Infrared image segmentation method based on local region self-organizing mapping algorithm

Abstract

The application discloses an infrared image segmentation method based on a local region self-organizing mapping algorithm, which comprises the following steps: firstly, a local region self-organizing mapping algorithm is designed, a plurality of feature bias fields are extracted by moving a local sliding window before iterative evolution of a level set, and a plurality of feature local data driving terms are constructed under a multiplicative bias field model framework; secondly, a plurality of feature global data driving terms are calculated, and a plurality of feature hybrid data driving terms are constructed through an adaptive weight function; then, an adaptive regularization function is used to regularize an energy value range of the plurality of feature hybrid data driving terms, and an initial level set is driven to perform iterative evolution; then, in the iterative process of the level set, a hyperbolic tangent function is used to keep the level set in the iterative evolution, and a symbol rule feature of positive outside and negative inside and a distance rule feature with a modulus value of 1 are kept; meanwhile, a mean filter template is used to continuously smooth the level set and eliminate redundant non-boundary contour lines; finally, a gradient descent method is used to continuously iteratively solve a minimum value of an energy function until a convergence criterion of the level set is reached or a maximum iteration number is reached, and then the method is stopped, at this time, a position of a zero level set is output, and image segmentation is completed. The local region self-organizing mapping algorithm can accurately mine a foreground contour from an infrared image, and has good segmentation speed and precision.

Description

Technical Field

[0001] This invention belongs to the field of infrared image segmentation, specifically an infrared image segmentation method based on a local region self-organizing mapping algorithm. Background Technology

[0002] The goal of infrared image segmentation is to extract the contours of several foreground targets from the input infrared image. The level set method (geometric active contour model) is one of the more common methods in image segmentation tasks. Its essential feature is to construct data-driven terms to guide the iterative evolution of the level set. Through gradient descent, a stable zero level set is iteratively solved as the final target boundary position, and several smooth and closed curves are used to characterize the size and shape of the foreground target contour.

[0003] However, traditional level set methods, such as RSF and LIF models, only consider a single gray-level feature of the input image when solving for the minimum energy function, thus only effectively segmenting a portion of visible light images. For infrared images with non-uniform gray levels, a single gray-level feature is often insufficient to characterize the detailed information of the target, easily leading to edge leakage, false targets, and other problems, resulting in a significant decrease in segmentation accuracy. Furthermore, traditional level set methods use a large number of convolutions to update the gray-level fitting values ​​inside and outside the contour line during the iterative evolution of the level set, generating significant computational overhead and thus greatly reducing segmentation speed. Summary of the Invention

[0004] To address the shortcomings of traditional level set methods in segmenting infrared images, this invention proposes an infrared image segmentation method based on the local region self-organizing map algorithm.

[0005] The technical solution to achieve the purpose of this invention is: an infrared image segmentation method based on a local region self-organizing map algorithm, comprising:

[0006] Step 1: Calculate the multi-feature map of the infrared input image, which includes a grayscale feature map, a roughness feature map, and a gradient feature map;

[0007] Step 2: Calculate the multi-feature bias field using a local region self-organizing map-based algorithm;

[0008] Step 3: Initialize the level set;

[0009] Step 4: Calculate the multi-feature local data driving term;

[0010] Step 5: Calculate the global data-driven terms for multiple features;

[0011] Step 6: Use an adaptive weighting function to combine local and global data-driven terms of multiple features to construct a multi-feature hybrid data-driven term;

[0012] Step 7: After regularizing the energy range of the multi-feature mixed data driving term using an adaptive regularization function, iteratively solve the level set to obtain the converged zero level set as the contour line of the target boundary, thus completing the contour segmentation.

[0013] Preferably, the roughness feature map is as follows:

[0014]

[0015] In the formula, w r (x,y) is a local window of size r×r located at pixel x=(x,y), μ r It is a local window w r The average gray value within (x,y) This is a grayscale feature map.

[0016] Preferably, gradient feature map Specifically:

[0017]

[0018] In the formula, G x (x,y),G y (x, y) are grayscale feature maps respectively. The gradients in the horizontal and vertical directions.

