposed real scenes. Then, a stable diffusion model is used to repair the large motion area to generate a series of pseudo-static images. In the detail-guided HDR generation stage, a shared high-frequency feature extraction module and a detail enhancement module are proposed, and high-quality HDR images are generated through these modules, which significantly reduce ghost artifacts and restore image details, solve the technical problems of insufficient accuracy and detail loss when the prior art processes the ghost problem caused by large motion in HDR imaging, and achieves the technical effect of processing the ghost problem caused by large motion in HDR imaging through a progressive generative network with high accuracy and no detail loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0027] Figure 1 is a flowchart of a high dynamic range imaging method based on a progressive generation network according to an embodiment of the present invention;

[0028] Figure 2 is a plurality of low dynamic range images with different exposures according to an embodiment of the present invention;

[0029] Figure 3 is a high dynamic range image according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] Example 1

[0033] According to an embodiment of the present invention, a high dynamic range imaging method based on a progressive generation network is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system comprising at least one set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] Figure 1is a flow chart of a high dynamic range imaging method based on a progressive generation network according to an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0035] Step S101, constructing a progressive generation network, wherein the progressive generation network includes: a pseudo-static low dynamic range generation network and a detail-guided high dynamic range generation network.

[0036] In the technical solution provided in the above step S101 of the present invention, a progressive generation network is constructed, and the progressive generation network consists of two parts, one part is a pseudo-static low dynamic range generation network, and the other part is a detail-guided high dynamic range generation network.

[0037] Step S102 : acquiring a plurality of low dynamic range images, wherein the plurality of low dynamic range images are reference frame images and non-reference frame images respectively.

[0038] In the technical solution provided in the above step S102 of the present invention, a plurality of low dynamic range images are obtained. Figure 2 are multiple frames of low dynamic range images with different exposures according to an embodiment of the present invention, such as Figure 2 As shown, the marks from left to right are 1, 2, and 3, respectively, wherein the images marked 1 and 3 are non-reference frame images, and the image marked 2 is a reference frame image.

[0039] Step S103, processing the multiple low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to the non-reference frame image.

[0040] In the technical solution provided in the above step S103 of the present invention, a plurality of low dynamic range images are input into a pseudo-static low dynamic range generation network to obtain an initial pseudo-static low dynamic range image corresponding to the non-reference frame image.

[0041] Step S104, inputting the initial pseudo-static low dynamic range image and the reference frame image into a detail-guided high dynamic range generation network, and outputting a high dynamic range image.

[0042] In the technical solution provided in the above step S104 of the present invention, the detail-guided high dynamic range generation network processes the initial pseudo-static low dynamic range image and the reference frame image to output a high dynamic range image.

[0043] The above method of this embodiment is further introduced below.

[0044] As an optional implementation method, step S103, the processing of multiple low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to a non-reference frame image includes: using a semantic segmentation model to segment each low dynamic range image respectively to generate an initial mask of multiple objects moving in each low dynamic range image; combining the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image to generate a target mask of the target object corresponding to the non-reference frame image, wherein the background initial mask of the reference frame image is determined by the initial masks of multiple objects moving in the reference frame image; and inputting the target mask of the target object corresponding to the non-reference frame image into a diffusion model to obtain a pseudo-static low dynamic range image corresponding to the non-reference frame image.

[0045] In this embodiment, when the multiple objects moving in each low dynamic range image are two objects, for example, a person and a car, the semantic segmentation model segments the two objects in each low dynamic range image respectively, generates an initial mask for each object moving in each low dynamic range image, multiplies the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image, and obtains a target mask of the target object corresponding to the non-reference frame image. The target object may be a person. The target mask of the target object corresponding to the non-reference frame image is input into the diffusion model to obtain a pseudo-static low dynamic range image corresponding to the non-reference frame image.

[0046] As an optional embodiment, the semantic segmentation model is used to segment each low dynamic range image respectively, and the expression for generating the initial mask of multiple moving objects in each low dynamic range image is:

[0047] M i =SAM(L i ,text),i=1,2,3

[0048] Where i represents the number of low dynamic range images. When i=1,3, L i Represents a non-reference frame image, when i=2, L i represents the reference frame image, text represents multiple objects moving in the low dynamic range image, M i An initial mask for each object representing motion in each low dynamic range image.

