A Multi-Feature Image Clutter Scale Quantization Method

The SLIC algorithm divides the image into superpixels and extracts multiple features for quantization, which solves the problems of scene object complexity and single feature extraction in existing technologies and improves the accuracy of clutter scale quantization.

CN118334358BActive Publication Date: 2026-11-17XIDIAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410307181.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-18
Publication Date
2026-11-17
Estimated Expiration
2044-03-18

AI Technical Summary

Technical Problem

Existing clutter scale quantization methods ignore the complexity and irregularity of objects in the scene, resulting in inaccurate evaluation results. Furthermore, the feature extraction methods are too simplistic and the quantization is not accurate enough.

Method used

A multi-feature image clutter scale quantization method is adopted. The original image is divided into multiple superpixels using the SLIC algorithm, and contrast, texture and contour features are extracted. The similarity between the target and the background is calculated and then fused and quantized.

Benefits of technology

By taking into account the complexity and irregularity of objects in the scene, the accuracy of feature extraction is improved, and the accuracy of clutter scale quantization results is also improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118334358B_ABST
    Figure CN118334358B_ABST
Patent Text Reader

Abstract

This invention discloses a multi-feature image clutter scale quantization method, comprising: determining the number of superpixels to be divided based on the number of pixels in the original image and the size of the target in the original image; dividing the original image into K superpixels using the SLIC algorithm; extracting image features that affect the difficulty of target acquisition from the target and each superpixel, including contrast features, texture features, and contour features; calculating the contrast similarity, texture similarity, and contour similarity between the target and the background based on the image features of the target and each superpixel; and fusing the contrast similarity, texture similarity, and contour similarity to obtain the clutter scale quantization result of the original image. This invention fully considers the complexity and irregularity of the objects that make up the scene, avoids dividing two completely unrelated objects into the same unit, and overcomes the problem of single feature extraction in existing clutter scale quantization models, thus resulting in more accurate clutter quantization results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method for multi-feature image clutter scale quantization. Background Technology

[0002] With the continuous improvement of the design level and manufacturing process of modern optoelectronic imaging systems, the impact of noise on image quality is decreasing, allowing observers to acquire more target details and scene content. The complexity of the scene content itself is gradually replacing noise as a crucial factor affecting the target acquisition performance of optoelectronic imaging systems. Therefore, to improve the evaluation accuracy of target acquisition performance of optoelectronic imaging systems, it is particularly important to quantitatively describe the impact of image background on target detection and acquisition, and to establish an accurate clutter scale quantification model.

[0003] Currently, scholars both domestically and internationally have proposed various clutter scale quantization methods, mainly divided into two categories: quantization methods based on human vision and quantization methods for automatic target recognition. Among them, quantization methods based on human vision include those based on mathematical definitions, early-stage human vision, and late-stage human vision; while quantization methods for automatic target recognition vary depending on the target detection algorithm.

[0004] However, existing clutter scale quantization methods have the following problems: First, existing clutter scale quantization methods use a rectangular window twice the size of the target to divide the image into multiple units for evaluation, ignoring the complexity and irregularity of the objects that make up the scene. This may result in two completely unrelated objects being classified into the same unit, affecting the accuracy of the evaluation results. Second, existing clutter scale quantization methods use relatively simple feature extraction methods, resulting in insufficient quantization accuracy. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a multi-feature image clutter scale quantization method. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] This invention provides a method for multi-feature image clutter scale quantization, comprising:

[0007] Based on the number of pixels in the original image and the size of the target in the original image, determine the number of superpixels K that need to be divided;

[0008] The original image is divided into K superpixels using the Simple Linear Iterative Clustering (SLIC) algorithm.

[0009] Extract the image features that affect the difficulty of target acquisition from the target and each superpixel. The image features include: contrast features, texture features, and contour features.

