A turbulence removal method and system based on image restoration

Abstract

The present invention discloses a de-turbulence method and system based on image restoration, the method comprising: S1: decomposing an image affected by turbulence using a non-subsampled wavelet transform and extracting features of the decomposed image; S2: performing turbulence detection based on adaptive threshold segmentation to obtain a turbulence region in the image affected by turbulence; S3: calculating and mapping the grayscale features of the image affected by turbulence, performing feature enhancement on the turbulence region obtained in step S2, and obtaining enhanced features of the turbulence region; S4: calculating adaptive weights and local similarity metrics based on the enhanced features of the turbulence region obtained in step S3, and using them to restore the image affected by turbulence to obtain a restored image; S5: combining the turbulence region in the restored image with the non-turbulence region in the original image to obtain a final de-turbulenced image. The present invention can accurately detect and restore image regions affected by turbulence to restore the clarity and quality of the image.

Description

Technical Field

[0001] The present invention relates to the technical field of turbulence removal, and in particular to a turbulence removal method and system based on image restoration. Background Art

[0002] In the field of image processing, de-turbulence technology is crucial, which aims to improve the image degradation problem caused by atmospheric turbulence. Atmospheric turbulence can cause dynamic blur and distortion in images, which is a severe challenge for imaging systems that require high clarity and stability. Traditional de-turbulence methods mainly rely on frequency domain analysis, such as Fourier transform, and time domain analysis, such as wavelet transform. Although these methods are effective in some cases, they have some inherent limitations. For example, Fourier transform is sensitive to noise and has difficulty in handling nonlinear and anisotropic turbulence effects; while wavelet transform can provide multi-scale information, it may lose important details during downsampling and is not flexible enough to respond to local features. In addition, existing technologies often lack pertinence and adaptability in feature extraction and image restoration, resulting in limited performance in complex environments. In particular, in terms of accurate detection and feature enhancement of turbulent areas, existing methods often find it difficult to achieve satisfactory results, and the restored image may not be effectively integrated into the original image, resulting in visual unnaturalness. Summary of the invention

[0003] In view of this, the present invention provides a de-turbulence method based on image restoration, which can accurately detect and restore image areas affected by turbulence to restore the clarity and quality of the image.

[0004] To achieve the above object, the present invention provides a turbulence removal method based on image restoration, comprising the following steps:

[0005] S1: Decompose the image affected by turbulence using non-subsampled wavelet transform and extract the features of the decomposed image;

[0006] S2: turbulence detection based on adaptive threshold segmentation to obtain the turbulence area in the image affected by turbulence;

[0007] S3: Calculate the grayscale features of the image affected by turbulence and perform mapping, perform feature enhancement on the turbulence area obtained in step S2, and obtain enhanced features of the turbulence area;

[0008] S4: Based on the turbulence region enhancement features obtained in step S3, an adaptive weight and a local similarity metric are calculated and used to repair the image affected by turbulence to obtain a repaired image;

[0009] S5: combining the turbulent region in the restored image with the non-turbulent region in the original image to obtain the final de-turbulent image;

[0010] As a further improvement method of the present invention:

[0011] Optionally, in step S1, decomposing the image affected by turbulence using a non-subsampled wavelet transform and extracting features of the decomposed image includes:

[0012] S11: Non-subsampled wavelet transform:

[0013] The image P affected by turbulence is subjected to non-subsampled wavelet transform to obtain image features in different directions and scales, specifically:

[0014]

[0015]

[0016] Among them, P LL , P LH , P HL and P HH are the approximate component and the detailed component in the horizontal direction, the detailed component in the vertical direction, and the detailed component in the diagonal direction; k = 0, 1, 2, 3; h LL (k), h LH (k), h HL (k) and h HH (k) are the wavelet coefficients of the approximate component, the detailed component in the horizontal direction, the detailed component in the vertical direction and the detailed component in the diagonal direction respectively; (x, y) are the pixel positions; P(2x-k, 2y-k), P(2x-k, 2y-1-k), P(2x-1-k, 2y-k) and P(2x-1-k, 2y-1-k) are the values ​​of P at the positions (2x-k, 2y-k), (2x-k, 2y-1-k), (2x-1-k, 2y-k) and (2x-1-k, 2y-1-k) respectively;

