Passive broadband warning process image visual enhancement method based on Gaussian filtering

Abstract

The invention provides a passive broadband warning process image vision enhancement method based on Gaussian filtering, aiming at solving the problems of multiple targets, high-dynamic noise mixing, low contrast ratio and poor visual effect in a passive sonar broadband warning image. According to the method, firstly, a nonlinear (multiplicative) hybrid signal is converted into an additive model through logarithmic transformation, and then a maximum between-class variance method is used for carrying out clustering segmentation calculation on a gray value in an image. And then, gamma transformation is carried out on the image by using the segmentation calculation threshold, and target enhancement and noise suppression are realized. And finally, extracting a target in the image through Gaussian filtering to generate a new image, thereby realizing contrast enhancement of the passive sonar broadband warning image, and remarkably improving the visual effect.

Description

Technical Field

[0001] The invention relates to underwater target image processing technology, mainly a passive broadband warning process image visual enhancement method based on Gaussian filtering. Background Art

[0002] Underwater target image processing technology plays a vital role in underwater detection, ocean exploration, and resource exploration. However, due to the complexity of the underwater acoustic channel and the influence of target radiation noise on the time-varying characteristics of the underwater acoustic channel, underwater images often suffer from low contrast, blurring, and noise interference, which seriously hinder image analysis and recognition. Traditional underwater image enhancement methods are often limited to simple linear filtering, exponential transformation, and histogram equalization, which are difficult to effectively process complex mixed signal history images. Therefore, an increasing number of researchers are beginning to explore more effective image enhancement methods to address problems such as multiple targets, high-dynamic noise mixing, and low contrast.

[0003] The most similar technical solution to the present invention is a Harbin Engineering University invention patent, which is a nonlinear enhancement method for side-scan sonar images. The main steps of this method include:

[0004] 1. Get the original image from the original sonar data file I

[0005] 2. Linearly map the grayscale values ​​in the original image I to the grayscale range [0,1] to obtain the normalized image I1

[0006] 3. Perform extreme neighborhood suppression on image I1 to obtain the suppressed image I2

[0007] 4. Perform Gaussian smoothing on the suppressed image I2 to obtain image I3

[0008] 5. For the image I3 after Gaussian smoothing, first calculate the ratio of the low grayscale area and the high grayscale area to the number of pixels in the whole image according to the imaging principle of the side scan sonar image and the distribution characteristics of its grayscale histogram, and then calculate the maximum grayscale value S of the low grayscale area. b , the minimum grayscale value T in the high grayscale area b and the middle gray area (S b , T b )

[0009] 6. Perform nonlinear correction on the three areas of low grayscale, high grayscale and middle grayscale divided in the Gaussian smoothed image I3. Perform gamma correction on the low grayscale and middle grayscale areas, and perform proportional correction on the high grayscale area to obtain the enhanced image I4.

[0010] Currently, commonly used sonar image grayscale correction methods include logarithmic transformation, time-varying gain (TVG), histogram equalization, gamma correction, and the Retinex algorithm. Logarithmic transformation can expand low-grayscale values ​​in sonar images and compress high-grayscale values, thereby emphasizing low-grayscale portions of the sonar image. Time-varying gain (TVG) compensates for the sonar image's history and achieves image enhancement by controlling propagation loss parameters. Histogram equalization evenly distributes the grayscale range of sonar images, improving overall image contrast. Its simple principle and ease of implementation have led to its widespread application. Gamma correction compresses or expands the grayscale of sonar images by appropriately varying the gamma correction exponent. The conventional Retinex algorithm divides sonar images into two components: luminance and reflectance. It achieves image grayscale correction by suppressing the influence of the luminance image on the reflectance image.

[0011] While the aforementioned method has some effectiveness in grayscale correction, the fixed parameter settings are not universally applicable across batches of sonar images, and cannot adaptively correct the grayscale range of targets in sonar images exposed to strong reverberation interference. To improve the visual quality of passive sonar broadband warning images, address the problems of multiple targets, high-dynamic noise, low contrast, and poor visual quality, and reduce the difficulty for sonar operators to interpret them, new technical solutions are needed to address these issues. Summary of the Invention

[0012] Aiming at the problems of multiple targets, high-noise dynamic mixing, low contrast and poor visual effect in passive sonar broadband warning images, the present invention provides a passive broadband warning process image visual enhancement method based on Gaussian filtering.

