Mixed level set image segmentation method based on distance regularization item and local feature enhancement

Abstract

The invention discloses a mixed level set image segmentation method based on a distance regularization item and local feature enhancement. The method comprises the following steps: acquiring an image to be segmented; constructing a local energy item, introducing a local variance into the local energy item, and improving the local energy item by using an internal and external local gray average value of a local region evolution curve; constructing a global energy item, constructing a weighting function by using the local gray average value and the global gray average value, and respectively applying the weighting function to the global energy item and the local energy item; a boundary item is constructed, an edge stop function is constructed by using a canny operator and an image convolution result, a reasonable order is selected by using an adaptive method, and a high-order boundary operator is applied to the boundary item; constructing a distance regularization term based on a logarithm and a polynomial, and introducing an edge indicator function into the distance regularization term; constructing an overall level set energy function; and finally, minimizing the total level set energy function by using a gradient flow descent method to obtain an image segmentation result.

Description

Technical Field

[0001] The invention relates to a hybrid level set image segmentation method based on a distance regularization term and local feature enhancement, and belongs to the technical field of image segmentation. Background Art

[0002] Image segmentation is an important step in image processing. Image segmentation divides an image into different regions according to its grayscale, texture and other features. Image segmentation based on the level set active contour model is widely used. Most level set image segmentation algorithms based on boundary mapping rely on image gradients. When the edges of the image are blurred or there is discontinuity in the gradient information, the level set performance is not good. The region-based level set image segmentation model can generally only focus on the local information or global information of the image. In order to overcome this defect, hybrid models have been proposed. Common hybrid models include hybrid models based on global and local region information and hybrid models based on edge and region information. Studies have shown that compared with a single model, the hybrid model can segment more complex images, has better robustness and noise resistance, and has better segmentation effects.

[0003] Level set image segmentation based on a hybrid model of global energy terms, local energy terms, boundary energy terms and distance regularization terms can obtain more comprehensive recognition of image regional information and edge information, avoid repeated initialization, and is expected to more accurately segment the background and target in the image. Summary of the invention

[0004] The purpose of the present invention is to address the defects and shortcomings of the above-mentioned prior art and propose a hybrid level set image segmentation method based on distance regularization term and local feature enhancement. The method uses global energy term, local energy term, boundary energy term and distance regularization term to form a hybrid model, uses global information and local information to form a weight function, and applies the weight function to the global term and the local term respectively. At the same time, the local variance is introduced into the local term to improve the grayscale mean of the local area in the local term. An adaptive high-order boundary operator is introduced into the boundary energy term, and a distance regularization term is formed by using polynomials and logarithms, and an edge indicator function is introduced into the distance regularization term. The method reduces the sensitivity of the level set initialization parameters, avoids repeated initialization, and effectively alleviates the grayscale unevenness, thereby enhancing the model's ability and robustness to segment grayscale uneven images.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] A hybrid level set image segmentation method based on distance regularization term and local feature enhancement, the method comprising:

[0007] Obtaining an image to be segmented;

[0008] Construct local energy terms, introduce local variance into the local energy terms, and improve the local energy terms by using the local grayscale means inside and outside the evolution curve;

[0009] Construct a global energy term, use the global grayscale mean and the local grayscale mean to construct a weight function, and apply the weight function to the global energy term and the local energy term respectively, which reduces the grayscale unevenness of the image;

[0010] Construct boundary terms, use the result of canny operator and image convolution to construct edge stop function, use adaptive method to select reasonable order, and apply high-order boundary operators to boundary terms;

[0011] Construct a distance regularization term based on logarithm and polynomial, and introduce the edge indicator function into the distance regularization term;

[0012] The overall level set energy function is constructed by combining the global energy term, local energy term, boundary term and distance regularization term;

[0013] Finally, the gradient flow descent method is used to minimize the overall level set energy function to obtain the image segmentation result.