[0017] S12: Feature extraction:

[0018] The local contrast C of the image affected by turbulence is calculated based on the image features of different directions and scales obtained in step S11. local , specifically:

[0019]

[0020] Among them, C local (x,y) is C local The value at (x, y); r is the window size; B r (x, y) is the pixel position within the range of r×r with (x, y) as the center; (u, v ...u, v) as the center; (u, v) is the pixel position within the range of r×r with (x r Elements in (x,y); is the difference value at point (u,v), calculated as:

[0021]

[0022] Among them, P LL (u,v),P LH (u,v),P HL (u,v) and P HH (u,v) are P LL , P LH , P HL and P HH The value at (u,v).

[0023] Optionally, in step S2, performing turbulence detection based on adaptive threshold segmentation to obtain a turbulence region in the image affected by turbulence includes:

[0024] Turbulence detection based on adaptive threshold segmentation, adaptive threshold T adapt Based on the local contrast C local The median value is determined as follows:

[0025] T adapt (x,y)=α·median 5×5 (C local (x,y)

[0026] Among them, α is the scaling factor; median 5×5 (C local (x,y)) is C in the range of 5×5 with (x,y) as the center. local The median value of T adapt (x,y) is T adapt The value at (x,y);

[0027] The turbulence area detection result S is determined as follows:

[0028]

[0029] Where S(x,y) is the value of S at (x,y), 1 represents the turbulent region, and 0 represents the non-turbulent region;

[0030] Optionally, in step S3, the grayscale features of the image affected by turbulence are calculated and mapped, and the features of the turbulence area obtained in step S2 are enhanced, including:

[0031] S31: Local grayscale difference statistics:

[0032] Calculate the gray level difference statistics LIDS in the local area p , specifically:

[0033]

[0034] Among them, LIDS p (x,y) is the LIDS in a square neighborhood centered at (x,y) with side length p p ; P(x+i,y+j) and P(xi,yj) are the values ​​of P at (x+i,y+j) and (xi,yj) respectively; i and j are integers in the range of -(p-1) / 2 to (p-1) / 2;

[0035] S32: Local grayscale centrality:

[0036] Calculate the pixel difference statistics LIC between the central position and the neighboring position in the local area p :

[0037]

[0038] Among them, LIC p (x,y) is the LIC in a square neighborhood centered at (x,y) with side length p. p ; P(x,y) is the value of P at (x,y);

[0039] S33: Feature Mapping and Enhancement:

[0040] Combined with LIDS p and LIC p Establishing feature map TRFM p , specifically:

[0041] TRFM p (x,y)=LIDS p (x,y)+β·LIDS p (x,y)

[0042] Among them, TRFM p (x,y) is the TRFM in a square neighborhood centered at (x,y) with side length p p ;β is the equilibrium parameter;

[0043] The turbulence region detection result S obtained in step S2 is used to enhance the turbulence region feature to obtain the turbulence region enhancement feature F enhanced , specifically:

[0044]

[0045] Among them, F enhanced (x,y) is F enhanced The value at (x,y); λ is the enhancement factor;

[0046] Optionally, in step S4, based on the turbulence region enhancement features obtained in step S3, an adaptive weight and a local similarity metric are calculated and used to repair the image affected by turbulence to obtain a repaired image, including:

[0047] S41: Calculate adaptive weights:

[0048] According to the turbulent region enhancement feature F obtained in step S3 enhanced Calculate the adaptive weight w, specifically:

[0049]

[0050] Where w(x,y) is the value of w at (x,y); Sum(F enhanced ) is for F enhanced sum;

[0051] S42: Calculate local similarity measure:

[0052] The local similarity metric ψ is used to quantify the similarity between the pixel to be repaired and the reference pixel, specifically:

[0053]

[0054] Where ψ(x,y,x',y') is the local similarity between pixels (dx,y) and (x',y'); d(x,y,x',y') is the Euclidean distance between pixels (x,y) and (x',y'); (x',y') is the pixel position within a 3×3 range centered at (x,y); σ1 is the spatial similarity parameter; σ2 is the intensity similarity parameter; e is a natural constant; P(x',y') is the value of P at (x',y'); It is the average pixel value within a 3×3 range centered at (x,y);

[0055] S43: Adaptive Image Inpainting:

[0056] Adaptive image restoration is performed using adaptive weights and local similarity metrics, specifically:

[0057]

[0058] Among them, P restored (x,y) is the value of P at (x,y) after repair; w(x',y') is the value of w at (x',y');

[0059] Optionally, in step S5, combining the turbulence region in the restored image with the non-turbulence region in the original image to obtain a final de-turbulenced image includes:

[0060] The turbulent area in the repaired image is combined with the non-turbulent area in the original image to obtain the final de-turbulent image P final , specifically:

[0061]

[0062] Among them, P final (x,y) is P final The value at (x,y);

[0063] The present invention also discloses a turbulence removal system based on image restoration, comprising:

[0064] Feature extraction module: Use non-subsampled wavelet transform to decompose the image affected by turbulence and extract the features of the decomposed image;

[0065] Turbulence detection module: turbulence detection based on adaptive threshold segmentation;

[0066] Feature enhancement module: calculates and maps the grayscale features of the image affected by turbulence, and performs feature enhancement on the turbulence area obtained in the turbulence detection module;

[0067] Inpainting module: Calculates adaptive weights and local similarity metrics for inpainting images affected by turbulence;

[0068] Fusion module: combines the turbulent regions in the inpainted image with the non-turbulent regions in the original image.

[0069] Beneficial effects:

[0070] The present invention uses non-subsampled wavelet transform to extract multi-scale and multi-directional features of images affected by turbulence, effectively capturing image details at different levels. Combined with adaptive threshold segmentation technology, it achieves accurate detection and positioning of turbulent areas, provides accurate target areas for subsequent image restoration, and significantly improves the accuracy and efficiency of turbulence removal processing.

[0071] By calculating local grayscale difference statistics and local grayscale centrality, the present invention innovatively enhances the image features in the turbulent region. This enhancement not only improves the texture contrast in the turbulent region, but also enhances the local edge information of the image, thereby providing richer and more accurate guidance information for image restoration.

[0072] This paper proposes an intelligent image restoration algorithm based on adaptive weights, which can dynamically adjust the restoration weights according to the characteristic strength of the turbulent area. This adaptive restoration strategy ensures that the restoration process can not only strengthen the damaged area in a targeted manner, but also maintain a natural transition and consistency with the surrounding non-turbulent areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic diagram of a process of a turbulence removal method based on image restoration according to an embodiment of the present invention;

[0074] Figure 2 FIG. 4 is a schematic diagram of a repair result according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0076] Embodiment 1:

[0077] A de-turbulence method based on image restoration, such as Figure 1 As shown, the following steps are included:

[0078] S1: Decomposing the image affected by turbulence using non-subsampled wavelet transform and extracting features of the decomposed image:

[0079] S11: Non-subsampled wavelet transform:

[0080] The image P affected by turbulence is subjected to non-subsampled wavelet transform to obtain image features in different directions and scales, specifically:

[0081]

[0082] Among them, P LL , P LH , P HL and P HH are the approximate component and the detailed component in the horizontal direction, the detailed component in the vertical direction, and the detailed component in the diagonal direction; k = 0, 1, 2, 3; h LL (k), h LH (k), h HL (k) and h HH (k) are the wavelet coefficients of the approximate component, the detailed component in the horizontal direction, the detailed component in the vertical direction and the detailed component in the diagonal direction respectively; (x, y) are the pixel positions; P(2x-k, 2y-k), P(2x-k, 2y-1-k), P(2x-1-k, 2y-k) and P(2x-1-k, 2y-1-k) are the values ​​of P at the positions (2x-k, 2y-k), (2x-k, 2y-1-k), (2x-1-k, 2y-k) and (2x-1-k, 2y-1-k) respectively;

