Infrared image denoising method and system based on NSST decomposition

Abstract

The invention discloses an infrared image denoising method and system based on NSST decomposition, and the method comprises the steps: S1, decomposing a transformer infrared image containing noise based on an NSST algorithm, and outputting a high-frequency sub-band coefficient, a low-frequency sub-band coefficient and a high-frequency sub-band coefficient; s2, based on the low-frequency sub-band in the S1, filtering and denoising the low-frequency sub-band by adopting an improved non-local mean algorithm, and outputting the denoised low-frequency sub-band; s3, based on the high-frequency sub-band coefficients in the S1, introducing local entropies for different high-frequency sub-band coefficients, and outputting each high-frequency sub-band after the local entropies are denoised; and S4, based on the low-frequency sub-band in S2 and each high-frequency sub-band in S3, obtaining a denoised image by adopting NSST inverse transformation, and completing infrared image denoising. According to the method, the high-frequency sub-band coefficients containing different frequency noise are denoised by adopting the local entropy threshold method, noise filtering parameters do not need to be manually set, denoising threshold parameters are established according to entropy information of the high-frequency sub-band coefficients, and the method has good adaptability.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method and system for denoising infrared images based on NSST decomposition. Background Technology

[0002] Infrared thermal imagers rely on thermal radiation imaging, but thermal radiation signals are typically weak. When inspecting transformers, they are easily affected by the complex surrounding electromagnetic environment and background, resulting in infrared images of transformers having higher noise levels, more clutter, and more interference points compared to visible light images. The main challenge in denoising transformer infrared images lies in the fact that the edges and texture details of the image exhibit characteristics similar to noise, making them difficult to distinguish and difficult to preserve the image's texture details while denoising.

[0003] Common noise reduction methods mainly include spatial domain noise reduction and transform domain noise reduction. Spatial domain noise reduction refers to processing image pixels within the pixel space, such as stretching. Although the structure of these algorithms is easy to understand, they cannot simultaneously reduce noise and preserve texture details, making it difficult to maintain image quality well. Transform domain noise reduction decomposes the image domain to extract more information before further noise reduction processing. The most common transform domain processing is wavelet transform, whose biggest advantage is that it can decompose information in different directions of the image. However, its directionality is relatively singular, lacking anisotropic expression.

[0004] Prior art 1 discloses an infrared image enhancement method for power equipment in the NSST domain based on an improved BEEPS filtering algorithm (CN113837974B). First, the infrared image is decomposed into high-frequency and low-frequency components using NSST transform. For the low-frequency component containing a large amount of target equipment information, in addition to BEEPS processing, the edges of the electrical equipment and the background are extracted and enhanced separately. For the high-frequency component, a multi-scale Retinex algorithm is used to denoise and enhance each high-frequency subband of each directional coefficient. Then, the enhanced infrared image is reconstructed from the enhanced low-frequency and high-frequency components. However, prior art 1 has the following drawbacks: the Gaussian filtering kernel function in the multi-scale Retinex algorithm is prone to visual artifacts during image denoising, and the edge preservation effect is poor.

[0005] Prior art 2 discloses an adaptive denoising method for NSCT domain infrared images based on quantum differential evolution algorithm (CN110298808B). First, the noisy infrared image is decomposed using NSCT to obtain a low-frequency sub-band image and a multi-scale high-frequency sub-band. Then, quantum coding and quantum gate updates are introduced into the differential evolution algorithm for optimization to prevent getting trapped in local optima. Next, for the multi-scale high-frequency sub-band, the quantum differential evolution algorithm is used to adaptively obtain the optimal threshold, and the high-frequency sub-band coefficients are denoised using a threshold processing function. Finally, the processed sub-band coefficients are inversely transformed to obtain the denoised infrared image. However, prior art 2 has the following drawback: the threshold function used when processing the high-frequency sub-band requires manually specified parameters and cannot calculate the denoising threshold based on the overall image features, thus lacking adaptability. Summary of the Invention

[0006] To address the shortcomings of current mainstream infrared image processing algorithms, this paper proposes an algorithm based on Nonsubsampled Shearlet Transform (NSST) combined with improved Non-Local Means (NLM) filtering. First, the NSST algorithm decomposes the transformer infrared image into different frequency bands. Compared to wavelet algorithms, it can resolve more sub-bands and achieve optimized noise reduction in multi-resolution images. Then, for the decomposed low-frequency image portion, an improved NLM filtering algorithm is used to remove noise while preserving image edge texture. For noise in the different high-frequency sub-bands, a local entropy thresholding method is used for filtering. The most significant advantage of this method is that it does not require manually set threshold parameters, providing a new approach to adaptive threshold setting in filtering algorithms.