[0019] Preferably, the specific method for calculating multi-feature bias fields using a local region self-organizing map algorithm is as follows:

[0020] Step 2.1: Apply the following to a multi-feature map of size m×n. At the edge position, pixels are expanded by copying the boundary with a size of w×w;

[0021] Step 2.2: Expand the multi-feature map by three pixels to a size of (m+2w, n+2w) respectively. Establish the xy pixel coordinate system with the left top corner as the origin (0,0);

[0022] Step 2.3: Set up a partial sliding window Ω z The center point is z = (x, y) and the size is (2w + 1). 2 ;

[0023] Step 2.4: Set the first local sliding window Ω in each of the multi-feature input maps. z＝(w,w) And randomly select an input sample vector, defined as X i =[x i1 ,x i2 ,...,xip ], convert it to power v = 0.5, i is the index of the input sample, p is the dimension of the feature, and an appropriate initial learning rate η(0) = 1 and initial dominant neighborhood are set.

[0024] Step 2.5: Initialize the weight vector W of the output layer neurons of the self-organizing map network using random numbers. q =[w q1 ,w q2 ,...,w qp ], where q is the index of the neuron;

[0025] Step 2.6: Calculate X' i With W q Manhattan distance x i ' o For the input sample vector X' i For the value of the o-th feature on the i-th sample, w qo The weight vector W q For the value of the o-th feature on the q-th neuron, select the best matching unit q with the smallest Manhattan distance. * =argmin q∈{1,2,...,p} d(X′ i W q ), and establish the optimal matching unit q * dominant neighborhood

[0026] Step 2.7: Update the best matching unit q * In the topology-dominant neighborhood The associated weight vector of neurons;

[0027] Step 2.8: Repeat steps 2.4 to 2.7 until the learning rate η(n) decays to 0, and obtain the corresponding multi-feature bias field fitting values ​​within the current local sliding window.

[0028] Step 2.9: Return to step 2.4 and proceed to the next local sliding window Ω. z Continue the above process until all pixels in the input feature image have been traversed, organize and sort all the offset field fitting values, and generate three grayscale feature offset fields of size m×n. Roughness feature offset field Gradient feature offset field

[0029] Preferably, the formula for updating the weights is:

[0030] For the best matching unit q*:

[0031]

[0032] For non-optimal matching unit q:

[0033] W q (n+1)=W q (n), q ≠ q*

[0034] Where η(n) is the learning rate. Let N represent the neighborhood function of the best-matching unit q*, where N represents the dominant neighborhood of the q-th neuron. Topological distance in This indicates that the q-th neuron is matched with the best-fit unit q. * The Manhattan distance between them.

[0035] Preferably, the sliding window movement rule in step 2.9 is as follows: the sliding window starts from the coordinates (w,w) of the multi-feature map after pixel expansion and ends at the coordinates (m+w,n+w). The center point z=(x,y) traverses row by row and pixel by pixel with a step size of 1 pixel until every relevant pixel on the image has been calculated and then stops.

[0036] Preferably, the specific method for calculating multi-feature local data-driven terms includes:

[0037] Step 4.1: Offset the grayscale feature field Roughness feature offset field The gradient feature offset field is imported into the multiplicative biased field model to obtain the local energy function.

[0038]

[0039] Where j = int, rou, grad, and φ represents the level set function. Represents a multi-feature input image. Represents a multi-feature bias plot, c ij K represents the fitted values ​​of multiple features inside and outside the contour line. σ (yx) is a Gaussian kernel function with standard deviation σ, M1(φ) = H ε (φ) is the membership function, representing the region within the contour line, M2(φ) = 1 - H ε (φ) represents the area outside the contour line. M1(φ) and M2(φ) are defined as follows:

[0040]

[0041] Among them, H ε (φ) represents the approximate Heaviside function, and ε is a constant, which is 1.

[0042] Step 4.2: For the local energy function With a fixed level set function φ, respectively for c 1j and c 2j Taking the partial derivative, we get:

[0043]

[0044] Step 4.3: For the local energy function Fixed m 1j and m 2j Taking the partial derivative with respect to the level set function φ, we obtain the gradient descent flow formula.

[0045]

[0046] For the above formula Multi-feature local data driving term d local Defined as:

[0047]

[0048] Preferably, the multi-feature global data-driven term d is calculated. global The specific method is as follows:

[0049] Step 5.1: Convert the grayscale feature map Roughness feature map Gradient feature map By embedding it within the framework of a CV model, the global energy function E for multiple features is obtained. global (φ):

[0050]

[0051] Step 5.2: For With a fixed level set function φ, respectively for m 1j and m 2j Taking the partial derivative, we get:

[0052]

[0053] Step 5.3: For the above Fixed m 1j and m 2j Taking the partial derivative with respect to the level set function φ, we get:

[0054]

[0055] For the above formula Multi-feature global data driving term d global The definition is as follows:

[0056]

[0057] Preferably, an adaptive weighting function is used to construct a multi-feature hybrid data-driven term d, which combines multi-feature local data-driven terms and global data-driven terms. hybrid Specifically:

[0058]

[0059] Where A(n) is the adaptive weight function, defined as: a is the amplification gain coefficient, b is the drift coefficient, and n is the independent variable, representing the number of iterations.

