Parameterization generation method and device of spatial variational fuzzy kernel and storage medium
By generating a spatial variational blur kernel through a hyperparameter generation network and a parser decoder, the problems of high computational complexity and insufficient stability in existing technologies are solved, achieving efficient and stable image restoration results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-22
- Publication Date
- 2026-04-14
AI Technical Summary
Existing technologies suffer from high computational complexity, insufficient stability, or difficulty in achieving both physical plausibility and computational reliability when modeling spatial variational fuzzy kernels, resulting in poor image restoration performance.
A hyperparameter generation network is used to generate a hyperparameter mapping graph. A parser decoder generates local fuzzy kernels position by position based on a parameterized fuzzy kernel model and performs normalization processing. Finally, the two parts are combined to generate a spatial variational fuzzy kernel.
This reduces model complexity, improves training stability and the physical plausibility of the generated fuzzy kernel, and enhances the quality of image restoration results.
Smart Images
Figure CN121860889A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer vision and image processing technology, specifically to a parameterized generation method, apparatus, and storage medium for spatial variational fuzzy kernels. Background Technology
[0002] In the fields of computer vision and image processing, image restoration is a fundamental and important research area, with typical tasks including image super-resolution, deblurring, and denoising. The goal of image restoration is to recover a potentially sharp image from a degraded observation image. In most image restoration algorithms, the way the image degradation process is modeled directly affects the design and performance of the restoration algorithm.
[0003] In general, the image degradation process can be described by the following model:
[0004] Among them, clear images With a point spread function (PSF) or fuzzy kernel Convolution is performed, followed by downsampling and noise contamination, resulting in a low-quality image. Fuzzy kernel The accuracy of the estimation directly determines the success or failure of the restoration.
[0005] Traditional degradation models assume Maintaining spatial invariance across the entire image, i.e., modeling the entire image using a spatially invariant blur kernel, simplifies the computation process, allowing the degradation process to be viewed as a linear time-invariant system. This has driven the development of numerous image restoration algorithms based on the unified blur kernel assumption. However, in real-world imaging scenarios, this assumption often proves difficult to uphold. Due to the complexity of the shooting process and imaging environment, blur typically exhibits significant spatial variations, for example: 1. Motion blur: Different objects in a scene move at different speeds and in different directions.
[0006] 2. Defocus blur: Due to the limited depth of field, objects outside the imaging plane produce depth-related blur.
[0007] 3. Optical aberration: The non-ideal imaging characteristics of a lens at different field of view and wavelengths.
[0008] 4. Atmospheric turbulence: In remote sensing or astronomical observations, atmospheric disturbances cause non-uniform ambiguity.
[0009] To address spatial variation degradation, existing technologies typically face two extreme choices: one is to still use a simple spatially invariant kernel model, which results in poor restoration; the other is to attempt to directly estimate an independent, high-dimensional blur kernel for each pixel or each image patch, but this brings insurmountable computational and optimization challenges.
[0010] For the modeling problem of spatial variational fuzziness, existing technologies mainly offer the following types of solutions.
[0011] The first type of method is based on the spatial invariance assumption of local blocks. This involves dividing the image into multiple local regions and assuming that the blur approximation within each region remains invariant. Then, blur kernel estimation is performed for each local region separately. While this method alleviates the spatial variation problem to some extent, it is prone to discontinuities at region boundaries, leading to block artifacts in the restoration result. Furthermore, the limited effective information within small-scale local regions makes the blur kernel estimation process highly sensitive to noise, exhibiting poor stability, and is highly dependent on the choice of block size and overlap parameters.
[0012] The second type of method attempts to directly regress the spatial variational fuzzy kernel using a deep learning model, that is, to predict a local fuzzy kernel for each spatial location in the image. Since the local fuzzy kernel itself is a high-dimensional data structure, this type of method usually needs to predict a huge kernel tensor, resulting in high computational and storage overhead. At the same time, directly regressing high-dimensional kernel parameters can easily lead to instability in the optimization process, and the fuzzy kernel output by the network often lacks explicit constraints on physical properties such as nonnegativity and energy conservation, making it difficult to guarantee its physical rationality.
[0013] The third type of method simplifies the blurring process based on a parametric model, describing the degradation of the entire image using a small number of global parameters. This type of method has certain advantages in terms of computational complexity, but due to the limited number of parameters, it is usually difficult to characterize complex, spatially non-uniform blurring patterns, thus limiting its applicability.