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for denoising infrared images based on NSST decomposition.

[0008] The present invention adopts the following technical solution.

[0009] The first aspect of this invention discloses a method for denoising infrared images based on NSST decomposition, comprising:

[0010] S1 decomposes the noisy transformer infrared image based on the NSST algorithm and outputs the high-frequency subband, low-frequency subband, and high-frequency subband coefficients.

[0011] S2, based on the low-frequency sub-band of S1, uses an improved nonlocal mean algorithm to filter and denoise the low-frequency sub-band, and outputs the denoised low-frequency sub-band.

[0012] S3, based on the high-frequency sub-band coefficients in S1, introduces local entropy values ​​for different high-frequency sub-band coefficients, and outputs each high-frequency sub-band after local entropy value denoising;

[0013] S4, based on the low-frequency subband of S2 and each high-frequency subband of S3, uses NSST inverse transform to obtain the denoised image, thus completing the infrared image denoising.

[0014] Preferably, the NSST algorithm decomposition process in S1 includes:

[0015] S11 performs NSP decomposition on the transformer infrared image and outputs high-frequency subband and low-frequency subband;

[0016] S12 performs directional localization on the high-frequency subband layer decomposed from S11, and outputs multi-directional high-frequency subband coefficients.

[0017] Preferably, in S2, the improved nonlocal mean algorithm is as follows:

[0018]

[0019] In the formula, v(j) represents the noise-distorted image; This is the filtered image; Ω represents the neighborhood search region for each pixel by weighting the neighborhood boxes; f(d(i,j)) is the improved weight function.

[0020] Preferably, the improved weighting function is:

[0021]

[0022] In the formula, d(i,j) is the Euclidean weighted distance; h is a parameter greater than 0 that controls the decay of the exponential function, thereby changing the weight of the Euclidean distance.

[0023] Preferably, the local entropy threshold is:

[0024]

[0025] Where TH represents the threshold of local entropy, Representing the cth i The local entropy matrix of the coefficients of each high-frequency subband, where mean(.) represents the matrix mean; c TH A parameter that reflects the overall noise level of an image.

[0026] Preferably,

[0027]

[0028] in, Let be the local entropy matrix of the original image.

[0029] Preferably, in S4, the coefficients of the low-frequency subband after denoising by the improved nonlocal mean algorithm and the coefficients of each high-frequency subband after local entropy denoising are subjected to NSST inverse transform to obtain the denoised infrared image.

[0030] A second aspect of the present invention provides a denoising system employing the above-described method, comprising:

[0031] NSST algorithm decomposition module, low-frequency subband filtering and denoising module, high-frequency subband denoising module, and NSST inverse transform module;

[0032] The NSST algorithm decomposition module is used to decompose the noisy transformer infrared image based on the NSST algorithm, and output the high-frequency subband, low-frequency subband and high-frequency subband coefficients.

[0033] The low-frequency subband filtering and denoising module is used to filter and denoise the low-frequency subband using an improved nonlocal mean algorithm, and outputs the denoised low-frequency subband.

[0034] The high-frequency subband denoising module is used to introduce local entropy values ​​for different high-frequency subband coefficients and output each high-frequency subband after local entropy denoising.

[0035] The NSST inverse transform module is used to obtain the denoised image by performing the NSST inverse transform, thus completing the infrared image denoising.

[0036] A third aspect of the present invention provides a terminal, including a processor and a storage medium;

[0037] The storage medium is used to store instructions;

[0038] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0039] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon.