[0014] Furthermore, the local variance is introduced into the local energy term, and the local grayscale mean inside and outside the evolution curve is used to improve the local energy term, including:

[0015] x i1 Indicates the gray value of the current pixel inside the curve. Indicates the current pixel x in the curve i1 The grayscale mean of the neighborhood centered at x i2 Represents the gray value of the current pixel outside the curve, Indicates the current pixel x outside the curve i2 The grayscale mean of the neighborhood centered at x. i is the gray value of the pixel with x as the neighborhood, Indicates the grayscale average of the pixels in the neighborhood of x. n1 indicates the grayscale average of the pixels in the neighborhood of the current x. i1 is the number of pixels in the evolution curve of the neighborhood, n2 represents the number of pixels in the neighborhood at the current x i2 is the number of pixels in the evolution curve of the neighborhood. σ(x) represents the variance of the local area centered on the current x, σ1(x) represents the variance inside the evolution curve centered on x, and σ2(x) represents the variance outside the evolution curve centered on x. Then the expressions of σ(x), σ1(x), and σ2(x) are:

[0016]

[0017] Use e3 and e4 to represent σi The difference between (x) and σ(x)

[0018] e3=σ1(x)-σ(x)

[0019] e4=σ2(x)-σ(x)

[0020] Use σ(x) to generate an adaptive function b and adjust the range of the neighborhood according to the size of the variance.

[0021]

[0022] Represents the gradient.

[0023] The local grayscale means inside and outside the curve evolution contour are defined as:

[0024]

[0025] f1(x) represents the local mean inside the curve profile, and f2(x) represents the local mean outside the curve profile.

[0026]

[0027] H represents a smooth unit step (Heaviside) function, and H is expressed as:

[0028]

[0029] K σ represents the Gaussian kernel function, and σ represents the Gaussian kernel.

[0030]

[0031] Introducing the adaptive function b into the local energy term, the local energy term is expressed as:

[0032] E L (C, f1(x), f2(x))

[0033] =λ1∫ in K σ (xy)(I(y)-f3(x)) 2 dy+λ2∫ out K σ (xy)(I(y)-f4(x)) 2 dy

[0034] +b(λ3(σ1(x)-σ(x))+λ4(σ2(x)-σ(x)))

[0035] λ1, λ2, λ3, λ4 represent various coefficients. K σ represents the Gaussian kernel function.

[0036]

[0037] Furthermore, a weight function is constructed using the global grayscale mean and the local grayscale mean, and the weight function is applied to the global energy term and the local energy term, respectively, including:

[0038] Generate weight function

[0039]

[0040] Where p is the parameter controlling the weight, t1(x) and t2(x) represent the mean of the global information inside and outside the contour, and l1(x) and l2(x) represent the mean of the local information inside and outside the contour:

[0041] t1=mean{I(x)|φ(x)>0,x∈Ω}

[0042] t2=mean{I(x)|φ(x)<0,x∈Ω}

[0043] l1(x)=mean({I(x)|φ(x)|>0}∩W k (x),x∈Ω)

[0044] l2(x)=mean({I(x)|φ(x)<0}∩W k (x), x∈Ω)

[0045] Where W k It is a Gaussian rectangular window with a standard deviation of 3 and a scale of 13*13.

[0046] The weight function is applied to the global energy term and the local energy term respectively, and the weight function is applied to the grayscale non-uniform image.

[0047] q(x)=ω(x)·I(x)

[0048] m1(x) and m2(x) are the internal and external fitting information of the region, respectively. The weight function is applied to the grayscale mean inside and outside the evolution curve:

[0049]

[0050] Where c1 and c2 are represented by

[0051]

[0052] H represents a smooth unit step (Heaviside) function, and H is expressed as:

[0053]

[0054] The global term is then expressed as

[0055]

[0056] At the same time, the weight function is applied to the local energy term (1-ω)E L , adjust the grayscale value of the local area.

[0057] Then the global energy term plus the local energy is expressed as:

[0058]

[0059] Among them, λ1, λ2, λ3, λ4, λ5, and λ6 are coefficients of each item.

[0060] Furthermore, the edge stop function is constructed using the result of the canny operator and the image convolution, and a reasonable order is selected using an adaptive method to apply high-order boundary operators to the boundary terms, including:

[0061] The canny operator is used to convolve the image to enhance edge detection; the edge stop function is constructed using the convolution result.

