Zero-sample low-light image enhancement method based on joint frequency domain prior guide diffusion

Abstract

The invention discloses a zero-sample low-light image enhancement method based on joint frequency domain prior guide diffusion, which relates to the field of optics and comprises the following steps: decomposing the frequency domain of a low-light image; performing forward diffusion and reverse sampling processing on the low-light image by using a Markov chain structure model, and supervising an optimization processing result in combination with a multi-modal text to obtain a sampling result; decomposing the frequency domain of the sampling result to obtain a sampling low-frequency domain, and performing conversion processing on the sampling low-frequency domain and the low-light low-frequency information by using Fourier transform to respectively obtain amplitude and phase information of the sampling low-frequency domain and the low-light low-frequency information; and in combination with the amplitude and phase information of the sampling low-frequency domain and the low-light low-frequency information and the low-light high-frequency information, on the basis of inverse fast Fourier transform and inverse discrete wavelet transform, an image enhancement output is obtained by updating a sampling result. According to the method, the problem of poor image quality of the low-light image is solved by combining the frequency domain prior guide diffusion method.

Description

Technical Field

[0001] The present invention relates to the field of optics, and in particular to a zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion. Background Art

[0002] Traditional low-light image enhancement methods mainly rely on optimizing the image's own parameters. Although these methods improve image quality to a certain extent, the manually designed prior information has poor adaptability. In complex and changeable low-light scenes, it is difficult to accurately and effectively handle different degrees of insufficient illumination and image degradation, resulting in unstable enhancement effects and significant performance differences between different images, making it difficult to meet the stringent requirements of practical applications for image quality consistency and stability. With the vigorous development of deep learning technology, research results on low-light image enhancement based on this technology have been remarkable. However, most studies have overly focused on using massive paired data to fit realistic lighting conditions. In the real world, it is difficult to obtain large-scale, high-quality paired data, and the debugging of the model is highly dependent on specific data sets, which severely limits the generalization ability of the model. This means that when the model faces complex and diverse low-light images in real scenes, it is often difficult to accurately restore the original clear appearance and rich details of the image, which greatly restricts its wide application in practical scenes.

[0003] In view of the above difficulties, unsupervised enhancement methods have gradually become a research hotspot. Strategies based on generative models to improve the perceived quality of low-light images have won certain recognition, among which the diffusion model stands out. It has attracted much attention due to its excellent generation effect and has emerged in the field of supervised image enhancement. However, when it is applied to unsupervised learning low-light image enhancement, due to the lack of prior information on lighting and content, its enhancement effect is greatly reduced in complex and changeable real-world scenes. For example, when processing unknown severely degraded low-light images, it is difficult to accurately grasp the structure and lighting characteristics of the image, resulting in a large gap between the visual effect of the generated image and the natural image, and it is impossible to achieve the ideal enhancement effect.

[0004] The wavelet domain and the Fourier frequency domain are closely related, which brings new opportunities to solve the problem of zero-sample low-light image enhancement. The wavelet low-frequency domain and Fourier amplitude both gather image illumination information, while the wavelet high-frequency domain and Fourier phase carry image structural information. Moreover, the low-frequency domain after wavelet decomposition can achieve better exposure effects than degraded images, and combined with the Fourier amplitude of normal images, it can accurately guide the restoration of illumination information. In addition, given that high-frequency information is easily damaged during the diffusion process, retaining high-frequency features helps stabilize the image structure and optimize data distribution. On this basis, integrating wavelet and Fourier frequency domains to construct a rich prior information embedded diffusion model has become a key breakthrough direction to make up for the lack of illumination and structural information in zero-sample enhancement and improve the stability and quality of the enhancement effect.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In order to overcome the above problems, the present invention aims to propose a zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion, which aims to make up for the lack of illumination and structural information in zero-sample enhancement and improve the stability and quality of low-light image enhancement effect.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] A zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion, the zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion includes:

[0009] S1, decomposing the frequency domain of the low-light image to obtain a low-light low-frequency domain, a low-light high-frequency domain, low-light low-frequency information, and low-light high-frequency information;

[0010] S2. Using the Markov chain structure model, the low-light image is forward diffused and reversely sampled, and the sampling result is obtained by combining the multimodal text supervision optimization processing result;

[0011] S3, decomposing the frequency domain of the sampling result to obtain a sampled low-frequency domain, converting the sampled low-frequency domain and the low-light low-frequency information using Fourier transform, and obtaining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information respectively;

[0012] S4, combining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information and the low-light high-frequency information, based on inverse fast Fourier transform and inverse discrete wavelet transform, and obtaining an image enhancement output by updating the sampling results;

[0013] The S2 includes:

[0014] Input the low-light image into the Markov chain structure model;

[0015] In the process of forward diffusion processing of low-light images, Gaussian noise is gradually added to the low-light images to obtain pure noise images;

[0016] In the process of reverse sampling of low-light images, the pure noise image is gradually denoised to obtain a preliminary sampling image.

[0017] Optionally, decomposing the frequency domain of the low-light image to obtain a low-light low-frequency domain, a low-light high-frequency domain, low-light low-frequency information, and low-light high-frequency information includes:

[0018] S11, using discrete wavelet transform, performing diffusion splitting on the low-light image to obtain a low-light low-frequency domain and a low-light high-frequency domain of the low-light image;

[0019] S12. Diffusion splitting is performed on the low-light and low-frequency domain according to discrete wavelet transform, so as to obtain low-light and low-frequency information and low-light and high-frequency information of the low-light and low-frequency domain respectively.

[0020] Optionally, S2 further includes:

[0021] Combining the multimodal model and non-reference brightness control constraints, multimodal text supervision is performed on the preliminary sampling results, and the sampling results are finally optimized.

[0022] Optionally, the expression of the forward diffusion process is:

[0023]

[0024] In the formula, q(x t |x 0 ) indicates that given the initial low-light image x 0 At time t the image x t The probability distribution of t represents the image data at time t; Represents the parameters in the diffusion process, controlling the Gaussian noise added to the image at each moment; represents the covariance matrix of the Gaussian distribution; I is the identity matrix.

[0025] Optionally, the reverse sampling process is expressed as:

[0026]

[0027] In the formula, Represents image data at time t Get the image data at time t-1 The probability distribution of ; N represents the normal distribution; Represents the image data restored at time t-1 during the reverse sampling process; represents the mean of the Gaussian distribution; represents the covariance matrix of Gaussian distribution; I represents the identity matrix; Controls the covariance size.

[0028] Optionally, the preliminary sampling results are supervised by multimodal text by combining the multimodal model and the non-reference brightness control constraint, and the final optimized sampling results include:

[0029] Based on the pre-trained multimodal model, input preset positive prompts and negative prompts into the text encoder to extract the text feature vector;

[0030] The image encoder is used to process the preliminary sampling results to extract image features, and the similarity loss between the image features and the text feature vector is calculated, and the difference between the image features and the text feature vector is determined according to the similarity loss;

[0031] Based on the difference between image features and text feature vectors, the brightness level learning parameters are optimized in combination with non-reference brightness control constraints. The preliminary sampling results are optimized and supervised according to the semantically guided calibration image feature space, and the sampling results are finally optimized.

[0032] Optionally, the expression of the non-reference brightness control constraint is:

[0033]

[0034] Where, L bri represents the non-reference brightness control constraint; represents the average intensity value of the mth non-overlapping local area; M represents the number of different areas in the image; and E represents the brightness level.