[0040] When the program is executed by the processor, it implements the steps of the above method.

[0041] The beneficial effects of this invention are that, compared with the prior art,

[0042] This invention uses the NSST algorithm to decompose infrared images into low-frequency sub-bands containing image contours and basic features, and a series of high-frequency sub-bands containing noise of different frequencies. By performing different processing on the low-frequency and high-frequency sub-bands, noise removal can be achieved while preserving image edge texture information.

[0043] This invention employs a local entropy thresholding method to denoise high-frequency subband coefficients containing noise of different frequencies. This method does not require manually setting noise filtering parameters, but instead establishes denoising threshold parameters based on the entropy information of the high-frequency subband coefficients themselves, thus exhibiting good adaptability.

[0044] This invention employs an improved nonlocal mean filtering algorithm to process the decomposed low-frequency sub-bands. Compared to the original nonlocal mean filtering algorithm, an improved weight function is introduced to avoid the shortcomings of the original exponential function in terms of insufficient accuracy in similarity measurement and unreasonable weight allocation, which can easily lead to blurred images after denoising. The improved weight function can significantly preserve image details while removing noise from the low-frequency sub-bands. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the infrared image denoising method based on NSST decomposition according to an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of non-subsampled shear wave transformation according to an embodiment of the present invention;

[0047] Figure 3 This is a comparison diagram of the Euclidean distance between the improved weight function and the original weight function in this embodiment of the invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0049] like Figure 1-3 As shown, this embodiment of the invention provides a method for denoising infrared images based on NSST decomposition, specifically including the following steps:

[0050] S1 decomposes the noisy transformer infrared image into multiple scales using the non-subsampled shearlet transform (NSST) algorithm, outputting the coefficients of the high-frequency subband, low-frequency subband, and high-frequency subband.

[0051] This paper describes a multi-scale decomposition of noisy transformer infrared images using the NSST algorithm. While maintaining low computational complexity, this method increases the number of constructed directions, enabling more accurate representation of multi-dimensional data such as curves and edges in the image. It also avoids the pseudo-Gibbs phenomenon observed in other multi-scale decomposition methods. Specific operations are as follows: Figure 2 As shown:

[0052] The main steps involved in the NSST algorithm decomposition are as follows:

[0053] S11 performs non-subsampled pyramid (NSP) decomposition on the transformer infrared image, outputting high-frequency subband and low-frequency subband.

[0054] A non-downsampled pyramid filter can resolve a transformer infrared image into high-frequency and low-frequency sub-bands. The outline and basic features of the transformer infrared image are mainly found in the low-frequency sub-band image, while the edges, noise, and details are mainly found in the high-frequency sub-band image.

[0055] Figure 2 The NSP decomposition in the model decomposes the original image f into one low-frequency sub-band image g and k high-frequency sub-band images l. i (i = 1, 2, 3, ... k), and the size of each frequency sub-band image is the same as the original image. In the process of multi-scale and multi-directional decomposition of the image, the number of layers has a great influence on the noise reduction effect of the image.

[0056] If the number of layers is too small, it will be impossible to accurately separate high-frequency and low-frequency information in the image. The low-frequency part will contain a large amount of information that should belong to the high-frequency part, reducing the noise reduction effect. If the number of layers is too large, although more detailed information can be obtained, the determination of the threshold function will become very complicated, which is also not conducive to the implementation of image noise reduction.

[0057] Therefore, in a preferred but non-limiting embodiment of this patent, the number of layers k is set to 4.

[0058] S12 performs directional localization on the high-frequency subband layers of the transformer infrared images of different levels decomposed by S11, and outputs multi-directional high-frequency subband coefficients.

[0059] In a preferred but non-limiting embodiment of the present invention, directional localization is mainly the process of obtaining the high-frequency subband coefficients by shearing the filter window.

[0060] First, a shearing filter window W is generated on the pseudopolar grid system using the Meyer wavelet. Then, to avoid downsampling, the window W is transformed to a Cartesian coordinate system, forming a new window function W0. new Finally, through W new and l i The multidirectional high-frequency subband coefficient C is obtained by convolution calculation between (i = 1, 2, 3, ... k). i .