[0083] h LL (k), h LH (k), h HL (k) and hHH (k) is calculated as follows:

[0084] h LL (k) = h L (k)*h L (k)

[0085] h LH (k) = h L (k)*h H (k)

[0086] h HL (k) = h H (k)*h L (k)

[0087] h HH (k) = h H (k)*h H (k)

[0088]

[0089] S12: Feature extraction:

[0090] The local contrast C of the image affected by turbulence is calculated based on the image features of different directions and scales obtained in step S11. local , specifically:

[0091]

[0092] Among them, C local (x,y) is C local The value at (x, y); r is the window size; B r (x, y) is the pixel position within the range of r×r with (x, y) as the center; (u, v ...u, v) as the center; (u, v) is the pixel position within the range of r×r with (x r Elements in (x,y); is the difference value at point (u,v), calculated as:

[0093]

[0094] Among them, P LL (u,v),P LH (u,v),P HL (u,v) and P HH (u,v) are P LL , P LH , P HL and P HH The value at (u,v).

[0095] The advantage of non-subsampled wavelet transform is that it can provide high-resolution multi-scale and multi-directional image features, which is crucial for accurately capturing subtle changes in images affected by turbulence. By avoiding the downsampling step in traditional wavelet transform, non-subsampled wavelet transform retains all the original information of the image, making it possible to more accurately identify and analyze the blur and distortion caused by turbulence in subsequent processing. In addition, the multi-directional decomposition capability of non-subsampled wavelet transform helps to fully understand the local features of the image, providing strong support for effectively extracting key image features and guiding subsequent turbulence detection and image restoration steps.

[0096] S2: Turbulence detection based on adaptive threshold segmentation to obtain the turbulence area in the image affected by turbulence:

[0097] Turbulence detection based on adaptive threshold segmentation, adaptive threshold T adapt Based on the local contrast C local The median value is determined as follows:

[0098] T adapt (x,y)=α·median 5×5 (C local (x,y)

[0099] Wherein, α is the scaling factor, which is 1.5 in this embodiment; median 5×5 (C local (x,y)) is C in the range of 5×5 with (x,y) as the center. local The median value of T adapt (x,y) is T adapt The value at (x,y);

[0100] The turbulence area detection result S is determined as follows:

[0101]

[0102] Among them, S(x,y) is the value of S at (x,y), 1 represents the turbulent region, and 0 represents the non-turbulent region.

[0103] The advantage of turbulence detection based on decomposed image features and adaptive threshold segmentation is that it combines the sensitivity of local image features and the flexibility of adaptive thresholds to accurately identify turbulent areas. This method can dynamically adjust the segmentation threshold to adapt to the various turbulence features that may exist in the image, improving the accuracy and reliability of detection. Adaptive threshold segmentation technology determines turbulent areas based on local contrast features, reducing dependence on global statistical characteristics, making the detection results more consistent with the actual image content.

[0104] S3: Calculate the grayscale features of the image affected by turbulence and perform mapping, perform feature enhancement on the turbulence area obtained in step S2, and obtain the enhanced features of the turbulence area:

[0105] S31: Local grayscale difference statistics:

[0106] Calculate the gray level difference statistics LIDS in the local area p , specifically:

[0107]

[0108] Among them, LIDS p (x,y) is the LIDS in a square neighborhood centered at (x,y) with side length p p ; P(x+i,y+j) and P(xi,yj) are the values ​​of P at (x+i,y+j) and (xi,yj) respectively; i and j are integers in the range of -(p-1) / 2 to (p-1) / 2;

[0109] S32: Local grayscale centrality:

[0110] Calculate the pixel difference statistics LIC between the central position and the neighboring position in the local area p :

[0111]

[0112] Among them, LIC p (x,y) is the LIC in a square neighborhood centered at (x,y) with side length p p ; P(x,y) is the value of P at (x,y);

[0113] S33: Feature Mapping and Enhancement:

[0114] Combined with LIDS p and LIC p Establishing feature map TRFM p , specifically:

[0115] TRFM p (x,y)=LIDS p (x,y)+β·LIDS p (x,y)

[0116] Among them, TRFM p (x,y) is the TRFM in a square neighborhood centered at (x,y) with side length p p ; β is a balance parameter, which is 0.5 in this embodiment;

[0117] The turbulence region detection result S obtained in step S2 is used to enhance the turbulence region feature to obtain the turbulence region enhancement feature F enhanced , specifically:

[0118]

[0119] Among them, F enhanced (x,y) is F enhanced The value at (x, y); λ is the enhancement factor, which is 1.8 in this embodiment.

[0120] Calculating and mapping the grayscale features of the image affected by turbulence, and then enhancing the features of the turbulent area can specifically enhance the significance of the image features in the turbulent area, thereby providing richer and more accurate guidance information for image restoration. Through the calculation of local grayscale difference statistics and local grayscale centrality, this step not only captures the subtle texture changes in the turbulent area, but also enhances the contrast of these areas, so that the damaged image details can be reconstructed more accurately in the subsequent restoration process. In addition, by adaptively enhancing the features of the turbulent area, this step also improves the adaptability of the algorithm to images of different turbulence levels and types, ensuring the consistency and naturalness of the restoration effect.

[0121] S4: Based on the turbulence region enhancement features obtained in step S3, the adaptive weights and local similarity metrics are calculated and used to repair the image affected by turbulence to obtain the repaired image:

[0122] S41: Calculate adaptive weights:

[0123] According to the turbulent region enhancement feature F obtained in step S3 enhanced Calculate the adaptive weight w, specifically:

[0124]

[0125] Where w(x,y) is the value of w at (x,y); Sum(F enhanced ) is for F enhanced sum;

[0126] S42: Calculate local similarity measure:

[0127] The local similarity metric ψ is used to quantify the similarity between the pixel to be repaired and the reference pixel, specifically:

[0128]

[0129] Where, ψ(x,y,x',y') is the local similarity between pixels (x,y) and (x',y'); d(x,y,x',y') is the Euclidean distance between pixels (x,y) and (x',y'); (x',y') is the pixel position within a 3×3 range centered at (x,y); σ1 is the spatial similarity parameter, which is 1 / 10 of the image width; σ2 is the intensity similarity parameter, which is 10% of the average grayscale value of the image; e is a natural constant; P(x',y') is the value of P at (x',y'); It is the average pixel value within a 3×3 range centered at (x,y);

[0130] S43: Adaptive Image Inpainting:

[0131] Adaptive image restoration is performed using adaptive weights and local similarity metrics, specifically:

[0132]

[0133] Among them, P restored (x,y) is the value of P at (x,y) after repair; w(x',y') is the value of w at (x',y');

[0134] Computing adaptive weights based on enhanced features in turbulent regions provides precise control and guidance for the image restoration process. This approach dynamically adjusts the restoration weight of each pixel by quantifying the strength of features in turbulent regions, ensuring that the restoration energy is concentrated in the areas where it is most needed. The calculation of adaptive weights enhances the algorithm's responsiveness to local features, allowing for detailed adjustments to specific areas affected by turbulence while maintaining overall image consistency. In addition, this feature-based weight allocation strategy also improves the flexibility and robustness of the algorithm, enabling it to adapt to different image content and turbulence levels.

[0135] S5: Combine the turbulent area in the restored image with the non-turbulent area in the original image to obtain the final de-turbulent image:

[0136] The turbulent area in the repaired image is combined with the non-turbulent area in the original image to obtain the final de-turbulent image P final ,like Figure 2 As shown, specifically:

[0137]

[0138] Among them, P final (x,y) is P final The value at (x,y).