[0014] In summary, existing technologies for modeling and generating spatial variational fuzzy kernels generally suffer from high computational complexity, insufficient modeling stability, or difficulty in simultaneously ensuring physical plausibility. Therefore, it remains necessary to propose a spatial variational fuzzy kernel generation scheme that can improve modeling efficiency while maintaining expressive power and physical reliability. Summary of the Invention
[0015] This invention provides a parameterized generation method, apparatus, and storage medium for spatial variational fuzzy kernels, which overcomes the aforementioned deficiencies in the prior art.
[0016] The purpose and effectiveness of the present invention—a parameterization generation method, apparatus, and storage medium for spatial variational fuzzy kernels—are achieved through the following specific technical means: A parameterized generation method for spatial variational fuzzy kernels includes: Acquire the input signal used to generate the spatial variational fuzzy kernel; The input signal is fed into a pre-trained hyperparameter generation network to generate a hyperparameter mapping map corresponding to the spatial resolution of the target image. The hyperparameter mapping map contains multiple hyperparameters at each spatial location to describe the morphology of the blur kernel. The hyperparameter mapping graph is input into the parsing decoder; For each spatial location in the hyperparameter mapping, the parser / decoder performs analytical calculations based on the hyperparameters of the corresponding location using a predefined parameterized fuzzy kernel model to generate the local fuzzy kernel corresponding to that location. The generated local fuzzy kernel is then normalized. The local fuzzy kernels generated at each spatial location are combined to obtain the final spatial variational fuzzy kernel.
[0017] In some embodiments of the method according to this disclosure, the input signal includes at least one of the following: a random noise vector, feature information extracted from a low-quality image to be processed, and conditional information related to the image restoration task.
[0018] In some embodiments of the method according to this disclosure, the hyperparameter mapping is a three-dimensional tensor whose spatial dimension corresponds to the spatial resolution of the target image and whose channel dimension corresponds to the number of hyperparameters at each spatial location.
[0019] In some embodiments of the method according to this disclosure, the hyperparameters are mapped by an activation function before the hyperparameter map is input to the parsing decoder, so as to limit the fuzz intensity parameter, scale parameter or orientation parameter to a preset physically reasonable range.
[0020] In some embodiments of the present disclosure, when the parser decoder generates a local fuzzy kernel for each spatial location, it includes: obtaining the hyperparameters corresponding to the spatial location; substituting the hyperparameters into a predefined parameterized fuzzy kernel model; and obtaining a two-dimensional discrete fuzzy kernel matrix through analytical calculation.
[0021] In some embodiments of the method according to this disclosure, the parameterized fuzzy kernel model is one of the following: anisotropic Gaussian fuzzy kernel model, mixture Gaussian fuzzy kernel model, or fuzzy kernel model based on polynomial Zernike coefficients.
[0022] In some embodiments of the present disclosure, the normalization process of the local fuzzy kernel includes: normalizing the elements in the local fuzzy kernel so that the sum of the elements is 1 or a preset value that satisfies the energy conservation constraint.
[0023] According to another aspect of this disclosure, a parameterization generation apparatus for a spatial variational fuzzy kernel is provided, comprising: The input module is used to acquire the input signal used to generate the spatial variational fuzzy kernel; The hyperparameter generation module is used to generate hyperparameter mapping graphs; The parsing and decoding module is used to generate local fuzzy kernels corresponding to each spatial location based on the parameterized fuzzy kernel model. The combination module is used to generate spatial variational fuzzy kernels.
[0024] According to another aspect of this disclosure, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described above.
[0025] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described above.
[0026] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a spatial variational fuzzy kernel generation framework based on hyperparameterized analytical decoding. By transforming the direct learning problem of high-dimensional fuzzy kernels into the learning of low-dimensional hyperparameter mappings and subsequent analytical computation, the parameter scale and learning difficulty of the model are fundamentally reduced. Compared with directly regressing high-dimensional fuzzy kernel tensors with spatial dimensions of H×W and kernel dimensions of k×k, this invention only needs to model a small number of hyperparameters at each spatial location, which significantly reduces the model complexity, helps alleviate overfitting problems, enables the model to be stably trained on smaller datasets, and maintains good generalization ability for unseen degradation patterns.
[0027] Furthermore, since the optimization object of this invention is a low-dimensional, continuously varying hyperparameter mapping graph, the optimization process is smoother and the gradient direction is more explicit compared to optimizing a high-dimensional, sparse fuzzy kernel tensor, thereby improving the stability of the training process and accelerating the convergence speed. Simultaneously, the parsing and decoding stage embeds a parameterized fuzzy kernel model with clear physical meaning, ensuring that the generated spatial variational fuzzy kernel naturally satisfies the physical requirements of nonnegativity, energy conservation, and smooth spatial transition.