[0010] Based on the target and the image features of each superpixel, calculate the contrast similarity, texture similarity, and contour similarity between the target and the background;

[0011] The contrast similarity, texture similarity, and contour similarity are fused to obtain the clutter scale quantization result of the original image.

[0012] In one embodiment of the present invention, the step of determining the number K of superpixels to be divided based on the number of pixels in the original image and the size of the target in the original image includes:

[0013] Determine the size l of the target in the original image, where, N x N y These represent the number of pixels in the horizontal and vertical directions of the target, respectively.

[0014] Count the number of pixels N in the original image;

[0015] Based on the target size l and the number of pixels N in the original image, determine the number of superpixels to be divided:

[0016] K = N / (4l) 2 ).

[0017] In one embodiment of the present invention, the step of dividing the original image into K superpixels using the Simple Linear Iterative Clustering (SLIC) algorithm includes:

[0018] After converting the original image from the RGB color space to the CIE-Lab color space, a 5-dimensional feature vector is constructed using the lab value (L, a, b) of each pixel and its position (x, y) in the original image; where L represents brightness, a represents red-green hue, b represents yellow-blue hue, and x and y represent the row and column of the pixel, respectively.

[0019] Based on the 5-dimensional feature vectors of pixels, K cluster centers C are initialized and generated. k =[L k ,a k ,b k ,x k ,y k ]; where L k a k b k Let x represent the brightness, red-green hue, and yellow-blue hue of the k-th cluster center, respectively. k y k Let represent the row and column of the k-th cluster center, respectively, and let d be the distance between two adjacent cluster centers;

[0020] Iterate through all pixels in the original image and assign each pixel to the category corresponding to the nearest cluster center within a 2d×2d range; the distance between the pixel and the cluster center is calculated using the following formula:

[0021]

[0022] D CIELab =‖Q(L,a,b)-C k (L k ,a k ,b k )‖;

[0023] D co =‖Q(x,y)-C k (x k ,y k )‖;

[0024] In the formula, m represents the preset weights, m = 20, Q(L,a,b), C k (L k ,a k ,b k ) represent pixel P and the k-th cluster center C, respectively. k The brightness, red-green hue, and yellow-blue hue, Q(x,y), C k (x k ,y k ) represent pixel Q and the k-th cluster center C, respectively. k The row and column in the original image;

[0025] The cluster centers are updated by calculating the average position of all pixels in each category;

[0026] Return to the step of traversing each pixel in the original image and assigning it to the category corresponding to the nearest cluster center within a 2d×2d range for each pixel, until a preset number of iterations is reached, resulting in K superpixels.

[0027] In one embodiment of the present invention, the contrast features of the target or each superpixel are extracted according to the following steps:

[0028] The R channel value range of 0-255 in the RGB color space is divided into 10 first intervals: [0,24], [25,49], ..., [224, 255]. The number of pixels (R1, R2, ..., R) within each first interval of the target or superpixel whose R channel value falls within that interval is then counted. 10 ;

[0029] The G channel value range of 0-255 in the RGB color space is divided into 10 second intervals: [0,24], [25,49], ..., [224, 255]. The number of pixels (G1, G2, ..., G) whose G channel values ​​within the target or superpixel fall within each of these second intervals is then counted. 10 ;

[0030] The range of B channel values ​​in the RGB color space from 0 to 255 is divided into 10 third intervals: [0, 24], [25, 49], ..., [224, 255]. The number of pixels (B1, B2, ..., B1) within the target or superpixel whose B channel values ​​fall within each of these third intervals is then counted. 10 ;

[0031] Obtain the contrast characteristics of the target or superpixel: T RGB =[R1,R2,...,R 10 ,G1,G2,...,G 10 ,B1,B2,...,B 10 ].

[0032] In one embodiment of the present invention, the texture features of the target or each superpixel are extracted according to the following steps:

[0033] The original image is converted into a grayscale image, and the grayscale co-occurrence matrix of the target and each superpixel is calculated.