[0013] The purpose of the present invention is achieved through the following technical solutions. A method for visual enhancement of passive broadband warning process images based on Gaussian filtering. The method first converts the nonlinear (multiplicative) mixed signal into an additive model through logarithmic transformation, and then uses the maximum inter-class variance method to perform cluster segmentation calculation on the grayscale values ​​in the image. Then, the image is gamma transformed using the segmentation calculation threshold to achieve target enhancement and noise suppression. Finally, the target in the image is extracted through Gaussian filtering to generate a new image, thereby achieving contrast enhancement of the passive sonar broadband warning image and significantly improving the visual effect. The specific steps include the following:

[0014] Step 1: Obtain the original image and preprocess it, including: performing logarithmic transformation on the original image I(x, y) to obtain the logarithmic transformed image I1(x, y), and performing normalization on I1(x, y) to obtain the normalized image I2(x, y);

[0015] Step 2: Use the maximum inter-class variance method to calculate the grayscale segmentation threshold vector θ = [θ1, θ2] of the pixel point of the image I2 (x, y). Use the segmentation threshold vector θ = [θ1, θ2] to divide the grayscale interval G of the pixel point I2 (x, y) into a low grayscale area G1, a medium grayscale area G2, and a high grayscale area G3;

[0016] Step 3: Use θ1 to perform Gamma correction on I2(x,y) to obtain the nonlinearly corrected image I3(x,y);

[0017] Step 4: Based on the grayscale segmentation result of step 3, a background image I4(x,y) containing low and medium grayscale areas and a target image I5(x,y) containing medium and high grayscale areas are obtained.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. Different normalization methods. Existing methods obtain the original image from the raw sonar data file and directly perform grayscale linear mapping to obtain the normalized image. However, the present invention first performs a logarithmic transformation on the original image, converting the multiplicative signal into an additive signal, and then performs grayscale linear mapping to obtain the normalized image, which helps improve the contrast of weak signals.

[0020] 2. Different clustering methods: The existing method clusters pixels based on the distribution characteristics of the grayscale histogram. However, the present invention clusters pixels using the maximum inter-class variance method, which is less sensitive to noise and has higher accuracy and efficiency.

[0021] 3. Different enhancement methods: The existing method performs gamma correction and scale correction on the image partitioned regions after Gaussian smoothing to achieve image enhancement. However, the present invention uses cluster segmentation thresholds to perform nonlinear correction and Gaussian smoothing on the normalized image partitioned regions, and outputs the difference between the images before and after filtering to achieve image enhancement. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art or ordinary technicians, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 Schematic diagram of the workflow of the present invention.

[0024] Figure 2 Schematic diagram of passive broadband warning image.

[0025] Figure 3Schematic diagram of the normalized grayscale relationship before and after correction.

[0026] Figure 4 Schematic diagram of the spatial energy spectrum of a single batch.

[0027] Figure 5 Schematic diagram of passive broadband warning-correction image.

[0028] Figure 6 Schematic diagram of passive broadband warning-background image.

[0029] Figure 7 Schematic diagram of passive broadband warning-target image.

[0030] Figure 8 Schematic diagram of passive broadband warning-weak target image.

[0031] Figure 9 Schematic diagram of passive broadband warning-strong target image.

[0032] Figure 10 Schematic diagram of passive broadband alert-visual enhancement image. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0034] This invention proposes a Gaussian-filtered visual enhancement method for passive broadband warning imagery. By introducing a logarithmic transformation to compress the image's dynamic range and enhance detail in low-contrast areas, the method converts multiplicative noise into additive noise. The logarithmically transformed image is then clustered and segmented using the maximum inter-class variance method, dividing the image into low, medium, and high grayscale regions based on grayscale values. Based on the segmentation results, the image is grayscale-corrected using a gamma transformation to further highlight target features, resulting in a target image containing medium and high grayscale regions and a background image containing low and medium grayscale regions. Finally, the two images are blurred using a Gaussian filter. The unfiltered and filtered images are then subtracted to extract the target's course. The resulting image is then added together to generate a new passive sonar broadband warning image, effectively improving image quality.

[0035] The workflow of the present invention is as follows Figure 1 As shown, the specific steps include:

[0036] Step 1: Get the original image I(x,y) and preprocess it, mainly including logarithmic transformation and normalization. Specifically, the following steps are included:

[0037] (1.1) Perform logarithmic transformation on the original image I(x,y) to reduce the dynamic range of the image, enhance the details of low-contrast areas, and convert the multiplicative noise in the image into additive noise to obtain the logarithmic transformed image I1(x,y), as shown in Figure 2 As shown, it can be expressed as:

[0038] I1(x,y)=λ*log(1+I(x,y))

[0039] Where λ represents the gain parameter, which is used to adjust the image contrast.