[0028] The spatial variational blur kernel generation method proposed in this invention can be embedded as a general functional module into image restoration tasks such as image super-resolution and deblurring, providing more accurate degradation prior information for downstream networks. Under the same network structure and training conditions, introducing the spatial variational blur kernel generated by this invention helps improve the objective evaluation index performance of image restoration results. Furthermore, due to the lightweight hyperparameter generation network structure and the ability to perform parallel computation point-by-point in the parsing and decoding process, this invention significantly outperforms schemes that directly maintain and manipulate high-dimensional blur kernel tensors in terms of computational and storage overhead, making it suitable for engineering implementation. Attached Figure Description
[0029] Figure 1 This is a schematic diagram of the parameterization generation method for spatial variational fuzzy kernels according to an exemplary embodiment of the present disclosure; Figure 2 This is a schematic diagram of a hyperparameter generation network according to an exemplary embodiment of the present disclosure; Figure 3 This is a schematic diagram illustrating the generation of a local blur kernel by a decoder according to an exemplary embodiment of the present disclosure; Figure 4 This is a visualization of the spatial variational fuzzy kernel according to an exemplary embodiment of this disclosure; Figure 5 This is a block diagram of a parameterization generation apparatus for a spatial variational fuzzy kernel according to an exemplary embodiment of the present disclosure; Figure 6 This is a block diagram of a computer device according to an exemplary embodiment of the present disclosure. Detailed Implementation
[0030] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0031] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.
[0032] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.
[0033] like Figure 1 As shown, this embodiment provides a parameterized generation method for spatial variational fuzzy kernels, the overall process of which includes the following steps: S1. Obtain the input signal used to generate the spatial variational fuzzy kernel; S2. Based on the input signal, a hyperparameter mapping graph is generated through a hyperparameter generation network; S3. Input the hyperparameter mapping graph into the parsing decoder to generate local fuzzy kernels position by position; S4. Normalize the generated local fuzzy kernels and combine them to obtain the spatial variational fuzzy kernel.
[0034] Through the above steps, spatial variational fuzzy kernels can be generated efficiently and stably while maintaining physical rationality.
[0035] In step S1, the input signal used to generate the spatial variational fuzzy kernel is obtained.
[0036] In one embodiment, the input signal can be a fixed random noise vector used to generate a spatial variational fuzzy kernel without external constraints.
[0037] In another embodiment, the input signal may be deep feature information extracted from the low-quality image to be processed, such as the feature representation obtained by encoding the low-quality image through a convolutional neural network, thereby associating the generated spatial variational blur kernel with the specific image content.
[0038] In a further embodiment, the input signal may also include other conditional information related to the image restoration task, such as imaging device parameters, shooting scene information, or prior estimation results.
[0039] By introducing different forms of input signals, the method of the present invention has strong flexibility and can be adapted to different application scenarios.
[0040] In step S2, the input signal is input to the hyperparameter generation network to generate a hyperparameter mapping diagram.
[0041] like Figure 2As shown, the hyperparameter generation network is used to map the input signal into a hyperparameter map with the same spatial resolution as the target image. The hyperparameter map is a three-dimensional tensor, where the first two dimensions correspond to spatial locations, and the third dimension corresponds to the number of hyperparameters at each spatial location.
[0042] In a basic embodiment, each spatial location corresponds to three hyperparameters, which are used to describe key control parameters in the parameterized fuzzy kernel model, such as the principal axis scale parameter, secondary axis scale parameter, and rotation angle parameter of the fuzzy kernel. In a specific example, C=3, and the three channels represent the major axis variance of the anisotropic Gaussian kernel. minor axis variance and rotation angle θ.
[0043] Understandably, the hyperparameter generation network does not directly output the complete fuzzy kernel, but only outputs low-dimensional control parameters with clear physical meaning, thereby significantly reducing the learning difficulty and parameter scale of the model.
[0044] Preferably, step S201 is introduced after step S2 to perform physical constraint processing on the hyperparameters in the hyperparameter mapping graph.
[0045] In one embodiment, the physical constraint processing is implemented through an activation function. For example, the sigmoid function is used to limit the blur intensity-related parameters to a preset positive range, and the Tanh function or a linear mapping is used to limit the direction parameters to a reasonable angular range.