[0060] Preferably, the specific method for using an adaptive regularization function to regularize the multi-feature mixed data driving term, iteratively solving the level set, and obtaining the converged zero level set as the contour line of the target boundary is as follows:

[0061] Step 7.1: Regularize the multi-feature mixed data driving term d opt :

[0062] The mixed data driver term d is regularized using the function erfreg(x). hybrid The range of values ​​is used to obtain the regularized hybrid data-driving term d. opt :

[0063] d opt =erfreg[d hybrid / γ1]=erfreg[(A(n)·d global +(1-A(n))·d local ) / γ1]

[0064] Where γ1 = 20, erfreg(x) is the adaptive regularization function, defined as: θ is the integration variable;

[0065] Step 7.2: Calculate the optimized gradient descent flow formula

[0066]

[0067] The sign and size of α can be manually adjusted to control the direction and speed of the level set's iterative evolution.

[0068] δ ε (φ) is defined as: φ represents the level set function;

[0069] Step 7.3: Solve for the zero level set using the iterative evolution formula and gradient descent:

[0070]

[0071] Where m and γ2 are positive constants, Δt represents the iteration time interval, φ(x,i) and φ(x,i-1) are the values ​​of the level set function when the iteration numbers are i and i-1, respectively; tanh(·) is the hyperbolic tangent function, M k×k (·) is a mean filter operator of size k×k; after calculation and update, return to step 4, and stop iterating until the convergence criterion is met, and output the final zero level set after convergence.

[0072] Compared with the prior art, the significant advantages of this invention are:

[0073] This invention not only considers the grayscale feature information of the image, but also additionally considers roughness feature information and gradient feature information to construct a multi-feature mixed data driving term d. hybrid This allows for a more accurate description of the details of foreground targets in infrared images.

[0074] This invention designs a method to complete the multi-feature migration field before the iterative evolution of the level set. The computational overhead of the model is greatly reduced, the image segmentation time is reduced, and the image segmentation efficiency is improved.

[0075] This invention innovatively integrates the local region self-organizing mapping algorithm into the multiplicative biased field model, constructs the energy function, and proposes a multi-feature local data-driven term.

[0076] This invention proposes an adaptive regularization function erfreg(x) to regularize the multi-feature mixed data driving term d. hybrid The energy range significantly improves segmentation accuracy and enhances the stability and robustness of image segmentation.

[0077] This invention combines the hyperbolic tangent function tanh(·) and mean filtering technique M k×k By combining (·), the sign and distance rule characteristics of the level set function are maintained during the iterative evolution of the level set function, and irrelevant contour lines at non-boundary locations are filtered out, thus improving the segmentation accuracy.

[0078] The present invention will now be further described with reference to the accompanying drawings. Attached Figure Description

[0079] Figure 1 This is a flowchart of the present invention.

[0080] Figure 2 Visualization of multi-feature offset fields.

[0081] Figure 3 This is a comparison chart of segmentation results from different models.

[0082] Figure 4 This is a comparison chart of the segmentation results before and after regularization of multi-feature mixed data-driven terms. Detailed Implementation

[0083] An infrared image segmentation method based on a local region self-organizing map (MOM) algorithm is proposed. First, an MOM-based algorithm is designed to extract multiple feature bias fields before the level set iterative evolution by moving a local sliding window. Within the framework of a multiplicative bias field model, a multi-feature local data-driven term is constructed, significantly reducing the computational overhead of the level set iterative evolution, shortening segmentation time, and improving segmentation efficiency. Second, a multi-feature global data-driven term is calculated, and a multi-feature mixed data-driven term is constructed using an adaptive weighting function. Next, an adaptive regularization function is used to regularize the energy range of the multi-feature mixed data-driven term, driving the initial level set iterative evolution, significantly improving segmentation accuracy and enhancing model stability and robustness. Then, during the level set iteration process, a hyperbolic tangent function is used to maintain the sign regularity features (positive exterior, negative interior) and distance regularity features with a modulus of 1 during iterative evolution. Simultaneously, a mean filter template is used to continuously smooth the level set and eliminate redundant non-boundary contour lines. Finally, the minimum value of the energy function is iteratively solved using gradient descent until the convergence criterion of the level set or the maximum number of iterations is reached. At this point, the position of the zero level set is output, completing image segmentation. Specific steps include:

[0084] Step 1: Calculate the multi-feature map of the infrared input image, including the grayscale feature map. Roughness feature map Gradient feature map Where x represents the coordinates of all pixels in the image, int represents the gray level, rou represents the roughness, and grad represents the gradient.