[0061] In a preferred but non-limiting embodiment of the invention, the high-frequency subband coefficients are c1, c2, c3 and c4.

[0062] S2, based on the low-frequency sub-band of the transformer infrared image output by S1, uses an improved non-local means (NLM) algorithm to filter and denoise the low-frequency sub-band, and outputs the denoised low-frequency sub-band.

[0063] It is generally believed that the low-frequency subband mainly contains the contour information of the image, and traditional transform domain denoising methods ignore the processing of the low-frequency subband. However, research has found that the low-frequency subband in the transform domain also contains noise and edge information. In order to ensure that more image details and textures are preserved while removing noise from the low-frequency subband, this invention designs an improved nonlocal mean filtering algorithm to process the noise in the low-frequency subband.

[0064] The Non-Local Means (NLM) algorithm is as follows:

[0065]

[0066] In the formula: v(j) represents the noisy, distorted image; This is an estimate of the original image u, which is the filtered image; Ω represents the neighborhood search region for each pixel using weighted neighborhood boxes; W(i,j) represents the weight coefficient between the current pixel's neighborhood N(j) and its contrasting block N(i) within the Ω range. This weight coefficient is typically measured using Gaussian-weighted Euclidean distance.

[0067]

[0068] Where: distance i，j denoted as ...

[0069] Nonlocal means algorithms are highly effective in image denoising, but there are still areas for improvement. The classic nonlocal means algorithm uses an exponential weight function, which lacks accuracy in similarity measurement and results in unreasonable weight allocation. This function assigns high weights to regions of high similarity, but the rapid decay and unreasonable weight distribution can easily lead to blurring of the denoised image. An ideal weight function should assign larger weights with slow decay in regions of high similarity, accelerate decay as it transitions from high to low similarity, and assign smaller weights, even approaching zero, in regions of low similarity. To make the trend of the weight function approximately match that of the ideal weight function, the improved weight function proposed in this invention is shown in the following equation.

[0070]

[0071] In the formula, d(i,j) is the Euclidean weighted distance; f(d(i,j)) is the improved weight function; h is a parameter greater than 0 that controls the decay of the exponential function and thus changes the weight of the Euclidean distance.

[0072] In a preferred but non-limiting embodiment of the invention, the value of h is 10.

[0073] Compared to the original weight function, the improved weight function makes the similarity measurement in the filtering process more accurate, avoiding the shortcomings of the original weight function, such as insufficient accuracy in similarity measurement and unreasonable weight allocation, which resulted in blurred images after processing. It also retains more image details while removing noise.

[0074] Therefore, the improved Non-Local Means (NLM) algorithm is as follows:

[0075]

[0076] In the formula: v(j) represents the noisy, distorted image; It is the estimated value of the original image u, which is the filtered image; Ω represents the neighborhood search area for each pixel by weighting the neighborhood boxes.

[0077] like Figure 3 As shown, when the Euclidean weighted distance between two image blocks is less than the set smoothing threshold, the two image blocks are considered similar, and the improved weighting function provided in this embodiment assigns them higher weights; when the Euclidean weighted distance between two image blocks is greater than the set smoothing threshold, the two image blocks are considered dissimilar, and the weighting function decreases rapidly, which can reduce the influence of dissimilar image blocks on pixels.

[0078] S3, based on the high-frequency sub-band coefficients output in S1, introduces local entropy values ​​for different high-frequency sub-band coefficients, and outputs each high-frequency sub-band after denoising based on the local entropy values.

[0079] The high-frequency subband coefficients c1, c2, c3, and c4 obtained from NSST decomposition contain different information. c1, c2, and c3 contain some noise and a large amount of high-frequency directional information, while c4 contains a large amount of noise information. Therefore, different filtering thresholds need to be established for processing. Traditional filtering methods based on NSST decomposition often use manually set filtering thresholds when processing low-frequency subband coefficients, which lacks adaptability. This embodiment of the invention introduces entropy values ​​to establish filtering thresholds for different high-frequency subband coefficients, thus avoiding the defects of manually setting filtering thresholds.