[0046] By applying the physical constraints described above, we can ensure that the hyperparameters are always within a physically meaningful range, thereby avoiding the generation of unreasonable or unrealizable fuzzy kernels.
[0047] It should be noted that the physical constraint processing does not change the spatial distribution relationship between hyperparameters, but only restricts their numerical range, and does not introduce additional computational complexity.
[0048] The hyperparameter mapping graph after physical constraint processing is input into the parsing decoder, which generates the corresponding local fuzzy kernel for each spatial location.
[0049] In step S3, the parser decoder is used to parse the hyperparameters in the hyperparameter map to generate the corresponding local fuzzy kernel.
[0050] In one embodiment, the parser decoder performs analytical calculations on the hyperparameters of each spatial location based on a predefined parameterized fuzzy kernel model, thereby generating a two-dimensional discrete local fuzzy kernel.
[0051] In a specific example, when the parameterized fuzzy kernel model is an anisotropic Gaussian model, the parsing decoder can construct the corresponding Gaussian function based on the hyperparameters corresponding to the spatial location. The parsing calculation method can be shown in the following formula, which is only used to illustrate one implementation of the parsing decoder:
[0052] in, It is a coordinate. After angle The result after rotation.
[0053]
[0054] In other specific examples, the parsing decoder can also generate a mixed blur kernel based on multiple sets of Gaussian parameters, or generate a blur kernel to describe optical aberrations based on Zernike polynomial coefficients.
[0055] The parser receives the corresponding hyperparameters at each spatial location and generates the local fuzzy kernel matrix corresponding to that location based on the parameterized model.
[0056] It is understood that the above anisotropic Gaussian model is only an exemplary implementation to illustrate how the parsing decoder generates local fuzzy kernels based on hyperparameters. This invention is not limited to this specific model form, and the parsing decoder can also use other parameterized fuzzy kernel models to perform parsing calculations on hyperparameters.
[0057] After generating the local fuzzy kernel, the local fuzzy kernel is normalized to ensure that the sum of its elements satisfies the energy conservation constraint.
[0058] like Figure 3 As shown, the parser decoder performs the following operations for each spatial location in the hyperparameter map: First, obtain the hyperparameters corresponding to this spatial location; Then, the hyperparameters are substituted into the predefined parameterized fuzzy kernel model, and a two-dimensional discrete local fuzzy kernel is generated through analytical calculation.
[0059] In a specific example, when an anisotropic Gaussian blur kernel model is used, the parsing decoder determines the scale and rotation relationship of the Gaussian function in different directions based on the hyperparameters, and calculates the blur kernel matrix at the corresponding position accordingly.
[0060] Understandably, the parsing and decoding process is a deterministic mathematical calculation process that does not rely on additional learnable parameters, thus ensuring the stability and interpretability of the decoding results.
[0061] In step S4, the generated local fuzzy kernels are normalized and combined to obtain the final spatial variational fuzzy kernel.
[0062] In one embodiment, the normalization process includes normalizing all elements in the local fuzzy kernel so that the sum of its elements satisfies a preset energy conservation constraint.
[0063] After normalization, the local fuzzy kernels generated at each spatial location are combined according to their spatial correspondence to form a complete spatial variational fuzzy kernel field.
[0064] The spatial variational blur kernel generated in the above manner changes continuously in space, which can effectively describe the blur characteristics that change with position in the real imaging process.
[0065] In a specific application scenario, the parameterized generation method of the spatial variational fuzzy kernel described in this invention is used to generate a corresponding spatial variational fuzzy kernel field for a test image.
[0066] As attached Figure 4 As shown, attached Figure 4 The image illustrates the distribution of local blur kernels generated at different spatial locations. Several representative spatial locations in the image are selected, and the local blur kernels generated at those locations are visualized in grayscale.
[0067] From the appendix Figure 4 It can be seen that the spatial variational fuzzy kernel generated by the method of the present invention exhibits continuous and smooth transition characteristics in space. The local fuzzy kernels at adjacent spatial locations have good continuity in shape and direction, and no obvious abrupt changes or discontinuities are observed.
[0068] Meanwhile, the local blur kernels generated at each spatial location all exhibit reasonable Gaussian or near-Gaussian shapes, consistent with the physical blur characteristics commonly found in actual imaging processes. This indicates that, by combining hyperparameter mapping and analytical decoding, this invention can ensure the ability to express spatial variations while making the generated blur kernels naturally meet the requirements of physical rationality.