[0062]

[0063] in is the gradient operator, G σ represents the canny operator, ρ controls the speed of curve evolution. The larger the ρ, the faster the curve evolves. α is the sensitivity of the edge index to noise.

[0064] Furthermore, an adaptive method is used to select a reasonable order α, and the sensitivity to noise is reduced while the curve evolution speed is increased by setting parameters, including:

[0065] Define the adaptive selection order:

[0066]

[0067] α is an integer, and the adaptive selection indicates that the order of the edge stop function starts from 1 and reaches α. When (G α+1 -G α +G α +G α-1 +…+G2-G1) / N is less than the threshold τ, α is taken as the order of the edge stop function, N is the number of pixels, and τ is a threshold.

[0068] Higher-order boundary operators are applied to the boundary energy terms:

[0069]

[0070] Furthermore, a distance regularization term is generated and the edge indicator function is introduced into the regularization term to avoid repeated initialization of the level set, including:

[0071] Constructing potential function penalty terms using logarithms, polynomials, and constants

[0072]

[0073] Represents the gradient.

[0074] The Gaussian kernel function is convolved with the image to generate an edge indication function.

[0075]

[0076] G σ represents the Gaussian kernel function.

[0077] Introducing the edge indicator function into the energy penalty term can smooth the image and reduce the impact of noise. The distance regularization term is expressed as:

[0078]

[0079] Furthermore, the level set energy term is generated and calculated using the gradient descent method to minimize the energy function and achieve image segmentation, including:

[0080] Combine global terms, local terms, boundary terms, and distance regularization terms.

[0081]

[0082] λ1, λ2, λ3, λ4, λ5, λ6, μ, ν, p are the parameters of each item

[0083] The overall energy function is calculated using the gradient flow descent method to minimize the energy function and obtain the image evolution curve.

[0084] Take λ1=λ2=λ3=λ4=λ5=λ6=1

[0085] Let I(y)-f3(x)=e1, I(y)-f4(x)=e2

[0086]

[0087] Where div is the divergence and δ is the Dirac function, which can be obtained by taking the derivative of H(φ):

[0088]

[0089] Beneficial effects:

[0090] 1. The present invention proposes a hybrid level set image segmentation method based on distance regularization term and local feature enhancement. This method constructs a distance regularization term that can better avoid level set initialization, and at the same time balances the uneven grayscale of the level set image through a weight function.

[0091] 2. The present invention adopts global terms, local terms, boundary terms and distance regularization terms, and adopts high-order adaptive boundary operators in boundary terms and introduces local variance in local terms, which can avoid the influence of noise on image segmentation and enhance the ability of level set to perform image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 It is an overall flow chart of a hybrid level set image segmentation method based on a distance regularization term and local feature enhancement according to an embodiment of the present invention.

[0093] Figure 2 It is a schematic diagram of the structure of global items and local items in an embodiment of the present invention.

[0094] Figure 3 4 is a schematic diagram of the structure of the edge item of the embodiment of the present invention.

[0095] Figure 4 Schematic diagram of the structure of the distance regularization term of an embodiment of the present invention. DETAILED DESCRIPTION

[0096] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0097] The present invention provides a hybrid level set image segmentation method based on distance regularization term and local feature enhancement, comprising:

[0098] Obtaining an image to be segmented;

[0099] Construct local energy terms, introduce local variance into the local energy terms, and improve the local energy terms by using the local grayscale means inside and outside the evolution curve;

[0100] Construct a global energy term, use the global grayscale mean and the local grayscale mean to construct a weight function, and apply the weight function to the global energy term and the local energy term respectively, which reduces the grayscale unevenness of the image;

[0101] Construct boundary terms, use the result of canny operator and image convolution to construct edge stop function, use adaptive method to select reasonable order, and apply high-order boundary operators to boundary terms;

[0102] Construct a distance regularization term based on logarithm and polynomial, and introduce the edge indicator function into the distance regularization term;

[0103] The overall level set energy function is constructed by combining the global energy term, local energy term, boundary term and distance regularization term;

[0104] Finally, the gradient flow descent method is used to minimize the overall level set energy function to obtain the image segmentation result.

[0105] like Figure 1 As shown, the hybrid level set image segmentation method based on distance regularization term and local feature enhancement provided in this embodiment includes the following steps:

[0106] Step 1) Obtain the image to be segmented.