[0080] In thermodynamics, entropy is directly proportional to the degree of disorder in a system. Applying the concept of entropy to images, it represents the amount of information in the image. Information theory dictates that image entropy is proportional to the amount of information, the change in grayscale, and the amount of edge information. The formula for calculating entropy is as follows:

[0081]

[0082] Where H f Let p be the entropy value. i,j f(i,j) represents the probability of the gray value of the corresponding pixel (i,j) in the image. The image size is m×n, and f(i,j) refers to the gray value of the pixel (i,j) in the image.

[0083] By incorporating the image's entropy value into the calculation of the threshold for distinguishing between noise and non-noise, the threshold for distinguishing between noise and non-noise is calculated using the image's entropy value, thus avoiding the drawbacks of manually setting the threshold.

[0084] When solving for the entropy value of a certain region within an image, it is called local entropy. In this embodiment of the invention, the local entropy thresholding method is used to process the high-frequency subband coefficients for denoising. Its calculation process is similar to the above two formulas. The only difference is that the range of coordinates (i,j) should be within the region being solved.

[0085] In a preferred but non-limiting embodiment of this invention, the window size for calculating local entropy is set to 9×9, i.e., m=n=9. By sliding the window, the local entropy of each pixel in the image can be calculated, forming the local entropy matrix H of the image. f .

[0086] For c1, c2, and c3, relatively low thresholds are used to preserve high-frequency directional information above the threshold as much as possible; while for c4, a relatively high threshold is used to separate noise as much as possible. Therefore, the following thresholds based on local entropy are derived:

[0087]

[0088] Where TH represents the threshold of local entropy, Representing the cth i The local entropy matrix of the coefficients of each high-frequency subband, where mean(.) represents the matrix mean; c TH The parameter reflecting the overall noise level of the image is obtained through the local entropy matrix of the original image. The local entropy matrix reflects the overall noise level of the image. Higher overall image noise results in a higher local entropy matrix. The larger the mean, the higher the threshold needs to be.

[0089] The first formula for the threshold TH is applied to the high-frequency subband coefficients c1, c2, and c3, and the second formula is applied to the high-frequency subband coefficient c4. As the number of decomposition layers increases, the grayscale information of the high-frequency subbands c1, c2, and c3 becomes increasingly rich, and the average value of the overall local entropy continuously increases. Therefore, the filtering thresholds of c1, c2 and c3 will increase continuously as a whole, and the threshold of c4 will be much higher than that of other high-frequency sub-bands. In each sub-image, values ​​less than TH are eliminated as noise. The biggest feature of the threshold method in this embodiment of the invention is that there are no parameters that need to be set manually, and the threshold of the filtering algorithm is adaptive.

[0090] S4, based on the denoised low-frequency sub-bands output by S2 and the denoised high-frequency sub-bands output by S3, uses NSST inverse transform to obtain the denoised image, thus completing the infrared image denoising.

[0091] The coefficients of the low-frequency subband after denoising by the improved nonlocal mean algorithm and the coefficients of each high-frequency subband after local entropy denoising are subjected to NSST inverse transform to finally obtain the denoised transformer infrared image.

[0092] To demonstrate the effectiveness of this denoising method, an experiment was designed to compare the present invention, the traditional NLM filtering algorithm, and wavelet threshold-based denoising methods. The peak signal-to-noise ratio (PSNR) and mean structural similarity (MSSIM) were used to evaluate the algorithm performance. The experimental data consisted of 50 transformer infrared images randomly fitted with Gaussian white noise with standard deviations of 10, 25, and 50. The experimental results are shown in the table below.

[0093] Table 1. Comparison of the present invention, traditional NLM, and wavelet thresholding denoising methods.

[0094]

[0095]

[0096] The experimental results show that the algorithm of this invention outperforms traditional NLM algorithm and wavelet thresholding denoising algorithm in both PNSR and MSSIM metrics.

[0097] This invention also provides a denoising system employing the above-described NSST-based infrared image decomposition denoising method, comprising:

[0098] NSST algorithm decomposition module, low-frequency subband filtering and denoising module, high-frequency subband denoising module, and NSST inverse transform module;

[0099] The NSST algorithm decomposition module is used to decompose the noisy transformer infrared image based on the NSST algorithm, and output the high-frequency subband, low-frequency subband and high-frequency subband coefficients.