[0034] Calculate the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix respectively:

[0035]

[0036]

[0037]

[0038]

[0039] In the formula, P(i,j) represents the element value in the i-th row and j-th column of the gray-level co-occurrence matrix, and M represents the number of pixels in the superpixel;

[0040] Based on the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix, the texture features of the target or superpixels are obtained:

[0041] T GLCM =[Con,Asm,Ent,H].

[0042] In one embodiment of the present invention, the contour features of the target or each superpixel are extracted according to the following steps:

[0043] Get the number of pixels in the target superpixel and the perimeter of the target superpixel;

[0044] The contour feature T of the target or superpixel is obtained by calculating the ratio of the number of pixels in the target / superpixel to the square of its perimeter. sh .

[0045] In one embodiment of the present invention, the similarity between the target and the background in terms of contrast, texture, and contour is calculated according to the following formulas:

[0046]

[0047] In the formula, T RGB,k T represents the contrast feature of the k-th superpixel. RGB,tar Indicates the contrast characteristics of the target;

[0048]

[0049] In the formula, T GLCM,k T represents the texture feature of the k-th superpixel. GLCM,tar Represents the texture features of the target;

[0050]

[0051] In the formula, T sh,k T represents the contour feature of the k-th superpixel. sh,tar Represents the contour features of the target.

[0052] In one embodiment of the present invention, the step of fusing the contrast similarity, texture similarity, and contour similarity to obtain the clutter scale quantization result of the original image includes:

[0053] Using the preset fusion coefficient a RGB a GLCM and a sh The contrast similarity, texture similarity, and contour similarity are fused to obtain the clutter scale quantization result of the original image:

[0054] S = a RGB S RGB +a GLCM S GLCM +a sh S sh .

[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0056] This invention provides a multi-feature image clutter scale quantization method. It uses the SLIC algorithm to divide the original image into multiple homogeneous superpixels, instead of using a rectangular window twice the size of the target to divide the original image into multiple units in the prior art. This fully considers the complexity and irregularity of the objects that make up the scene, avoids dividing two completely unrelated objects into the same unit, and helps to improve the accuracy of subsequent feature extraction and mitigate the adverse effects on the image clutter scale quantization results.

[0057] Furthermore, this invention extracts image features that affect the difficulty of target acquisition from the target and each superpixel, such as contrast features, texture features, and contour features, overcoming the problem of single feature extraction in existing clutter scale quantization models. Therefore, the clutter quantization results are more accurate.

[0058] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0059] Figure 1 This is a flowchart of a multi-feature image clutter scale quantization method provided in an embodiment of the present invention. Detailed Implementation

[0060] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0061] Figure 1 This is a flowchart of a multi-feature image clutter scale quantization method provided in an embodiment of the present invention.

[0062] like Figure 1 As shown, this embodiment of the invention provides a method for multi-feature image clutter scale quantization, including:

[0063] S1. Determine the number of superpixels K to be divided based on the number of pixels in the original image and the size of the target in the original image;

[0064] S2. The original image is divided into K superpixels using the Simple Linear Iterative Clustering (SLIC) algorithm.

[0065] S3. Extract the image features that affect the difficulty of target acquisition from the target and each superpixel. The image features include: contrast features, texture features and contour features.

[0066] S4. Based on the image features of the target and each superpixel, calculate the contrast similarity, texture similarity, and contour similarity between the target and the background;

[0067] S5. The contrast similarity, texture similarity and contour similarity are fused to obtain the clutter scale quantization result of the original image.

[0068] Optionally, step S1, which involves determining the number of superpixels K to be divided based on the number of pixels in the original image and the size of the target in the original image, includes:

[0069] S101. Determine the size l of the target in the original image, where, N x N y These represent the number of pixels in the horizontal and vertical directions of the target, respectively.