[0035] Optionally, combining the amplitude and phase information of the sampled low frequency domain and the low light low frequency information and the low light high frequency information, based on the inverse fast Fourier transform and the inverse discrete wavelet transform, by updating the sampling results, the image enhancement output includes:

[0036] Combining the wavelet low-frequency domain and Fourier amplitude information, the brightness prior is constructed;

[0037] Combining the wavelet high-frequency domain and Fourier phase information, the high-frequency information and phase of the low-light image are used to replace the high-frequency domain and phase of the sampling result, and a priori iterative guided sampling is constructed;

[0038] By performing multimodal text supervision on the prior iterative guided sampling, the updated sampling results are obtained;

[0039] Combining the updated sampling results with the low-light and high-frequency domains, an inverse discrete wavelet transform is performed to obtain the image enhancement output.

[0040] Optionally, the prior iterative guide sampling process is expressed as:

[0041]

[0042] In the formula, represents the prior iterative guided sampling result; IDWT represents inverse discrete wavelet transform; IFFT represents inverse fast Fourier transform; θ represents brightness learning factor; amp t Indicates the amplitude of the sampled low-frequency domain; amp L Represents the amplitude of low-light and low-frequency information; pha L The phase representing low-light and low-frequency information; Represents low-light high-frequency information.

[0043] Optionally, the expression for the image enhancement output is:

[0044]

[0045] In the formula, I E Represents the enhanced output of the image; denoise represents the denoising operation; IDWT represents the inverse discrete wavelet transform; Indicates the updated sampling result; H L Represents the low-light high-frequency domain.

[0046] Compared with the prior art, the present application has the following beneficial effects: the present invention enhances low-light images by combining Markov chain structure model, discrete wavelet transform, Fourier transform and network optimization. The enhanced low-light images have natural and realistic colors, accurately restore and enrich the image details, and the visual presentation is consistent with the human eye perception. It can effectively reconstruct the image visual effects in complex low-light scenes, improve the visual quality and recognizability, and strongly support its practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0048] Figure 1 is a flowchart of a zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion according to an embodiment of the present invention;

[0049] Figure 2 is an overall implementation flow chart of a zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion according to an embodiment of the present invention;

[0050] Figure 3 It is a visual qualitative comparison diagram of the results of using different enhancement methods in the zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] According to an embodiment of the present invention, a zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion is provided.

[0053] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1-Figure 3As shown, according to an embodiment of the present invention, a zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion comprises the following steps:

[0054] S1. Decompose the frequency domain of the low-light image to obtain a low-light low-frequency domain, a low-light high-frequency domain, low-light low-frequency information, and low-light high-frequency information.

[0055] Preferably, decomposing the frequency domain of the low-light image to obtain the low-light low-frequency domain, the low-light high-frequency domain, the low-light low-frequency information and the low-light high-frequency information includes:

[0056] S11, using discrete wavelet transform, performing diffusion splitting on the low-light image to obtain a low-light low-frequency domain and a low-light high-frequency domain of the low-light image;

[0057] S12. Diffusion splitting is performed on the low-light and low-frequency domain according to discrete wavelet transform, so as to obtain low-light and low-frequency information and low-light and high-frequency information of the low-light and low-frequency domain respectively.

[0058] It should be explained that the input low-light image I L Perform discrete wavelet transform (DWT) diffusion process to the low frequency domain, and input low light image I L Split into low-light low-frequency domain L L and low light high frequency domain H L , for low light and low frequency domain L L Perform discrete wavelet transform (DWT) again to retain the low-light low-frequency domain L L Low light and low frequency information and low light high frequency information This information will be used to construct priors and guide the update of the sampling process.

[0059] S2. Use the Markov chain structure model to perform forward diffusion and reverse sampling processing on the low-light image, and combine the multimodal text supervision optimization processing results to obtain the sampling results.

[0060] Preferably, S2 includes:

[0061] Input the low-light image into the Markov chain structure model;

[0062] In the process of forward diffusion processing of low-light images, Gaussian noise is gradually added to the low-light images to obtain pure noise images;

[0063] In the process of reverse sampling of low-light images, the pure noise image is gradually denoised to obtain a preliminary sampling image;

[0064] Combining the multimodal model and non-reference brightness control constraints, multimodal text supervision is performed on the preliminary sampling results, and the sampling results are finally optimized.