[0085] Step 1.1: Use Matlab code rgb2gray to calculate the grayscale feature map of the infrared input image.

[0086] Step 1.2: Use the formula Calculate roughness feature map Among them, w r (x,y) is a local window of size r×r located at pixel x=(x,y), μ r It is a local window w r The average gray value within (x,y).

[0087] Step 1.3: Use the formula Calculate gradient feature map

[0088] Step 2: Calculate the multi-feature offset field using a local region-based self-organizing map algorithm. Where y represents the local sliding window Ω z The coordinates of all pixels within the area.

[0089] Step 2.1: Apply the following to a multi-feature map of size m×n. At the edge positions, pixel expansion is performed using a boundary copying method. Specifically, the upper boundary row is copied upwards by w rows, the lower boundary row is copied downwards by w rows, the left boundary row is copied to the left by w columns, and the right boundary row is copied to the right by w columns, thereby generating a multi-feature map of size (m+2w)×(n+2w) after expansion.

[0090] Step 2.2: Expand the multi-feature map by three pixels each. Establish a pixel coordinate system for x and y pixels with the left top corner as the origin (0,0) to facilitate local sliding window Ω. z Feature extraction at edge locations.

[0091] Step 2.3: Set up a partial sliding window Ω z The center point is z = (x, y) and the size is (2w + 1). 2

[0092] Step 2.4: Input the multi-feature image separately In the middle, set the first local sliding window Ω z＝(w,w) And randomly select an input sample, defined as X i =[x i1 ,x i2 ,...,x ip ], and convert it using exponentiation. i is the index of the input sample, and p is the dimension of the feature. Set an appropriate initial training rate η(0) and initial dominant neighborhood.

[0093] Step 2.5: Initialize the weight vector W of the output layer neurons of the self-organizing map network using small random numbers. q =[w q1 ,w q2 ,...,w qp ], where q is the index of the neuron.

[0094] Step 2.6: Calculate X' i With W q Manhattan distance Select the best matching unit q with the smallest distance * =argmin q∈{1,2,...,p} d(X′ i W q ), and establish the optimal matching unit q * dominant neighborhood

[0095] Where x′ io For the input sample vector X' i For the value of the o-th feature on the i-th sample, w qo The weight vector W q The value of the o-th feature on the q-th neuron.

[0096] Step 2.7: Update the best matching unit q * In the topology-dominant neighborhood The association weight vector of neurons. The weight update formula is defined as:

[0097] For the best matching unit q*:

[0098]

[0099] For non-optimal matching unit q:

[0100] W q (n+1)=W q (n), q ≠ q*

[0101] Where η(n) represents the learning rate, Let N represent the neighborhood function of the best-matching unit q*, where N represents the dominant neighborhood of the q-th neuron. Topological distance in This indicates that the q-th neuron is matched with the best-fit unit q. * The Manhattan distance between them. According to h q (n,N) can determine the dominant neighborhood of the best matching unit q* with radius σ(n). This neighborhood is a Gaussian-shaped region centered on the best-matching unit q*. Neurons closer to q* are more significantly affected, while those farther away are less affected. As the number of iterations n increases, the learning rate and the dominant neighborhood... Gradually decrease.

[0102] Step 2.8: Repeat steps 2.4 to 2.7, iterating continuously until the learning rate η(n) decreases to 0. At this point, the neuron weights gradually converge to the cluster centers to represent the current multi-feature input map. Below, a local sliding window Ω at the center point z. z The corresponding multi-feature offset field fitting value b j (y)

[0103] Step 2.9: Return to step 2.4 and select the next partial sliding window Ω. z Continuing the above process (sliding window movement rules: the sliding window starts from the multi-feature map after pixel augmentation) Starting at coordinates (w, w) and ending at coordinates (m+w, n+w), the center point z = (x, y) is traversed row by row and pixel by pixel with a step size of 1 pixel until every relevant pixel in the image has been calculated. Finally, all cluster centers are sorted and arranged, generating three grayscale feature offset fields of size m×n. Roughness feature offset field Gradient feature offset field

[0104] Step 3: Initialization phase of the level set:

[0105] Step 3.1: Set the values ​​of the following parameters: Variable coefficients: α, k, n; Fixed coefficients: w = 10, σ = 1.5, γ1 = 20, γ2 = 0.2, Δt = 1, ε = 1, ε1 = 10 -3 In this context, parameter α controls the magnitude and direction of the iteration speed of the level set, parameter k controls the size of the average value filter template, parameter n represents the iteration number, and parameter w controls the local sliding window Ω. z Size, parameter σ represents the Gaussian kernel function K σ The standard deviation of (yx), parameter γ1 = 20 represents the slope of the erfreg function, parameter γ2 = 0.2 represents the slope of the tanh function, Δt = 1 represents the iteration time interval, and ε1 = 10 -3 This represents the threshold used to determine whether a level set has converged.

[0106] Step 3.2: Initialize the level set function as follows:

[0107] Step 4: Calculate the multi-feature local data driving term d local :

[0108] Step 4.1: Combine the above three offset fields Importing the multiplicative biased field model yields the local energy function.

[0109]

[0110] Among them, K σ (yx) is a Gaussian kernel function with standard deviation σ, M1(φ) and M2(φ) represent membership functions, M1(φ) = H ε (φ) represents the area within the contour line, M2(φ) = 1 - H ε (φ) represents the area outside the contour line. M1(φ) and M2(φ) are defined as follows:

[0111]

[0112] Step 4.2: For the above With a fixed level set function φ, respectively for c 1j and c 2j Taking the partial derivative, we get:

[0113]

[0114] Step 4.3 For the above Fixed m 1j and m 2j Taking the partial derivative with respect to the level set function φ, we get:

[0115]

[0116] Among them, the multi-feature local data driving term d local Defined as follows

[0117]

[0118] Step 5: Calculate the multi-feature global data-driven term d global :

[0119] Step 5.1: Convert the grayscale feature map Roughness feature map Gradient feature map Embedded within the framework of a CV model,

[0120]

[0121] Step 5.2: For the above With a fixed level set function φ, respectively for m 1j and m 2j Taking the partial derivative, we get:

[0122]

[0123] Step 5.3: For the above Fixed m 1j and m 2j Taking the partial derivative with respect to the level set function φ, we get:

[0124]

[0125] Among them, the multi-feature local data driving term d global The definition is as follows:

[0126]

[0127] Step 6: Using an adaptive weighting function, combine local and global data-driven terms of multiple features to construct a multi-feature mixed data-driven term, d. hybrid Specifically:

[0128]

[0129] Where A(n) is the adaptive weight function, defined as: a is the amplification gain coefficient, and b is the drift coefficient. The entire adaptive weighting function divides the level set evolution into three stages: In this paper, a = 3, and when b = 22, in the first stage, when the number of iterations n increases from 0 to 20, A(n) stabilizes near 0, i.e., the multi-feature local data driving term d... local The first stage is dominant, accurately locating and identifying the target's boundary information, guiding the level set to evolve towards the target's boundary. This process may produce local minima when encountering situations where there is uneven grayscale within the target. In the second stage, as the number of iterations n increases from 21 to 25, A(n) rapidly transitions from 0 to 1. At this point, the multi-feature global and local data-driven term d... global and d local Together, they drive the evolution curve toward the target boundary. In the third stage, when the number of iterations n is greater than 25, A(n) stabilizes around 1, and the multi-feature global data drives the term d. global It dominates, eliminates the local minima that may be generated in the first stage, and drives the evolution curve to continue to converge toward the global optimal boundary, thus obtaining the final zero-level set.

[0130] Step 7: After regularizing the energy range of the multi-feature mixed data driving term using an adaptive regularization function, iteratively solve the level set to obtain the convergent zero level set as the contour line of the target boundary, thus completing the contour segmentation.

[0131] Step 7.1: Regularize the multi-feature mixed data driving term d opt :

[0132] Due to the significant differences in contrast between different input images, the multi-feature mixed data driving term d hybrid Uneven distribution of energy values ​​can easily lead to erroneous segmentation results, greatly reducing the robustness and accuracy of the model.

[0133] Therefore, the mixed data driving term d is regularized using the function erfreg(x). hybrid The range of values ​​is determined, and a regularized hybrid data-driven term d is proposed. opt :

[0134] d opt =erfreg[d hybrid / γ1]=erfreg[(A(n)·d global +(1-A(n))·d local ) / γ1]

[0135] Where γ1 = 20, erfreg(x) is the adaptive regularization function, defined as: θ is the integration variable.