[0070] S102. Count the number of pixels N in the original image;

[0071] S103. Based on the target size l and the number of pixels N in the original image, determine the number of superpixels to be divided:

[0072] K = N / (4l) 2 ).

[0073] In this embodiment, the Simple Linear Iterative Clustering (SLIC) algorithm can be used to group neighboring pixels with similar features in the original image into the same superpixel. Specifically, step S2, which uses the SLIC algorithm to divide the original image into K superpixels, includes:

[0074] S201. After converting the original image from RGB color space to CIE-Lab color space, a 5-dimensional feature vector is constructed using the lab value (L,a,b) of each pixel and its position (x,y) in the original image; where L represents brightness, a represents red-green hue, b represents yellow-blue hue, and x and y represent the row and column of the pixel, respectively.

[0075] S202. Initialize and generate K cluster centers C based on the 5-dimensional feature vector of each pixel. k =[L k ,a k ,b k ,x k ,y k ]; where L k a k b k Let x represent the brightness, red-green hue, and yellow-blue hue of the k-th cluster center, respectively. k y k Let represent the row and column of the k-th cluster center, respectively, and let d be the distance between two adjacent cluster centers;

[0076] S203. Traverse each pixel in the original image and assign it to the category corresponding to the nearest cluster center within a 2d×2d range for each pixel; wherein, the distance between the pixel and the cluster center is calculated according to the following formula:

[0077]

[0078] D CIELab =‖Q(L,a,b)-C k (L k ,a k ,b k )‖;

[0079] GD co =‖Q(x,y)-C k (x k ,y k )‖;

[0080] In the formula, m represents the preset weights, m = 20, Q(L,a,b), C k (L k ,a k ,b k ) represent pixel Q and the k-th cluster center C, respectively. k The brightness, red-green hue, and yellow-blue hue, Q(x,y), C k (x k ,y k ) represent pixel Q and the k-th cluster center C, respectively. k The row and column it is located in in the original image;

[0081] S204. Update the cluster center by calculating the average position of all pixels in each category;

[0082] S205. Return to the step of traversing each pixel in the original image and assigning it to the category corresponding to the nearest cluster center within a 2d×2d range for each pixel, until the preset number of iterations is reached, resulting in K superpixels.

[0083] Furthermore, this embodiment extracts image features of the target and each superpixel from the perspectives of contrast, texture, and shape contour.

[0084] Specifically, the contrast features of the target or each superpixel are extracted according to the following steps:

[0085] The R channel value range of 0-255 in the RGB color space is divided into 10 first intervals: [0,24], [25,49], ..., [224,255]. The number of pixels (R1, R2, ..., R) within each first interval of the target or superpixel is then counted. 10 ;

[0086] The G channel value range of 0-255 in the RGB color space is divided into 10 second intervals: [0,24], [25,49], ..., [224, 255]. The number of pixels (G1, G2, ..., G) whose G channel values ​​within the target or superpixel fall within each second interval is then counted. 10 ;

[0087] The range of B channel values ​​in the RGB color space from 0 to 255 is divided into 10 third intervals: [0, 24], [25, 49], ..., [224, 255]. The number of pixels (B1, B2, ..., B1) within the target or superpixel whose B channel values ​​fall within each of these third intervals is then counted. 10 ;

[0088] Obtain the contrast characteristics of the target or superpixel: T RGB =[R1,R2,...,R 10 ,G1,G2,...,G 10 ,B1,B2,...,B 10 ].

[0089] For example, the texture features of the target or each superpixel are extracted according to the following steps:

[0090] The original image is converted to a grayscale image, and the grayscale co-occurrence matrix of the target and each superpixel is calculated.

[0091] Calculate the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix respectively:

[0092]

[0093]

[0094]

[0095]

[0096] In the formula, P(i,j) represents the element value in the i-th row and j-th column of the gray-level co-occurrence matrix, and M represents the number of pixels in the superpixel;

[0097] Based on the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix, the texture features of the target or superpixels are obtained:

[0098] T GLCM =[Con,Asm,Ent,H].