[0065] Preferably, the expression of the forward diffusion process is:

[0066]

[0067] In the formula, q(x t |x 0 ) indicates that given the initial low-light image x 0 At time t the image x t The probability distribution of t represents the image data at time t; Represents the parameters in the diffusion process, controlling the Gaussian noise added to the image at each moment; represents the covariance matrix of the Gaussian distribution; I is the identity matrix.

[0068] Preferably, the reverse sampling process is expressed as:

[0069]

[0070] In the formula, Represents image data at time t Get the image data at time t-1 The probability distribution of ; N represents the normal distribution; Represents the image data restored at time t-1 during the reverse sampling process; represents the mean of the Gaussian distribution; represents the covariance matrix of Gaussian distribution; I represents the identity matrix; Controls the covariance size.

[0071] Preferably, the multimodal model and the non-reference brightness control constraint are combined to perform multimodal text supervision on the preliminary sampling results, and the final optimized sampling results include:

[0072] Based on the pre-trained multimodal model, input preset positive prompts and negative prompts into the text encoder to extract the text feature vector;

[0073] The image encoder is used to process the preliminary sampling results to extract image features, and the similarity loss between the image features and the text feature vector is calculated, and the difference between the image features and the text feature vector is determined according to the similarity loss;

[0074] Based on the difference between image features and text feature vectors, the brightness level learning parameters are optimized in combination with non-reference brightness control constraints. The preliminary sampling results are optimized and supervised according to the semantically guided calibration image feature space, and the sampling results are finally optimized.

[0075] Preferably, the expression of the non-reference brightness control constraint is:

[0076]

[0077] Where, L bri represents the non-reference brightness control constraint; represents the average intensity value of the mth non-overlapping local area; M represents the number of different areas in the image; and E represents the brightness level.

[0078] It should be explained that the forward propagation process q of the Markov chain structure is to pass the initial low-light image x 0 This is achieved by gradually adding Gaussian noise until T time steps approximate the pure noise data x T The noise in the forward propagation process q is independent and follows a normal distribution. The expression of the forward diffusion process is:

[0079]

[0080] In the formula, q(x t |x 0 ) indicates that given the initial low-light image x 0 At time t the image x t The probability distribution of t represents the image data at time t; Represents the parameters in the diffusion process, controlling the Gaussian noise added to the image at each moment. a t =1-β t , t∈[1,…,T], β t Used to determine the parameters for adding Gaussian noise during forward diffusion; represents the covariance matrix of the Gaussian distribution; I is the identity matrix.

[0081] Reverse sampling process θ Mainly for pure noise x T Remove noise to restore image x 0 , the expression of the reverse sampling process is:

[0082]

[0083] In the formula, Represents image data at time t Get the image data at time t-1 The probability distribution of ; N represents the normal distribution; Represents the image data restored at time t-1 during the reverse sampling process; represents the mean of the Gaussian distribution, where ∈θ(x t ,t) represents the estimated noise; represents the covariance matrix of Gaussian distribution; I represents the identity matrix; Indicates the control covariance size.

[0084] Combining the multimodal model and non-reference brightness control constraints, multimodal text supervision is performed on the preliminary sampled images, the pre-trained CLIP model is introduced, and the preset positive prompt T is input to the text encoder. p and negative prompt T n Extract the feature vector and use the image encoder to sample the result x at each step t Extract features and then calculate the similarity loss L between the image and text vector in the CLIP latent space TG ,According to this supervised enhancement process, the image feature space is aligned with the semantic guidance ability to improve the image perception quality.

[0085] The expression of similarity loss LTG is:

[0086]

[0087] Where LTG represents the similarity loss between image and text vector; Φ image Represents the image encoder, which is used to sample the result x t Perform feature extraction and convert image data into feature vectors; Φ text T represents the text encoder, which is used to extract features from the preset text prompts and convert the text information into feature vectors; p Indicates that the preset is positive; T n Indicates preset negative prompt; T j represents the text prompt; j represents the sample number; e cos An index representing the cosine similarity between image features and text features.