[0040] (1.2) Traverse every pixel of the image and obtain the maximum value I of the logarithmic transformation image I1(x,y) 1max and minimum value I 1min , using I 1max and I 1min Normalize I1(x,y) to get the normalized image I2(x,y), which can be expressed as:

[0041]

[0042] Step 2: Use the maximum inter-class variance method to calculate the grayscale segmentation threshold vector θ = [θ1, θ2] of the pixel point of the image I2 (x, y). Use the segmentation threshold vector θ = [θ1, θ2] to divide the grayscale interval G of the pixel point I2 (x, y) into a low grayscale area G1, a medium grayscale area G2, and a high grayscale area G3, which can be expressed as:

[0043] G=G1∪G2∪G3

[0044] Among them, G∈[0,1], G1∈[0,θ1], G2∈(θ1,θ2], G3∈(θ2,1].

[0045] The specific process includes the following:

[0046] (2.1) Calculate the grayscale mean value I of the pixel I2(x,y) 2ave .

[0047] (2.2) Traverse the segmentation threshold vector θ = [θ1, θ2], and the traversal range is the pixel grayscale interval G.

[0048] (2.3) According to the currently traversed thresholds θ1 and θ2, the pixel grayscale interval G is divided into a low grayscale area G1, a medium grayscale area G2 and a high grayscale area G3.

[0049] (2.4) Calculate the proportion of G1, G2 and G3 pixel points to the total number of pixels P1, P2, P3 and the grayscale mean G 1ave , G 2ave , G 3ave .

[0050] (2.5) Calculate the inter-class variance σ of G1, G2 and G3 2 , which can be expressed as:

[0051] σ 2 = Ρ1(G 1ave -I 2ave ) 2 +P2(G 2ave -I 2ave ) 2 +P3(G 3ave -I 2ave ) 2

[0052] (2.6) Repeat the process (2.2)-(2.5), with σ 2 The maximum threshold vector θ is the final segmentation threshold vector θ, and the grayscale partitions G1, G2 and G3 are determined.

[0053] Step 3: Use θ1 to perform Gamma correction on I2(x,y) to obtain the nonlinearly corrected image I3(x,y), which can be expressed as:

[0054] I3(x,y)=(I2(x,y) / θ1) γ

[0055] Where γ is the correction Gamma index, and the normalized grayscale relationship before and after correction is as follows: Figure 3 As shown, the single batch spatial energy spectrum is corrected, as shown in Figure 4 As shown, I2(x,y) is corrected to image I3(x,y), as shown Figure 5 It is found that the spatial energy spectrum contrast can be effectively improved.

[0056] Step 4: Based on the grayscale segmentation results of step 3, we obtain the background image I4(x, y) containing low and medium grayscale areas and the target image I5(x, y) containing medium and high grayscale areas, which can be expressed as:

[0057]

[0058] Construct a Gaussian kernel function f(x,y) of size 1×n, which can be expressed as:

[0059]

[0060] Where σ is the standard deviation, which determines the degree of diffusion of the Gaussian function, that is, the smoothness of the filter.

[0061] Convolve the background image I4(x,y) and the target image I5(x,y) with f(x,y) and then calculate the difference to obtain the passive broadband warning weak target image I 4_mask (x, y) and obtain the passive broadband warning strong target image I 5_mask (x,y), I 4_mask (x,y) and I 5_mask (x, y) are added together to obtain the passive broadband warning visual enhancement image I6(x, y) containing all targets, which can be expressed as:

[0062] I 4_mask (x,y)=I4(x,y)-I4(x,y)*f(x,y)

[0063] I 5_mask (x,y)=I5(x,y)-I5(x,y)*f(x,y)

[0064] I6(x,y)=I 4_mask (x,y)+I 5_mask (x,y).

[0065] The present invention uses the normalized image maximum inter-class variance segmentation threshold to perform grayscale correction, such as Figures 2 to 5 As shown, observe Figures 2 to 5 ,It can be found that using the maximum inter-class variance segmentation threshold to perform Gamma correction on the normalized image can simultaneously suppress background noise and enhance target brightness, effectively improving image contrast.

[0066] The present invention performs Gaussian blur on the partition results of image clustering, and outputs the difference superposition results of the images before and after blurring to achieve image enhancement. Figures 6 to 10 As shown in the figure, the present invention first uses the maximum inter-class variance segmentation threshold to divide the normalized image into a background image and a target image. Then, a Gaussian mask is generated to blur the background image and the target image separately. The difference between the blurring and the background image is then taken to extract the target trajectory in the background image and the target image. Compared with the method of directly using Gaussian blur background estimation to perform subtraction and extract edges, the partitioning method of the present invention can better preserve the target trajectory details in low-grayscale areas.