[0099] It should be understood that the contrast Con of the gray-level co-occurrence matrix reflects the sharpness of the image and the depth of the texture grooves, the energy Asm reflects the coarseness of the texture and the uniformity of the gray-level distribution, the entropy Ent of the gray-level co-occurrence matrix reflects randomness, and the inverse variance H reflects the degree of local texture variation.

[0100] Further, extract the contour features of the target or each superpixel according to the following steps:

[0101] Get the number of pixels in the target superpixel and the perimeter of the target superpixel;

[0102] The contour feature T of the target or superpixel is obtained by calculating the ratio of the number of pixels in the target / superpixel to the square of its perimeter. sh .

[0103] It should be noted that for narrower targets or superpixel blocks, the contour feature T... sh The smaller the value, the more rounded the region, and the better the contour feature T. sh The larger the value, the better.

[0104] Optionally, in step S4, the similarity between the target and the background in contrast, texture, and contour is calculated according to the following formulas:

[0105]

[0106] In the formula, T RGB,k T represents the contrast feature of the k-th superpixel. RGB,tar Indicates the contrast characteristics of the target;

[0107]

[0108] In the formula, T GLCM,k T represents the texture feature of the k-th superpixel. GLCM,tar Represents the texture features of the target;

[0109]

[0110] In the formula, T sh,k T represents the contour feature of the k-th superpixel. sh,tar Represents the contour features of the target.

[0111] Step S5, which involves fusing contrast similarity, texture similarity, and contour similarity to determine the clutter scale of the original image, includes:

[0112] Using the preset fusion coefficient a RGB a GLCM and a shBy fusing contrast similarity, texture similarity, and contour similarity, the clutter scale of the original image is obtained:

[0113] S = a RGB S RGB +a GLCM S GLCM +a sh S sh .

[0114] As can be seen from the above embodiments, the beneficial effects of the present invention are as follows:

[0115] This invention provides a multi-feature image clutter scale quantization method. It uses the SLIC algorithm to divide the original image into multiple homogeneous superpixels, instead of using a rectangular window twice the size of the target to divide the original image into multiple units in the prior art. This fully considers the complexity and irregularity of the objects that make up the scene, avoids dividing two completely unrelated objects into the same unit, and helps to improve the accuracy of subsequent feature extraction and mitigate the adverse effects on the image clutter scale quantization results.

[0116] Furthermore, this invention extracts image features that affect the difficulty of target acquisition from the target and each superpixel, such as contrast features, texture features, and contour features, overcoming the problem of single feature extraction in existing clutter scale quantization models. Therefore, the clutter quantization results are more accurate.