[0049] As an optional embodiment, the expression for combining the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image to generate the target mask of the target object corresponding to the non-reference frame image is:

[0050] M′ i =M i×(1-M2),i=1,3

[0051] Among them, M′ i It represents the target mask of the target object corresponding to the non-reference frame image, 1 represents the mask is 1, M2 represents the initial mask of multiple objects moving in the reference frame image, the background initial mask of the reference frame image, and 1-M2 represents the background initial mask of the reference frame image.

[0052] As an optional implementation manner, the target mask of the target object corresponding to the non-reference frame image is input into the diffusion model to obtain the expression of the pseudo-static low dynamic range image corresponding to the non-reference frame image:

[0053] L′ i =SDB(M′ i ,L i ),i=1,3

[0054] Among them, L′ i is a pseudo-static low dynamic range image corresponding to the non-reference frame image, and SDB represents a stable diffusion model.

[0055] As an optional implementation method, step S104, inputting the initial pseudo-static low dynamic range image and the reference frame image into the detail-guided high dynamic range generation network to output the high dynamic range image, includes: performing Gamma correction on the initial pseudo-static low dynamic range image and the reference frame image to obtain the corrected initial pseudo-static low dynamic range image and the reference frame image; connecting the initial pseudo-static low dynamic range image and the reference frame image and the corrected initial pseudo-static low dynamic range image and the reference frame image in the channel dimension to obtain multiple initial features of the target pseudo-static low dynamic range image, wherein the target pseudo-static low dynamic range image is a collection of the initial pseudo-static low dynamic range image and the reference frame image; and using a feature extractor to extract features from the multiple initial features respectively. Obtain each target feature corresponding to each initial feature; connect each target feature to obtain an initial feature map; process the initial feature map through a channel attention mechanism to obtain a feature guide corresponding to the initial feature map; extract features of the target pseudo-static low dynamic range image through Laplace transform to obtain the initial high-frequency features corresponding to the target pseudo-static low dynamic range image; fuse the initial high-frequency features and feature guides, and perform a convolution operation on the fused features to obtain the final high-frequency features; extract features from the initial feature map through a feature extractor to obtain the features of the initial feature map; process the final high-frequency features and the features of the initial feature map through multiple detail enhancement modules, and fuse the features processed by each detail enhancement module to output a high dynamic range image.

[0056] In this embodiment, the set of initial pseudo-static low dynamic range images and reference frame images {L′1, L2, L′3} is subjected to Gamma correction to obtain the corrected initial pseudo-static low dynamic range images and reference frame images. The purpose of the correction is to enhance their dynamic range and adapt to exposure changes. The multiple initial features of the target pseudo-static low dynamic range image can be represented by {X1, X2, X3}. The feature extractor is used to extract features from the multiple initial features to obtain each target feature corresponding to each initial feature. Each target feature is represented by {F1, F2, F3}, and each target feature is connected into a single initial feature map S. The initial feature map is processed by the channel attention mechanism, and the expression of the feature guidance corresponding to the initial feature map is obtained as follows:

[0057] FG 1×1×3 =Conv1(Sigmoid9Conv3(MaxPool(S H×W×3C )

[0058] +AvgPool(S H×W×3C ))))

[0059] Among them, Conv1() and Conv3() represent 1x1 convolution operations and 3x3 convolution operations respectively, MaxPool represents the maximum pooling operation, AvgPool represents the average pooling operation, and FG 1×1×3 is the feature guide corresponding to the initial feature map, H, W are the width and height of the initial feature map, and 3C is the number of channels of the initial feature map.

[0060] The initial high-frequency features and feature guides are fused, and the fused features are convolved to obtain the final high-frequency feature expression:

[0061]

[0062] in, is the final high frequency characteristic, HF' H×W×3 is the initial high frequency feature.

[0063] The target pseudo-static low dynamic range image {L′1, L2, L′3} is subjected to feature extraction through Laplace transform to obtain initial high-frequency features HF1, HF2, HF3 corresponding to the target pseudo-static low dynamic range image.