[0107] Step 2) If the image is a color image, convert the image into a grayscale image.

[0108] Step 3) In this embodiment, local variance is introduced into the local energy term, and the grayscale mean of the local energy term is improved by using the grayscale mean of the local area inside and outside the contour.

[0109] Introduce local variance in the local energy term. i1 Indicates the gray value of the current pixel inside the curve. Indicates the current pixel x in the curve i1 The grayscale mean of the neighborhood centered at x i2 Represents the gray value of the current pixel outside the curve, Indicates the current pixel x outside the curve i2 The grayscale mean of the neighborhood centered at x. i is the gray value of the pixel with x as the neighborhood, Indicates the grayscale average of the pixels in the neighborhood of x. n1 indicates the grayscale average of the pixels in the neighborhood of the current x. i1 is the number of pixels in the evolution curve of the neighborhood, n2 represents the number of pixels in the neighborhood at the current x i2 is the number of pixels in the evolution curve of the neighborhood. σ(x) represents the variance of the local area centered on the current x, σ1(x) represents the variance inside the evolution curve centered on x, and σ2(x) represents the variance outside the evolution curve centered on x. Then the expressions of σ(x), σ1(x), and σ2(x) are:

[0110]

[0111] Use e3 and e4 to represent σ i The difference between (x) and σ(x)

[0112] e3=σ1(x)-σ(x)

[0113] e4=σ2(x)-σ(x)

[0114] Use σ(x) to generate an adaptive function b and adjust the range of the neighborhood according to the size of the variance.

[0115]

[0116] Represents the gradient.

[0117] The local area energy term is improved by using the grayscale mean of the local area inside and outside the contour. The local grayscale mean f3(x) and f4(x) in the local term are defined as

[0118]

[0119] Among them, f1(x) and f2(x) are the grayscale means of the local areas inside and outside the contour, respectively.

[0120]

[0121] In the formula, K σ represents the Gaussian kernel function.

[0122]

[0123] H represents a smooth unit step (Heaviside) function, and H is expressed as:

[0124]

[0125] The adaptive function is introduced into the local variance, and the local energy term is constructed according to the improved local grayscale mean:

[0126] E L (C, f1(x), f2(x)) = λ1∫ in K σ (xy)(I(y)-f3(x)) 2 dy+λ2∫ out K σ (xy)(I(y)-f4(x)) 2 dy+b(λ3(σ1(x)-σ(x))+λ4(σ2(x)-σ(x)))

[0127] Step 4) In this embodiment, the weight function is composed of the mean of the global information inside and outside the contour and the mean of the local grayscale inside and outside the contour.

[0128]

[0129] Where p is the parameter controlling the weight, t1 and t2 represent the mean of the global information inside and outside the contour, and l1 and l2 represent the mean of the local information inside and outside the contour:

[0130] t1=mean{I(x)|φ(x)>0,x∈0}

[0131] t2=mean{I(x)|φ(x)<0,x∈Ω}

[0132] l1(x)=mean({I(x)|φ(x)>0}∩W k (x), x∈Ω)

[0133] l2(x)=mean({I(x)|φ(x)<0}∩W k (x),x∈Ω)

[0134] Where W k It is a Gaussian rectangular window with a standard deviation of 3 and a scale of 13*13.

[0135] Apply the weight function to the global term and the local term separately.

[0136]

[0137] Among them, λ1, λ2, λ3, λ4, λ5, and λ6 are coefficients of each item.

[0138] q(x) represents the grayscale mean of the image after the weight is applied

[0139] q(x)=ω(x)·I(x)

[0140] m1 and m2 are the inner and outer fitting information of the region respectively:

[0141]

[0142] c1 and c2 are the global grayscale means inside and outside the curve of the original grayscale image without weighting.

[0143]

[0144] Step 5) In this embodiment, the canny operator image is convolved, and the result of the convolution is used to set the edge stop function.

[0145]

[0146] G σ Represents the canny operator, ρ controls the speed of curve evolution. The larger ρ is, the faster the curve evolves. α is the sensitivity of the edge index to noise.