[0117] In the description of this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0118] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for multi-feature image clutter scale quantization, characterized in that, include: Based on the number of pixels in the original image and the size of the target in the original image, determine the number of superpixels needed for segmentation. ; The original image was divided into groups using the Simple Linear Iterative Clustering (SLIC) algorithm. One superpixel; Extract the image features that affect the difficulty of target acquisition from the target and each superpixel. The image features include: contrast features, texture features, and contour features. Based on the target and the image features of each superpixel, calculate the contrast similarity, texture similarity, and contour similarity between the target and the background; The contrast similarity, texture similarity, and contour similarity are fused to obtain the clutter scale quantization result of the original image; Specifically, the number of superpixels to be divided is determined based on the number of pixels in the original image and the size of the target in the original image. The steps include: Determine the size of the target in the original image l ,in, , , These represent the number of pixels in the horizontal and vertical directions of the target, respectively. Count the number of pixels in the original image. ; Based on the size of the target l and the number of pixels in the original image Determine the number of superpixels needed for the division: 。 2. The multi-feature image clutter scale quantization method according to claim 1, characterized in that, The original image was divided into groups using the Simple Linear Iterative Clustering (SLIC) algorithm. The steps for each superpixel include: After converting the original image from the RGB color space to the CIE-Lab color space, the lab value of each pixel is used. and its position in the original image. This constitutes a 5-dimensional feature vector; where, Indicates brightness, Indicates red-green hue. Indicates the degree of yellow-blue. , These represent the row and column where the pixel is located, respectively. Initialize and generate 5-dimensional feature vectors based on pixels. Cluster centers ;in, , , They represent the first k The brightness, red-green hue, and yellow-blue hue of each cluster center , They represent the first k The rows and columns containing the cluster centers, and the distance between any two adjacent cluster centers are... d ; Iterate through each pixel in the original image, and at each pixel 2 d ×2 d Within a certain range, each pixel is assigned to the category corresponding to the nearest cluster center; the distance between a pixel and the cluster center is calculated using the following formula: ; ; ; In the formula, Indicates the preset weight. , , Representing pixels and the k Cluster centers Brightness, red-green hue, and yellow-blue hue, , Representing pixels and the k Cluster centers The row and column in the original image; The cluster centers are updated by calculating the average position of all pixels in each category; Return to each pixel in the original image, and at each pixel 2 d ×2 d The process of assigning elements to the nearest cluster center within a given range is repeated until a preset number of iterations is reached, resulting in the desired partitioning. One superpixel.

3. The multi-feature image clutter scale quantization method according to claim 1, characterized in that, Extract the contrast features of the target or each superpixel using the following steps: The R channel value range of 0-255 in the RGB color space is divided into 10 first intervals: [0,24], [25,49], ..., [224,255]. The number of pixels whose R channel value within the target or superpixel falls within each of the first intervals is then counted. ; The G channel value range of 0-255 in the RGB color space is divided into 10 second intervals: [0,24], [25,49], ..., [224,255]. The number of pixels whose G channel value within the target or superpixel falls within each of these second intervals is then counted. ; The range of B channel values ​​in the RGB color space from 0 to 255 is divided into 10 third intervals: [0, 24], [25, 49], ..., [224, 255]. The number of pixels whose B channel values ​​within the target or superpixel fall within each of these third intervals is then counted. ; Obtain the contrast features of the target or superpixel: .

4. The multi-feature image clutter scale quantization method according to claim 3, characterized in that, Extract the texture features of the target or each superpixel using the following steps: The original image is converted into a grayscale image, and the grayscale co-occurrence matrix of the target and each superpixel is calculated. Calculate the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix respectively: ; ; ; ; In the formula, Represents the gray-level co-occurrence matrix of the th i line, number j The element values ​​of the column, This indicates the number of pixels in the superpixel; Based on the contrast, energy, entropy, and inverse variance of the gray-level co-occurrence matrix, the texture features of the target or superpixels are obtained: 。 5. The multi-feature image clutter scale quantization method according to claim 4, characterized in that, Extract the contour features of the target or each superpixel using the following steps: Get the number of pixels in the target superpixel and the perimeter of the target superpixel; The contour features of the target or superpixel are obtained by calculating the ratio of the number of pixels in the target / superpixel to the square of its perimeter. .

6. The multi-feature image clutter scale quantization method according to claim 5, characterized in that, The following formulas are used to calculate the contrast similarity, texture similarity, and contour similarity between the target and the background: ; In the formula, Indicates the first k Contrast characteristics of each superpixel Indicates the contrast characteristics of the target; ; In the formula, Indicates the first k Texture features of a superpixel Represents the texture features of the target; ; In the formula, Indicates the first k The contour features of each superpixel Represents the contour features of the target.

7. The multi-feature image clutter scale quantization method according to claim 6, characterized in that, The step of fusing the contrast similarity, texture similarity, and contour similarity to obtain the clutter scale quantization result of the original image includes: Using preset fusion coefficients , and The contrast similarity, texture similarity, and contour similarity are fused to obtain the clutter scale quantization result of the original image: 。