[0069] It should be noted that the appendix Figure 4 The kernel field distribution shown is merely a visualization result of the method of the present invention in a specific example, used to illustrate one effect of the method of the present invention in generating spatial variational fuzzy kernels, and does not constitute a limitation on the scope of protection of the present invention.
[0070] In a specific application scenario, the parameterized generation method of the spatial variational fuzzy kernel described above is applied to the image super-resolution task to provide degradation modeling priors for the image restoration network.
[0071] In this application scenario, the Set5 dataset was used as the test dataset, and the same set of low-quality images were processed using both a spatially invariant blur kernel and a spatial variational blur kernel generated based on the method of this invention. The two methods were compared and tested under the same network structure and training parameters.
[0072] method PSNR SSIM Space-invariant kernel 23.17 0.6889 Spatial variational kernel 29.18 0.8387 The test results are shown in the table above. When using the spatially invariant fuzzy kernel, the peak signal-to-noise ratio (PSNR) is 23.17 dB and the structural similarity index (SSIM) is 0.6889. When using the spatial variational fuzzy kernel generated by the method of this invention, the PSNR is improved to 29.18 dB and the SSIM is improved to 0.8387.
[0073] The comparison results above show that, under the same test conditions, the spatial variational blur kernel generated by the method of this invention can more accurately characterize the blur characteristics that change with spatial position during image degradation, thereby significantly improving the quality of image restoration results.
[0074] It should be noted that the above experimental data are only used to illustrate the effect of the method of the present invention in a specific application scenario, and do not constitute a limitation on the scope of protection of the present invention.
[0075] The parameterization generation method for spatial variational fuzzy kernels provided in this embodiment can be executed in smart terminals, computer terminals, network devices, chips, chip modules, or similar computing devices.
[0076] Corresponding to the parameterization generation method for spatial variational fuzzy kernels described above, this embodiment also provides a parameterization generation device for spatial variational fuzzy kernels. These will be described in detail below. Specifically, as... Figure 5 As shown, the parameterization generation device for the spatial variational fuzzy kernel in this embodiment includes: Input module 1 is used to acquire the input signal used to generate the spatial variational fuzzy kernel; Hyperparameter generation module 2 is used to generate hyperparameter mapping graphs; The parsing and decoding module 3 is used to generate local fuzzy kernels corresponding to each spatial location based on the parameterized fuzzy kernel model; Combination module 4 is used to generate spatial variational fuzzy kernels.
[0077] It should be noted that the embodiment of the parameterization generation device for spatial variational fuzzy kernels provided in this application is based on the same inventive concept as the above-described embodiment of the parameterization generation method for spatial variational fuzzy kernels.
[0078] It should be noted that the parameterization generation device for the spatial variational fuzzy kernel in this embodiment can be, for example, a separate chip, a chip module, or an electronic device, or it can be a chip or a chip module integrated into an electronic device. Regarding the modules / units included in the various devices and products described in the above embodiments, they can be software modules / units, hardware modules / units, or a combination of both. For example, for various devices and products applied to or integrated into a chip, all of their included modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs running on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits; for various devices and products applied to or integrated into a chip module, all of their included modules / units can be implemented using hardware methods such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The components can be implemented using software programs that run on the processor integrated within the chip module. The remaining (if any) modules / units can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into the terminal, each of its components / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or in different components within the terminal. Alternatively, at least some modules / units can be implemented using software programs that run on the processor integrated within the terminal, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits.
[0079] This invention provides a computer device. The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a parameterized generation method for spatial variational fuzzy kernels. Figure 6 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.
[0080] like Figure 6 As shown, the computer device can be in the form of a general-purpose computing device, such as a server device. The components of the computer device may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including memory 32 and processor 31).
[0081] Bus 33 includes a data bus, an address bus, and a control bus.
[0082] The memory 32 may include volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.
[0083] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.
[0084] The processor 31 executes various functional applications and data processing by running computer programs stored in the memory 32, such as the parameterization generation method of the spatial variational fuzzy kernel in this embodiment of the invention.
[0085] The computer device can also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed through the input / output (I / O) interface 35. Furthermore, the model-generating device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via a network adapter 36. Figure 6 As shown, network adapter 36 communicates with other modules of the model-generated device 30 via bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0086] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0087] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the parameterization generation method of spatial variational fuzzy kernel.
[0088] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0089] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to execute steps in the parameterized generation method for implementing the spatial variational fuzzy kernel.
[0090] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.
[0091] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved. This embodiment does not impose any limitations on these steps.