[0088] Introducing non-reference brightness control constraint L bri , the brightness level learning parameter θ can be optimized to ensure good visual perception and brightness distribution of the sampling results, and enhance the stability and reliability of the image enhancement effect. Non-reference brightness control constraint L bri The expression is:

[0089]

[0090] Where, L bri represents the non-reference brightness control constraint; represents the average intensity value of the mth non-overlapping local area; M represents the number of different areas in the image; and E represents the brightness level.

[0091] S3. Decompose the frequency domain of the sampling result by discrete wavelet transform to obtain the sampled low-frequency domain of the sampling result. According to the sampled low-frequency domain and the low-light low-frequency information, use Fourier transform to obtain the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information respectively.

[0092] It should be explained that the expressions for the amplitude and phase information of the sampled low-frequency domain and low-light low-frequency information are:

[0093]

[0094] Where, amp t Indicates the amplitude of the sampled low-frequency domain; pha t Indicates the phase of low-light low-frequency domain; amp L Represents the amplitude of low-light and low-frequency information; pha L It represents the phase of low-light low-frequency information; FFT stands for Fourier transform.

[0095] S4. Combining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information and the low-light high-frequency information, based on inverse fast Fourier transform and inverse discrete wavelet transform, the image enhancement output is obtained by updating the sampling results.

[0096] Preferably, combining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information and the low-light high-frequency information, based on the inverse fast Fourier transform and the inverse discrete wavelet transform, by updating the sampling results, the image enhancement output includes:

[0097] Combining the wavelet low-frequency domain and Fourier amplitude information, the brightness prior is constructed;

[0098] Combining the wavelet high-frequency domain and Fourier phase information, the high-frequency information and phase of the low-light image are used to replace the high-frequency domain and phase of the sampling result, and a priori iterative guided sampling is constructed;

[0099] By performing multimodal text supervision on the prior iterative guided sampling, the updated sampling results are obtained;

[0100] Combining the updated sampling results with the low-light and high-frequency domains, an inverse discrete wavelet transform is performed to obtain the image enhancement output.

[0101] Preferably, the expression of the a priori iterative guided sampling process is:

[0102]

[0103] In the formula, represents the prior iterative guided sampling result; IDWT represents inverse discrete wavelet transform; IFFT represents inverse fast Fourier transform; θ represents brightness learning factor; amp t Indicates the amplitude of the sampled low-frequency domain; amp LRepresents the amplitude of low-light and low-frequency information; pha L The phase representing low-light and low-frequency information; Represents low-light high-frequency information.

[0104] Preferably, the expression of the image enhancement output is:

[0105]

[0106] In the formula, I E Represents the enhanced output of the image; denoise represents the denoising operation; IDWT represents the inverse discrete wavelet transform; Indicates the updated sampling result; H L Represents low light high frequency domain

[0107] It should be explained that, since the wavelet low-frequency domain and Fourier amplitude are closely related to the brightness and structural information of the image, the two are combined to construct the brightness prior. Considering that the wavelet high-frequency information and sampling phase are prone to produce random details that interfere with the distribution of image data, the low-light high-frequency information of the low-light image is used. and phase pha L Instead, it guides the generation of sampling content and maintains the stability of data distribution, ensuring the fidelity and consistency of data distribution.

[0108] like Figure 2 As shown in the figure, the overall implementation flow chart of the method is shown. In the experimental simulation process, the experimental framework is built on a single NVIDIA Tesla V100 GPU based on the PyTorch framework. The unconditional 256×256 diffusion model pre-trained on ImageNet is adopted. The pre-trained weights learned on large-scale image data give the model excellent initial feature extraction and expression capabilities, accelerate the convergence of low-light image enhancement tasks, and improve generalization. The total diffusion step size T is precisely set to 1000 to strictly control the gradual process of the image from the original pure state to the complete noise submergence. The alternating optimization interval step size S is determined to be 200, and the optimization adjustment is performed every 200 steps during inverse sampling to balance the computational cost and model learning efficiency, ensure the accurate optimization of model parameters in the key links of denoising, illumination compensation and detail restoration, and make the generated image gradually approach the distribution of natural image data.