[0067] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the scope of the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for visual enhancement of passive broadband warning process images based on Gaussian filtering, characterized by: The steps are as follows: Step 1: Obtain the original image and preprocess it, including: performing logarithmic transformation on the original image I(x, y) to obtain the logarithmic transformed image I1(x, y), and performing normalization on I1(x, y) to obtain the normalized image I2(x, y); Step 2: Use the maximum inter-class variance method to calculate the grayscale segmentation threshold vector θ = [θ1, θ2] of the pixel point of the image I2 (x, y). Use the segmentation threshold vector θ = [θ1, θ2] to divide the grayscale interval G of the pixel point I2 (x, y) into a low grayscale area G1, a medium grayscale area G2, and a high grayscale area G3; Step 3: Use θ1 to perform Gamma correction on I2(x,y) to obtain the nonlinearly corrected image I3(x,y); Step 4: Based on the grayscale segmentation result of step 3, a background image I4(x,y) containing low and medium grayscale areas and a target image I5(x,y) containing medium and high grayscale areas are obtained.

2. The method for visual enhancement of passive broadband warning history images based on Gaussian filtering according to claim 1 is characterized in that: The step 1 specifically includes the following process: (1.1) Perform logarithmic transformation on the original image I(x,y) to reduce the dynamic range of the image, enhance the details of low-contrast areas, and convert the multiplicative noise in the image into additive noise. The resulting logarithmic transformed image I1(x,y) is expressed as: I1(x,y)=λ*log(1+I(x,y)) Where λ represents the gain parameter, which is used to adjust the image contrast; (1.2) Traverse every pixel of the image and obtain the maximum value I of the logarithmic transformation image I1(x,y) 1max and minimum value I 1min , using I 1max and I 1min Normalize I1(x,y) to get the normalized image I2(x,y), which is expressed as:

3. The method for visual enhancement of passive broadband warning history images based on Gaussian filtering according to claim 2, characterized in that: In step 2, the maximum inter-class variance method is used to calculate the grayscale segmentation threshold vector θ=[θ1,θ2] of the pixel point of the image I2(x,y). The grayscale interval G of the pixel point I2(x,y) is divided into a low grayscale area G1, a medium grayscale area G2 and a high grayscale area G3 using the segmentation threshold vector θ=[θ1,θ2], which can be expressed as: G=G1∪G2∪G3 Among them G∈[0,1], G1∈[0,θ1], G2∈(θ1,θ2], G3∈(θ2,1]; The specific process includes the following: (2.1) Calculate the grayscale mean value I of the pixel I2(x,y) 2ave ; (2.2) Traverse the segmentation threshold vector θ = [θ1, θ2], and the traversal range is the pixel grayscale interval G; (2.3) Based on the currently traversed thresholds θ1 and θ2, the pixel grayscale interval G is divided into a low grayscale area G1, a medium grayscale area G2, and a high grayscale area G3; (2.4) Calculate the proportion of G1, G2 and G3 pixel points to the total number of pixels P1, P2, P3 and the grayscale mean G 1ave , G 2ave , G 3ave ; (2.5) Calculate the inter-class variance σ of G1, G2 and G3 2 , expressed as: s 2 =P1(G 1ave -I 2ave ) 2 +P2(G 2ave -I 2ave ) 2 +P3(G 3ave -I 2ave ) 2 (2.6) Repeat the process (2.2)-(2.5), with σ 2 The maximum threshold vector θ is the final segmentation threshold vector θ, and the grayscale partitions G1, G2 and G3 are determined.

4. The method for visual enhancement of passive broadband warning history images based on Gaussian filtering according to claim 3 is characterized in that: In step 3, Gamma correction is performed on I2(x, y) using θ1 to obtain the nonlinearly corrected image I3(x, y), which is expressed as: I3(x,y)=(I2(x,y) / θ1) γ Where γ is the correction Gamma index, which is the normalized grayscale relationship before and after correction.

5. The method for visual enhancement of passive broadband warning history images based on Gaussian filtering according to claim 4 is characterized in that: In step 4, based on the grayscale segmentation result of step 3, a background image I4(x, y) containing low and medium grayscale areas and a target image I5(x, y) containing medium and high grayscale areas are obtained, which can be expressed as: Construct a Gaussian kernel function f(x,y) of size 1×n, expressed as: Where σ is the standard deviation, which determines the degree of diffusion of the Gaussian function, that is, the smoothness of the filter; Convolve the background image I4(x,y) and the target image I5(x,y) with f(x,y) and then calculate the difference to obtain the passive broadband warning weak target image I 4_mask (x, y) and obtain the passive broadband warning strong target image I 5_mask (x,y), I 4_mask (x,y) and I 5_mask (x, y) are added together to obtain the passive broadband warning visual enhancement image I6(x, y) containing all targets, which is expressed as: Yo 4_mask (x,y)=I4(x,y)-I4(x,y)*f(x,y) I 5_mask (x,y)=I5(x,y)-I5(x,y)*f(x,y) I6(x,y)=I 4_mask (x,y)+I 5_mask (x,y)。

Images