[0064] The feature extractor extracts the features of the initial feature map and obtains the features f′∈R of the initial feature map. h×w×c , the feature extractor consists of a 7x7 depthwise separable convolution and multiple 3x3 and 1x1 small convolution layers.

[0065] Each detail enhancement module consists of an average separation block (ASB) and a maximum separation block (MSB). ASB uses an average pooling (AvgPool) operation to transform the input feature f′∈R h×w×c Converted to low frequency features f l ∈R h / 2×w / 2×c , and then upsampled by bilinear interpolation to generate upsampled features f u ∈R h×w×c The expression is as follows:

[0066] f l =AvgPool(f'),f u =UpSampling(f l )

[0067] By taking the upsampled low-frequency features f u Subtract the feature f′ from the initial feature map to obtain a specific high-frequency feature f h The same applies when using MSB.

[0068] In order to supplement the rich and complete detail information, the shared final high-frequency feature HF is used s As a guide, specific high-frequency features are enhanced through the self-attention mechanism (SA), which can be expressed as:

[0069] f h =SA(HF s ×(f'-f u ))

[0070] Figure 3 is a high dynamic range image according to an embodiment of the present invention.

[0071] As an optional implementation manner, the initial pseudo-static low dynamic range image and the reference frame image are subjected to Gamma correction, and the expressions of the corrected initial pseudo-static low dynamic range image and the reference frame image are obtained as follows:

[0072]

[0073] Among them, when i=1,3, H i represents the corrected non-pseudo-static low dynamic range image, γ represents the Gamma correction parameter, γ = 2.2, H2 represents the corrected reference frame image, t i Represents image L i ′, and t2 represents the exposure time of image L2.

[0074] As an optional embodiment, the initial pseudo-static low dynamic range image and the reference frame image as well as the corrected initial pseudo-static low dynamic range image and the reference frame image are connected in the channel dimension to obtain multiple initial features of the target pseudo-static low dynamic range image:

[0075] X i =Concat(L′ i ,H i ),i=1,2,3

[0076] Among them, when i=1,3, X i represents the initial features of the pseudo-static low dynamic range image. When i=2, X i Represents the initial features of the reference frame image.

[0077] In an embodiment of the present invention, a progressive generation network is constructed, wherein the progressive generation network includes: a pseudo-static low dynamic range generation network and a detail-guided high dynamic range generation network; a plurality of low dynamic range images are obtained, wherein the plurality of low dynamic range images are reference frame images and non-reference frame images respectively; the plurality of low dynamic range images are processed by the pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to the non-reference frame image; the initial pseudo-static low dynamic range image and the reference frame image are input into the detail-guided high dynamic range generation network to output a high dynamic range image, thereby solving the technical problem of insufficient accuracy and loss of details when the prior art processes the ghost problem caused by large motion in HDR imaging, and achieving the technical effect of processing the ghost problem caused by large motion in HDR imaging by the progressive generation network with high accuracy and no loss of details.

[0078] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0079] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0081] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0082] In addition, each functional unit in each embodiment of the present invention may be integrated into a first processing unit, or each unit may exist physically separately, 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 software functional units.

[0083] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A high dynamic range imaging method based on a progressive generative network, characterized in that: include: Constructing a progressive generation network, wherein the progressive generation network includes: a pseudo-static low dynamic range generation network and a detail-guided high dynamic range generation network; Acquire a plurality of low dynamic range images, wherein the plurality of low dynamic range images are reference frame images and non-reference frame images respectively; Processing multiple low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to a non-reference frame image; The initial pseudo-static low dynamic range image and the reference frame image are input into the detail-guided high dynamic range generation network, and a high dynamic range image is output.

2. The method according to claim 1, characterized in that The method of processing a plurality of low dynamic range images through a pseudo-static low dynamic range generation network to generate an initial pseudo-static low dynamic range image corresponding to a non-reference frame image includes: Using the semantic segmentation model to segment each low dynamic range image separately, generating initial masks of multiple objects moving in each low dynamic range image; Combining an initial mask of a target object in a non-reference frame image with an initial background mask of a reference frame image to generate a target mask of the target object corresponding to the non-reference frame image, wherein the initial background mask of the reference frame image is determined by initial masks of a plurality of objects moving in the reference frame image; The target mask of the target object corresponding to the non-reference frame image is input into the diffusion model to obtain a pseudo-static low dynamic range image corresponding to the non-reference frame image.