[0139] Combining the turbulent areas in the inpainted image with the non-turbulent areas in the original image ensures that the final image retains the original quality and details of the non-turbulent areas while removing the turbulence. This approach avoids the degradation of the overall image quality by intelligently blending the inpainted and unaffected areas, while reducing possible inpainting artifacts.

[0140] Embodiment 2:

[0141] A de-turbulence system based on image restoration, including the following five modules:

[0142] Feature extraction module: Use non-subsampled wavelet transform to decompose the image affected by turbulence and extract the features of the decomposed image;

[0143] Turbulence detection module: turbulence detection based on adaptive threshold segmentation;

[0144] Feature enhancement module: calculates and maps the grayscale features of the image affected by turbulence, and performs feature enhancement on the turbulence area obtained in the turbulence detection module;

[0145] Inpainting module: Calculates adaptive weights and local similarity metrics for inpainting images affected by turbulence;

[0146] Fusion module: combines the turbulent regions in the inpainted image with the non-turbulent regions in the original image.

[0147] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0148] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0149] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A turbulence removal method based on image restoration, characterized in that: The following steps are involved: S1: Decompose the image affected by turbulence using non-subsampled wavelet transform and extract the features of the decomposed image; S2: turbulence detection based on adaptive threshold segmentation to obtain the turbulence area in the image affected by turbulence; S3: Calculate the grayscale features of the image affected by turbulence and perform mapping, perform feature enhancement on the turbulence area obtained in step S2, and obtain enhanced features of the turbulence area; The step S3 includes the following steps: S31: Local grayscale difference statistics: Calculate the gray level difference statistics LIDS in the local area p , specifically: Among them, LIDS p (x,y) is the LIDS in a square neighborhood centered at (x,y) with side length p p ; P(x+i,y+j) and P(xi,yj) are the values ​​of P at (x+i,y+j) and (xi,yj) respectively; i and j are integers in the range of -(p-1) / 2 to (p-1) / 2; S32: Local grayscale centrality: Calculate the pixel difference statistics LIC between the central position and the neighboring position in the local area p : Among them, LIC p (x,y) is the LIC in a square neighborhood centered at (x,y) with side length p p ; P(x,y) is the value of P at (x,y); S33: Feature Mapping and Enhancement: Combined with LIDS p and LIC p Establishing feature map TRFM p , specifically: TRFM p (x,y)=LIDS p (x,y)+β·LIDS p (x,y) Among them, TRFM p (x,y) is the TRFM in a square neighborhood centered at (x,y) with side length p p ;β is the equilibrium parameter; The turbulence region detection result S obtained in step S2 is used to enhance the turbulence region feature to obtain the turbulence region enhancement feature F enhanced , specifically: Among them, F enhanced (x,y) is F enhanced The value at (x,y); λ is the enhancement factor; S4: Based on the turbulence region enhancement features obtained in step S3, an adaptive weight and a local similarity metric are calculated and used to repair the image affected by turbulence to obtain a repaired image; S5: Combine the turbulent region in the restored image with the non-turbulent region in the original image to obtain the final de-turbulent image.