[0109] In order to verify the effectiveness of the present invention, test images are selected from LOL (Deep Retinex Decomposition for Low-Light Enhancement) and SICE (Single Image Contrast Enhancement) paired data sets, and low-light images are collected from LIME (Low-light Image Enhancement via Illumination Map Estimation), DICM (Contrast enhancement based on layered difference representation), and MEF (Power-constrained contrast enhancement for emissive displays based on histogram equalization) data sets to form an “Unpaired-set” (unpaired data set). The paired dataset is evaluated using PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index), LPIPS (Learned Perceptual Image Patch Similarity), and FID (Frechet Inception Distance). The unpaired dataset is evaluated using MUSIQ (Multi-Scale Image Quality Assessment) and LOE (Low-Light Enhancement).

[0110] As shown in Table 1 below, the quantitative evaluation table of different unsupervised learning methods on the benchmark dataset. Among them, the zero-shot low-light image enhancement method based on joint frequency domain prior guided diffusion has outstanding indicators such as PSNR of 20.922, SSIM of 0.811, LPIPS as low as 0.281, and FID of 63.601 in the category of zero-shot methods. Its comprehensive performance ranks first, fully demonstrating the effectiveness and strong generalization of the model. Accurate quantification proves that it has significant advantages in improving image quality, optimizing structural similarity, reducing perceptual differences, and matching data distribution.

[0111] Table 1 Quantitative evaluation of different unsupervised learning methods on benchmark datasets

[0112]

[0113]

[0114] like Figure 3 As shown in the figure, the visual qualitative comparison of the results of different enhancement methods. Compared with other models, the method proposed in this paper has natural and realistic colors, accurately restores rich image details, and the visual presentation is consistent with human eye perception. It can effectively reconstruct the image visual effect in complex low-light scenes, improve visual quality and recognizability, and strongly support its practical application value.

[0115] Specifically, in order to facilitate better understanding by those skilled in the art, the relevant embodiments of the present application and the technical terms or some nouns that may be involved in the present application are now explained.

[0116] To sum up, with the help of the above-mentioned technical scheme of the present invention, by combining the Markov chain structure model, discrete wavelet transform, Fourier transform and network optimization, the low-light image is enhanced. The enhanced low-light image has natural and realistic colors, accurately restores and enriches the image details, and the visual presentation is consistent with the perception of the human eye. It can effectively reconstruct the image visual effect in complex low-light scenes, improve the visual quality and recognizability, and strongly support its practical application value.

[0117] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion, characterized in that: include: S1, decomposing the frequency domain of the low-light image to obtain a low-light low-frequency domain, a low-light high-frequency domain, low-light low-frequency information, and low-light high-frequency information; S2. Using the Markov chain structure model, the low-light image is forward diffused and reversely sampled, and the sampling result is obtained by combining the multimodal text supervision optimization processing result; S3, decomposing the frequency domain of the sampling result to obtain a sampled low-frequency domain, converting the sampled low-frequency domain and the low-light low-frequency information using Fourier transform, and obtaining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information respectively; S4, combining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information and the low-light high-frequency information, based on inverse fast Fourier transform and inverse discrete wavelet transform, and obtaining an image enhancement output by updating the sampling results; The S2 includes: Input the low-light image into the Markov chain structure model; In the process of forward diffusion processing of low-light images, Gaussian noise is gradually added to the low-light images to obtain pure noise images; In the process of reverse sampling of low-light images, the pure noise image is gradually denoised to obtain a preliminary sampling image.