3. The method according to claim 2, characterized in that The expression for segmenting each low dynamic range image by using the semantic segmentation model and generating the initial mask of multiple moving objects in each low dynamic range image is: M i =AM(L i ,text),i=1,2,3 Where i represents the number of low dynamic range images. When i=1,3, L i Represents a non-reference frame image, when i=2, L i represents the reference frame image, text represents multiple objects moving in the low dynamic range image, M i An initial mask for each object representing motion in each low dynamic range image.

4. The method according to claim 2, characterized in that: The expression for combining the initial mask of the target object in the non-reference frame image with the background initial mask of the reference frame image to generate the target mask of the target object corresponding to the non-reference frame image is: M i ′= i ×(1-2),i=1,3 Among them, M i ′ represents the target mask of the target object corresponding to the non-reference frame image, 1 represents the mask is 1, M2 represents the initial mask of multiple objects moving in the reference frame image, the background initial mask of the reference frame image, and 1-M2 represents the background initial mask of the reference frame image.

5. The method according to claim 2, characterized in that: The target mask of the target object corresponding to the non-reference frame image is input into the diffusion model to obtain the expression of the pseudo-static low dynamic range image corresponding to the non-reference frame image: L′ i =SDB(M i ′, i ),i=1,3 Among them, L′ i is a pseudo-static low dynamic range image corresponding to the non-reference frame image, and SDB represents a stable diffusion model.

6. The method according to claim 1, characterized in that The initial pseudo-static low dynamic range image and the reference frame image are input into a detail-guided high dynamic range generation network to output a high dynamic range image, including: Gamma correction is performed on the initial pseudo-static low dynamic range image and the reference frame image to obtain a corrected initial pseudo-static low dynamic range image and the reference frame image; The initial pseudo-static low dynamic range image and the reference frame image as well as the corrected initial pseudo-static low dynamic range image and the reference frame image are connected in the channel dimension to obtain a plurality of initial features of the target pseudo-static low dynamic range image, wherein the target pseudo-static low dynamic range image is a collection of the initial pseudo-static low dynamic range image and the reference frame image; Using a feature extractor to extract features from multiple initial features respectively, to obtain each target feature corresponding to each initial feature; Connect each target feature to obtain the initial feature map; The initial feature map is processed through the channel attention mechanism to obtain the feature guidance corresponding to the initial feature map; Extract features of the target pseudo-static low dynamic range image by Laplace transform to obtain initial high-frequency features corresponding to the target pseudo-static low dynamic range image; The initial high-frequency features and feature guides are fused, and the fused features are convolved to obtain the final high-frequency features; Extract features from the initial feature map using a feature extractor to obtain features of the initial feature map; The final high-frequency features and the features of the initial feature map are processed through multiple detail enhancement modules, and the features processed by each detail enhancement module are fused to output a high dynamic range image.

7. The method according to claim 6, characterized in that The initial pseudo-static low dynamic range image and the reference frame image are subjected to Gamma correction, and the expressions for obtaining the corrected initial pseudo-static low dynamic range image and the reference frame image are: Among them, when i=1,3, H i represents the corrected non-pseudo-static low dynamic range image, γ represents the Gamma correction parameter, γ = 2.2, H2 represents the corrected reference frame image, t i Represents image L i ′, and t2 represents the exposure time of image L2.

8. The method according to claim 6, characterized in that The initial pseudo-static low dynamic range image and the reference frame image as well as the corrected initial pseudo-static low dynamic range image and the reference frame image are connected in the channel dimension to obtain multiple initial features of the target pseudo-static low dynamic range image: X i =Concat(L′ i ,H i ),i=1,2,3 Among them, when i=1,3, X i represents the initial features of the pseudo-static low dynamic range image. When i=2, X i Represents the initial features of the reference frame image.

9. A processor, characterized in that: The processor is used to run a program, wherein the program executes the method according to any one of claims 1 to 8 when running.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 8.

Images