[0136] Step 7.2: Calculate the optimized gradient descent flow formula

[0137]

[0138] The sign and size of α can be manually adjusted to control the direction and speed of the level set's iterative evolution, δ ε (φ) is defined as:

[0139] The essence of level set evolution is that the initial level set (usually a rectangular box within an input image) is driven by data-driven terms and iteratively evolved through gradient descent. During this process, the initial level set continuously shrinks or expands, approaching the target boundary. When the level set satisfies the convergence criterion, the zero level set (i.e., the target boundary) is obtained.

[0140] Step 7.3: Solve for the zero level set using the iterative evolution formula and gradient descent:

[0141]

[0142] Where m and γ2 are positive constants, Δt represents the iteration time interval, φ(x,i) and φ(x,i-1) are the values ​​of the level set function when the iteration numbers are i and i-1, respectively; tanh(·) is the hyperbolic tangent function, used to normalize the range of the level set function so that it maintains the sign property of negative inside and positive outside and the distance property of always having a modulus of 1 during the iteration evolution of the level set. k×k (·) is a mean filter operator of size k×k, used to continuously smooth the evolution curve, maintain its smoothness, and filter out irrelevant contour lines outside the boundary, thereby improving the segmentation efficiency.

[0143] After the calculation and update, return to step 4 and continue iterating until the convergence criterion is met. Then output the final zero level set after convergence.

[0144] The convergence criteria for the iterative evolution of the level set are: (1) If the number of iterations reaches the maximum number of iterations n, the iteration stops and the partitioning result is output, regardless of whether the level set has converged at this time; (2) If the difference between the level sets satisfies the convergence criterion: |φ(x,i)-φ(x,i-1)|≤ε1, then ε1=10 -3 Regardless of whether the maximum number of iterations n has been reached, the iteration stops and the segmentation result is output. Finally, the final converged zero-level set is used as the contour line of the target boundary, completing the contour segmentation task.

[0145] Example 1

[0146] To demonstrate the feasibility of the proposed local region self-organizing mapping algorithm, computer experiments were conducted to verify the performance of the proposed model, and comparative experiments were performed with two similar models based on a combination of global and local approaches.

[0147] A. Experimental Environment

[0148] The experimental environment for this invention was Matlab 2022a, with an AMD Ryzen R7 5800H 3.2GHz CPU, 40GB of RAM, and an NVIDIA GeForce RTX 3060 Laptop 6GB GPU. In all the segmentation results, the green box represents the initial level set, and the red curve represents the final zero level set.

[0149] B. Experimental parameters:

[0150] GLSEPF model: c0=2, ε=1.5, λ1=λ2=1, μ=10, σ=(4~20)

[0151] GLMF model: c0=1, λ=0.3, α=400, ε=1, w=5, σ1=1, σ2=3

[0152] The model of this invention:

[0153] C. Experimental Results:

[0154] (1) Segmentation Experiment

[0155] Three infrared images with uneven grayscale were selected, and the model proposed in this invention was used to extract the contours of foreground targets from these three images. For example... Figure 2 As shown, the first column is the grayscale image with the initial level set, and the second column is the grayscale feature offset field. The second column is the roughness feature offset field. The third column is the gradient feature offset field. The fourth column shows the final segmentation result with zero level (i.e., the target boundary location). Furthermore, Table 1 displays the segmentation time and segmentation accuracy (IOU / DSC) for these three images, demonstrating that the proposed local region self-organizing map-based algorithm can extract the contours of foreground targets from unevenly grayscale infrared images with less time consumption and higher accuracy.

[0156] Table 1 shows the segmentation time and accuracy of the model of the present invention for three infrared images.

[0157] Image number Segmentation time (seconds) Segmentation precision (IOU / DSC) 1 2.306 0.953 / 0.976 2 1.121 0.952 / 0.975 3 1.492 0.978 / 0.989

[0158] (2) Comparative Experiment

[0159] Five infrared images with uneven grayscale were selected, and segmentation was performed on the five images using the Global and Local Signed Energy-based Pressure Force (GLSEPF) model, the Global and local multi-feature fusion model (GLMF) model, and the model of this invention, respectively. The segmentation results were then compared. Figure 3 As shown, the first column is the original image with the initial level set. The second column is the visualized segmentation result of the GLSEPF model, the third column is the visualized segmentation result of the GLMF model, and the fourth column is the visualized segmentation result of the model proposed in this invention. For Figure 3 In terms of visualization results, the model of this invention can produce a clear and accurate red zero-level set to characterize the boundary position of the target, while the other two models produce inaccurate or even failed segmentation results.