2. The de-turbulence method based on image restoration according to claim 1, characterized in that: The step S1 includes the following steps: S11: Non-subsampled wavelet transform: The image P affected by turbulence is subjected to non-subsampled wavelet transform to obtain image features in different directions and scales, specifically: Among them, P LL , P LH , P HL and P HH are the approximate component and the detailed component in the horizontal direction, the detailed component in the vertical direction, and the detailed component in the diagonal direction; k = 0, 1, 2, 3; h LL (k), h LH (k), h HL (k) and h HH (k) are the wavelet coefficients of the approximate component, the detailed component in the horizontal direction, the detailed component in the vertical direction, and the detailed component in the diagonal direction, respectively; (x, y) are the pixel positions; P(2x-k, 2y-k), P(2x-k, 2y-1-k), P(2x-1-k, 2y-k), and P(2x-1-k, 2y-1-k) are the values ​​of P at the positions (2x-k, 2y-k), (2x-k, 2y-1-k), (2x-1-k, 2y-k), and (2x-1-k, 2y-1-k), respectively; h LL (k), h LH (k), h HL (k) and h HH (k) is calculated as follows: h LL (k)=h L (k)*h L (k) h LH (k)=h L (k)*h H (k) h HL (k)=h H (k)*h L (k) h HH (k)=h H (k)*h H (k) S12: Feature extraction: The local contrast C of the image affected by turbulence is calculated based on the image features of different directions and scales obtained in step S11 local , specifically: Among them, C local (x,y) is C local The value at (x, y); r is the window size; B r (x,y) is the pixel position within the range of r×r with (x,y as the center); (u,v) is the element in Br(x,y); is the difference value at point (u,v), calculated as: Among them, P LL (u,v),P LH (u,v),P HL (u,v) and P HH (u,v) are P LL , P LH , P HL and P HH The value at (u,v).

3. The de-turbulence method based on image restoration according to claim 2, characterized in that: The step S2 includes the following steps: Turbulence detection based on adaptive threshold segmentation, adaptive threshold T adapt Based on the local contrast C local The median value is determined as follows: T adapt (x,y)=α·median 5×5 (C local (x,y)) Among them, α is the scaling factor; median 5×5 (C local (x,y)) is C in the range of 5×5 with (x,y) as the center. local The median value of T adapt (x,y) is T adpat The value at (x,y); The turbulence area detection result S is determined as follows: Among them, S(x,y) is the value of S at (x,y), 1 represents the turbulent region, and 0 represents the non-turbulent region.

4. The de-turbulence method based on image restoration according to claim 3, characterized in that: The step S4 includes the following steps: S41: Calculate adaptive weights: According to the turbulent region enhancement feature F obtained in step S3 enhanced Calculate the adaptive weight w, specifically: Where w(x,y) is the value of w at (x,y); Sum(F enhanced ) is for F enhanced sum; S42: Calculate local similarity measure: The local similarity metric ψ is used to quantify the similarity between the pixel to be repaired and the reference pixel, specifically: Where ψ(x,y,x',y') is the local similarity between pixels (x,y) and (x',y'); d(x,y,x',y') is the Euclidean distance between pixels (x,y) and (x',y'); (x',y') is the pixel position within a 3×3 range centered at (x,y); σ1 is the spatial similarity parameter; σ2 is the intensity similarity parameter; e is a natural constant; P(x',y') is the value of P at (x',y'); It is the average pixel value within a 3×3 range centered at (x,y); S43: Adaptive Image Inpainting: Adaptive image restoration is performed using adaptive weights and local similarity metrics, specifically: Among them, P restored (x,y) is the value of P at (x,y) after repair; w(x',y') is the value of w at (x',y').

5. The de-turbulence method based on image restoration according to claim 4, characterized in that: The step S5 includes the following steps: The turbulent area in the repaired image is combined with the non-turbulent area in the original image to obtain the final de-turbulent image P final , specifically: Among them, P final (x,y) is P final The value at (x,y).

6. A turbulence removal system based on image restoration, characterized in that: include: Feature extraction module: uses non-subsampled wavelet transform to decompose the image affected by turbulence and extract the features of the decomposed image; Turbulence detection module: turbulence detection based on adaptive threshold segmentation; Feature enhancement module: calculates and maps the grayscale features of the image affected by turbulence, and performs feature enhancement on the turbulence area obtained in the turbulence detection module; Inpainting module: Calculates adaptive weights and local similarity metrics for inpainting images affected by turbulence; Fusion module: combines the turbulent areas in the restored image with the non-turbulent areas in the original image; To realize a de-turbulence method based on image restoration as described in any one of claims 1-5.

Images