[0100] The low-frequency subband filtering and denoising module is used to filter and denoise the low-frequency subband using an improved nonlocal mean algorithm, and outputs the denoised low-frequency subband.

[0101] The high-frequency subband denoising module is used to introduce local entropy values ​​for different high-frequency subband coefficients and output each high-frequency subband after local entropy denoising.

[0102] The NSST inverse transform module is used to obtain the denoised image by performing the NSST inverse transform, thus completing the infrared image denoising.

[0103] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0104] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0105] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0106] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for denoising infrared images based on NSST decomposition, characterized in that, include: S1 decomposes the noisy transformer infrared image based on the NSST algorithm and outputs the high-frequency subband, low-frequency subband, and high-frequency subband coefficients. S2, based on the low-frequency sub-band of S1, uses an improved nonlocal mean algorithm to filter and denoise the low-frequency sub-band, and outputs the denoised low-frequency sub-band. S3, based on the high-frequency sub-band coefficients in S1, introduces local entropy values ​​for different high-frequency sub-band coefficients, and outputs each high-frequency sub-band after local entropy value denoising; S4, based on the low-frequency subband of S2 and each high-frequency subband of S3, uses NSST inverse transform to obtain the denoised image, thus completing the infrared image denoising.

2. The infrared image denoising method based on NSST decomposition according to claim 1, characterized in that: The NSST algorithm decomposition process in S1 includes: S11 performs NSP decomposition on the transformer infrared image and outputs high-frequency subband and low-frequency subband; S12 performs directional localization on the high-frequency subband layer decomposed from S11, and outputs multi-directional high-frequency subband coefficients.

3. The infrared image denoising method based on NSST decomposition according to claim 1, characterized in that: In S2, the improved nonlocal mean algorithm is as follows: In the formula, v(j) represents the noise-distorted image; This is the filtered image; Ω represents the neighborhood search region for each pixel by weighting the neighborhood boxes; f(d(i,j)) is the improved weight function.

4. The infrared image denoising method based on NSST decomposition according to claim 3, characterized in that: The improved weighting function is as follows: In the formula, d(i,j) is the Euclidean weighted distance; h is a parameter greater than 0 that controls the decay of the exponential function, thereby changing the weight of the Euclidean distance.

5. The infrared image denoising method based on NSST decomposition according to claim 1, characterized in that: The local entropy threshold is: Where TH represents the threshold of local entropy, Representing the cth i The local entropy matrix of the coefficients of each high-frequency subband, where mean(.) represents the matrix mean; c TH A parameter that reflects the overall noise level of an image.

6. The infrared image denoising method based on NSST decomposition according to claim 5, characterized in that: in, Let be the local entropy matrix of the original image.

7. The infrared image denoising method based on NSST decomposition according to claim 1, characterized in that: In S4, the coefficients of the low-frequency subbands after denoising by the improved nonlocal mean algorithm and the coefficients of each high-frequency subband after local entropy denoising are subjected to NSST inverse transform to obtain the denoised infrared image.

8. A denoising system employing the NSST decomposition infrared image denoising method according to any one of claims 1-7, characterized in that, include: The module includes an NSST algorithm decomposition module, a low-frequency subband filtering and denoising module, a high-frequency subband denoising module, and an NSST inverse transform module. The NSST algorithm decomposition module is used to decompose the noisy transformer infrared image based on the NSST algorithm, and output the high-frequency subband, low-frequency subband and high-frequency subband coefficients. The low-frequency subband filtering and denoising module is used to filter and denoise the low-frequency subband using an improved nonlocal mean algorithm, and outputs the denoised low-frequency subband. The high-frequency subband denoising module is used to introduce local entropy values ​​for different high-frequency subband coefficients and output each high-frequency subband after local entropy denoising. The NSST inverse transform module is used to obtain the denoised image by performing the NSST inverse transform, thus completing the infrared image denoising.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images