[0092] It should be understood that the terms "system," "device," "unit," and / or "module" used in this embodiment are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they can be replaced by other expressions.
[0093] This embodiment uses a flowchart to illustrate the operations performed by the system according to this embodiment. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, each step can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.
[0094] As shown in this embodiment, unless the context explicitly indicates an exception, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate that explicitly identified steps and elements are included, and these steps and elements do not constitute an exclusive list; the method or apparatus may also include other steps or elements.
[0095] The definitions used in this embodiment, such as the terms "having," "may have," "comprising," or "may include," indicate the presence of corresponding functions, operations, elements, etc., in this embodiment, but do not limit the presence of one or more other functions, operations, elements, etc. Furthermore, it should be understood that the terms "comprising" or "having," as used herein, indicate the presence of features, numbers, steps, operations, elements, components, or combinations thereof described in the specification, without excluding the presence or addition of one or more other features, numbers, steps, operations, elements, components, or combinations thereof.
[0096] In this embodiment, the definitions of "A or B", "at least one of A and / or B" or "one or more of A and / or B" as used herein include any and all combinations of the words listed therewith. For example, "A or B", "at least one of A and / or B" or "one or more of A and / or B" means (1) including at least one A, (2) including at least one B, or (3) including both at least one A and at least one B.
[0097] The definitions of "first" and "second" in this embodiment, and the descriptions of "first" and "second" appearing in this embodiment, are only for illustration and to distinguish the objects being described. They do not indicate any order and do not represent a special limitation on the number of devices in this embodiment, nor do they constitute any limitation on this embodiment. For example, the first element can be referred to as the second element without departing from the scope of this disclosure, and similarly, the second element can be referred to as the first element.
[0098] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.
Claims
1. A parameterized generation method for spatial variational fuzzy kernels, characterized in that, include: Acquire the input signal used to generate the spatial variational fuzzy kernel; The input signal is fed into a pre-trained hyperparameter generation network to generate a hyperparameter mapping map corresponding to the spatial resolution of the target image. The hyperparameter mapping map contains multiple hyperparameters at each spatial location to describe the morphology of the blur kernel. The hyperparameter mapping graph is input into the parsing decoder; For each spatial location in the hyperparameter mapping, the parser / decoder performs analytical calculations based on the hyperparameters of the corresponding location using a predefined parameterized fuzzy kernel model to generate the local fuzzy kernel corresponding to that location. The generated local fuzzy kernel is then normalized. The local fuzzy kernels generated at each spatial location are combined to obtain the final spatial variational fuzzy kernel.
2. The method as described in claim 1, characterized in that, The input signal includes at least one of the following: a random noise vector, feature information extracted from the low-quality image to be processed, and conditional information related to the image restoration task.
3. The method as described in claim 1, characterized in that, The hyperparameter mapping is a three-dimensional tensor, whose spatial dimension corresponds to the spatial resolution of the target image, and whose channel dimension corresponds to the number of hyperparameters at each spatial location.
4. The method as described in claim 1, characterized in that, Before inputting the hyperparameter mapping graph into the parsing decoder, the hyperparameters are mapped by an activation function to limit the blur intensity parameter, scale parameter, or orientation parameter to a preset physical reasonable range.
5. The method as described in claim 1, characterized in that, When the parser / decoder generates a local blur kernel for each spatial location, it includes: Obtain the hyperparameters corresponding to this spatial location; The hyperparameters are substituted into a predefined parameterized fuzzy kernel model, and the two-dimensional discrete fuzzy kernel matrix is obtained through analytical calculation.
6. The method as described in claim 5, characterized in that, The parameterized fuzzy kernel model is one of the following: anisotropic Gaussian fuzzy kernel model, mixed Gaussian fuzzy kernel model, or fuzzy kernel model based on polynomial Zernike coefficients.
7. The method as described in claim 1, characterized in that, The normalization process for the local fuzzy kernel includes: normalizing the elements in the local fuzzy kernel so that the sum of its elements is 1 or a preset value that satisfies the energy conservation constraint.
8. A parameterization generation device for spatial variational fuzzy kernels, characterized in that, include: The input module is used to acquire the input signal used to generate the spatial variational fuzzy kernel; The hyperparameter generation module is used to generate hyperparameter mapping graphs; The parsing and decoding module is used to generate local fuzzy kernels corresponding to each spatial location based on the parameterized fuzzy kernel model. The combination module is used to generate spatial variational fuzzy kernels.
9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.