2. The zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion according to claim 1, characterized in that: Decomposing the frequency domain of the low-light image to obtain the low-light low-frequency domain, the low-light high-frequency domain, the low-light low-frequency information and the low-light high-frequency information includes: S11, using discrete wavelet transform, performing diffusion splitting on the low-light image to obtain a low-light low-frequency domain and a low-light high-frequency domain of the low-light image; S12. Diffusion splitting is performed on the low-light and low-frequency domain according to discrete wavelet transform, so as to obtain low-light and low-frequency information and low-light and high-frequency information of the low-light and low-frequency domain respectively.

3. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 1, characterized in that: The S2 further includes: Combining the multimodal model and non-reference brightness control constraints, multimodal text supervision is performed on the preliminary sampling results, and the sampling results are finally optimized.

4. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 1, characterized in that: The expression of the forward diffusion process is: In the formula, q(x t |x0) represents the image x at time t given the initial low-light image x0 t The probability distribution of t represents the image data at time t; Represents the parameters in the diffusion process, controlling the Gaussian noise added to the image at each moment; represents the covariance matrix of the Gaussian distribution; I is the identity matrix.

5. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 1, characterized in that: The expression of the reverse sampling process is: In the formula, Represents image data at time t Get the image data at time t-1 The probability distribution of ; N represents the normal distribution; Represents the image data restored at time t-1 during the reverse sampling process; represents the mean of the Gaussian distribution; represents the covariance matrix of Gaussian distribution; I represents the identity matrix; Controls the covariance size.

6. The zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion according to claim 3, characterized in that: The multimodal model and the non-reference brightness control constraint are combined to perform multimodal text supervision on the preliminary sampling results, and the final optimized sampling results include: Based on the pre-trained multimodal model, input preset positive prompts and negative prompts into the text encoder to extract the text feature vector; The image encoder is used to process the preliminary sampling results to extract image features, and the similarity loss between the image features and the text feature vector is calculated, and the difference between the image features and the text feature vector is determined according to the similarity loss; Based on the difference between image features and text feature vectors, the brightness level learning parameters are optimized in combination with non-reference brightness control constraints. The preliminary sampling results are optimized and supervised according to the semantically guided calibration image feature space, and the sampling results are finally optimized.

7. The zero-sample low-light image enhancement method based on joint frequency domain prior-guided diffusion according to claim 6, characterized in that: The expression of the non-reference brightness control constraint is: Where, L bri represents the non-reference brightness control constraint; represents the average intensity value of the mth non-overlapping local area; M represents the number of different areas in the image; and E represents the brightness level.

8. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 1, characterized in that: The image enhancement output obtained by combining the amplitude and phase information of the sampled low-frequency domain and the low-light low-frequency information and the low-light high-frequency information based on the inverse fast Fourier transform and the inverse discrete wavelet transform by updating the sampling results includes: Combining the wavelet low-frequency domain and Fourier amplitude information, the brightness prior is constructed; Combining the wavelet high-frequency domain and Fourier phase information, the high-frequency information and phase of the low-light image are used to replace the high-frequency domain and phase of the sampling result, and a priori iterative guided sampling is constructed; By performing multimodal text supervision on the prior iterative guided sampling, the updated sampling results are obtained; Combining the updated sampling results with the low-light and high-frequency domains, an inverse discrete wavelet transform is performed to obtain the image enhancement output.

9. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 8, characterized in that: The expression of the a priori iterative guided sampling process is: In the formula, represents the prior iterative guided sampling result; IDWT stands for inverse discrete wavelet transform; IFFT stands for inverse fast Fourier transform; θ stands for brightness learning factor; amp t Indicates the amplitude of the sampled low-frequency domain; amp L Indicates the amplitude of low-light and low-frequency information; pha L The phase representing low-light and low-frequency information; Represents low-light high-frequency information.

10. The zero-sample low-light image enhancement method based on joint frequency domain prior guided diffusion according to claim 8, characterized in that: The expression of the image enhancement output is: In the formula, I E Represents the enhanced output of the image; denoise represents the denoising operation; IDWT stands for Inverse Discrete Wavelet Transform; Indicates the updated sampling result; H L Represents the low-light high-frequency domain.

Images