[0160] Furthermore, Tables 2 and 3 summarize the segmentation time and segmentation accuracy of the three models mentioned above. The two metrics used to evaluate segmentation accuracy, the Jaccard similarity coefficient (JSC) and the Dice similarity coefficient (DSC), are defined as follows: Where G0 represents the foreground region extracted after segmentation by each model. This represents the foreground region of the ground truth map. The evaluation metrics JSC and DSC range from 0 to 1. The closer their values ​​are to 1, the higher the similarity between the foreground regions and the higher the segmentation accuracy; conversely, the lower the similarity, the lower the segmentation accuracy. On average, the model of this invention is superior to the other two models in terms of segmentation time and segmentation accuracy.

[0161] Table 2. Segmentation time (seconds) for the three models. Bold text indicates the optimal value.

[0162]

[0163]

[0164] Table 3. Segmentation accuracy (JSC / DSC) of the three models. Bold text indicates the best value.

[0165]

[0166] Example 2

[0167] This invention proposes an adaptive regularization function. This function can regularize the original multi-feature mixed data driver term d. hybrid The value range is used to generate the optimized multi-feature mixed data driving term d. opt This uniformly confines the originally fluctuating energy range to between -1 and 1, such as... Figure 4 As shown, the first and second columns represent the original image and the ground truth image with initial contours, respectively, and the third column represents the term d driven by the original multi-feature mixed data. hybrid The fourth column shows the 3D visualization of the corresponding final level set, and the fifth column shows the segmentation result driven by the optimized multi-feature mixed data driving term d. opt The fifth column shows the 3D visualization of the final level set driven by the segmentation. The red curve in the 3D visualization represents the zero level set. Therefore... Figure 4 It can be seen that the multi-feature mixed data driving term d hybrid The energy range is normalized by the adaptive regularization function erfreg(x), which eliminates the problem of getting trapped in false targets, thereby improving the segmentation efficiency and stability.

Claims

1. An infrared image segmentation method based on a local region self-organizing mapping algorithm, characterized in that, Comprising: Step 1: calculating multi-feature maps of the infrared input image, the multi-feature maps comprising a gray-scale feature map, a roughness feature map, and a gradient feature map; Step 2: calculating multi-feature bias fields using a local region self-organizing mapping algorithm; Step 3: initializing a level set; Step 4: calculating a multi-feature local data-driven term; Step 5: calculating a multi-feature global data-driven term; Step 6: Using the adaptive weight function, combine the multi-feature local data-driven term and the global data-driven term to construct a multi-feature hybrid data-driven term d hybrid Specifically: wherein A(n) is an adaptive weight function defined as: a is an amplification gain coefficient, b is a drift coefficient, n is an independent variable, indicating the number of iterations, d global is a multi-feature global data-driven term, d local is a multi-feature local data-driven term; int represents a gray scale, rou represents roughness, and grad represents a gradient; Step 7: using an adaptive regularization function to regularize the energy value range of the multi-feature hybrid data-driven term, iteratively solving the level set, obtaining a converged zero level set as the contour line of the target boundary, completing contour segmentation, and the specific method is: Step 7.1: Regularizing the multi-feature hybrid data-driven term d opt : The mixed data driven term d is regularized by the function erfreg(x) hybrid The value range of the mixed data driven term d opt is obtained d opt = erfreg [d hybrid / γ1] = erfreg [(A(n) · d global + (1 - A(n)) · d local / γ1] where γ1= 20, erfreg(x) is an adaptive regularization function defined as: θ is the integral variable; Step 7.2: Calculate the optimized gradient descent flow formula Wherein, the alpha symbol and size can be manually adjusted, control the direction and speed size of the level set iteration evolution, and δ(φ) is defined as: ε = 1, and φ represents the level set function; Step 7.3: using an iterative evolution formula and solving the zero level set by a gradient descent method: where m and γ2 are normal numbers, Δt represents the iteration time interval, φ(x, i), φ(x, i-1) are the values of the level set function when the iteration number is i, i-1 respectively; tanh(·) is the hyperbolic tangent function, M k×k (·) is a mean filter operator with size k x k; after the calculation is updated, return to step 4, stop iteration until the convergence criterion is met, and output the final zero level set after convergence. 2.The local region self-organizing map algorithm based infrared image segmentation method according to claim 1, characterized in that, The roughness feature map is specifically: wherein w r (x,y) is a local window of size r x r at pixel point x = (x,y), μ r is the average gray value within the local window w r (x,y), is the gray feature map. 3.The local region self-organizing map algorithm based infrared image segmentation method according to claim 1, characterized in that, Gradient feature map Specifically: where G x (x,y), G y (x,y) are the horizontal and vertical gradients of the gray scale feature map respectively.

4. The method according to claim 1, wherein, The specific method for calculating the multi-feature bias field using a local region self-organizing mapping algorithm is: Step 2.1: Pixel expansion is performed in the edge position of the multi-feature map with the size of m x n respectively by the method of boundary replication with the size of w x w; Step 2.1: Pixel expansion is performed in the edge position of the multi-feature map with the size of m x n respectively by the method of boundary replication with the size of w x w; Step 2.2: the multi-feature maps of size (m+2w, n+2w) respectively after three-pixel padding The x-y pixel coordinate system is established with the left top corner of the multi-feature map as the coordinate origin (0, 0). Step 2.3: Set local sliding window Ω z : center point is z = (x,y) and size is (2w+1) 2 ; Step 2.4: Set the first local sliding window Ω in the multi-feature input image z＝(w,w) and randomly select the input sample vector, defined as X i = [x i1 , x i2 ,..., x ip ], and use the power transformation to v = 0.5, i is the index of the input sample, p is the dimension of the feature, and set the appropriate initial learning rate η(0) = 1 and the initial dominant neighborhood Step 2.5: Initialize the weight vectors W of the output layer neurons of the self-organizing map network using the random numbers q = [w q1 , w q2 ,..., w qp ], q is the index of the neuron; Step 2.6: Compute Manhattan distance d(X', W) between X' and W i q x i ' o i qo q * q∈{1,2,...,p} i q * ​​​​​​​​​​​ Step 2.7: Update best matching unit q * In topologically dominant neighborhood The associated weight vector of the central neuron; Step 2.8: repeating steps 2.4 to 2.7 until the learning rate η(n) decays to 0, and obtaining the corresponding multi-feature bias field fitting value in the current local sliding window; Step 2.9: Return to Step 2.4 and continue the process with the next local sliding window Ω z Continue the above process until all relevant pixels in the input image are traversed, and all the bias field fitting values are sorted and ordered to generate three gray-scale feature bias fields with size m x n, respectively Roughness feature bias field Gradient feature bias field 5. The method according to claim 4, wherein, The weight update formula is: For the best matching unit q*: q = q* For the non-best matching unit q: W q (n+1) = W q (n), q≠q* where η(n) is the learning rate, is a neighborhood function that indicates the best matching unit q*, N indicates the topological distance of the qth neuron in the dominant neighborhood , is the Manhattan distance between the qth neuron and the best matching unit q * .

6. The method according to claim 4, wherein, The moving rule of the local sliding window in step 2.9: the local sliding window starts from the coordinates (w, w) of the pixel expanded multi-feature map and ends at the coordinates (m+w, n+w), and the center point z=(x, y) traverses row by row and pixel by pixel with a step size of 1 pixel.

7. The local region self-organizing map algorithm based infrared image segmentation method according to claim 1, characterized in that, The specific method for calculating the multi-feature local data-driven term includes: Step 4.1: Shift the grayscale feature field by Roughness feature shift field Gradient feature shift field where j = int, rou, grad, and φ represents the level set function, represents the multi-feature input image, represents the multi-feature bias field image, c ij represents the multi-feature fitting value inside and outside the contour line, K σ (y-x) is a Gaussian kernel function with a standard deviation of σ, M1(φ) = H ε (φ) is a membership function, representing the area inside the contour line, M2(φ) = 1-H ε (φ) represents the area outside the contour line, M1(φ), M2(φ) are defined as shown below: M2(φ) = 1 - M1(φ) where H ε (φ) denotes the approximate Heaviside function and ε is a constant, taken to be 1. Step 4.2: For the local energy function Fixing the level set function φ, we solve for c 1j and c 2j Taking the partial derivative gives: Step 4.3: For the local energy function Fix m 1j and m 2j Taking the partial derivative of the level set function φ gives the gradient descent flow equation For the above equation The multi-feature local data-driven term d local is defined as:

8. The method according to claim 1, wherein, Computing multi-feature global data-driven term d global The specific method is: Step 5.1: Obtain the gray scale feature map Roughness feature map Gradient feature map In the framework of embedding the CV model, obtain the multi-feature global energy function E global (φ) Step 5.2: For Fix the level set function φ, and solve m 1j and m 2j The partial derivative is: Step 5.3: For the above Fix m 1j and m 2j The partial derivative of the level set function φ gives: For the above formula The multi-feature global data driven term d global is defined as follows: 。

Images