[0147] Use an adaptive method to select a reasonable order α as follows:

[0148]

[0149] α is an integer, and the adaptive selection indicates that the order of the edge stop function starts from 1 and reaches α. When (G α+1 -G α +G α +G α-1 +…+G2-G1) / N is less than the threshold τ, α is taken as the order of the edge stop function. N is the number of pixels, and τ is a threshold.

[0150] Apply higher-order boundary operators to the boundary terms of the level set energy terms.

[0151]

[0152] Step 6) Use logarithms and polynomials to establish distance regularization terms.

[0153] Combining potential function penalties using logarithmic, polynomial, and constant

[0154]

[0155] Introducing the edge indication function into the energy penalty term can smooth the image and reduce the impact of noise.

[0156]

[0157] G σ represents the Gaussian kernel function.

[0158] Then the distance regularization term is expressed as:

[0159]

[0160] Step 7) Combine the global term, local term, edge term and distance regularization term.

[0161]

[0162] λ1, λ2, λ3, λ4, λ5, λ6, μ, ν, p are the parameters of each item

[0163] Step 8) In order to minimize the energy function, the gradient descent method is used to evolve the segmentation curve:

[0164] Take λ1=λ2=λ3=λ4=λ5=λ6=1

[0165] Let I(y)-f3(x)=e1, I(y)-f4(x)=e2

[0166]

[0167] Where div is the divergence and δ is the Dirac function, which can be obtained by taking the derivative of H(φ):

[0168]

[0169] When the curve evolution stops, the segmentation curve of the image can be obtained.

[0170] The hybrid level set image segmentation method based on distance regularization term and local feature enhancement proposed in the present invention can effectively reduce the sensitivity of level set initialization to parameters, increase the segmentation effect of the model for images with uneven grayscale, and thus improve the accuracy of image segmentation, compared with the general level set model.

[0171] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0172] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0173] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0174] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0175] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A hybrid level set image segmentation method based on distance regularization term and local feature enhancement, characterized in that: include: Obtaining an image to be segmented; Construct local energy terms, introduce local variance into the local energy terms, and improve the local energy terms by using the local grayscale means inside and outside the evolution curve; Construct a global energy term, use the global grayscale mean and the local grayscale mean to construct a weight function, and apply the weight function to the global energy term and the local energy term respectively, which reduces the grayscale unevenness of the image; Construct boundary terms, use the result of canny operator and image convolution to construct edge stop function, use adaptive method to select reasonable order, and apply high-order boundary operators to boundary terms; Construct a distance regularization term based on logarithm and polynomial, and introduce the edge indicator function into the distance regularization term; The overall level set energy function is constructed by combining the global energy term, local energy term, boundary term and distance regularization term; Finally, the gradient flow descent method is used to minimize the overall level set energy function to obtain the image segmentation result.

2. The hybrid level set image segmentation method based on distance regularization term and local feature enhancement according to claim 1, characterized in that: The local variance is introduced into the local energy term, and the local grayscale mean inside and outside the evolution curve is used to improve the local energy term, including: x i1 Indicates the gray value of the current pixel inside the curve. Indicates the current pixel x in the curve i1 The grayscale mean of the neighborhood centered at x i2 Represents the gray value of the current pixel outside the curve, Indicates the current pixel x outside the curve i2 The grayscale mean of the neighborhood centered at x i is the gray value of the pixel with x as the neighborhood, represents the grayscale average of the pixels in the neighborhood of x, and n1 represents the grayscale average of the pixels in the neighborhood of the current x. i1 is the number of pixels in the evolution curve of the neighborhood, n2 represents the number of pixels in the neighborhood at the current x i2 is the number of pixels in the evolution curve of the neighborhood, σ(x) represents the variance of the local area centered on the current x, σ1(x) represents the variance inside the evolution curve centered on x, and σ2(x) represents the variance outside the evolution curve centered on x. Then the expressions of σ(x), σ1(x), and σ2(x) are: Introduce local xi and use e3 and e4 to represent σ i The difference between (x) and σ(x) e3=σ1(x)-σ(x) e4=σ2(x)-σ(x) Use σ(x) to generate an adaptive function b and adjust the range of the neighborhood according to the size of the variance; represents the gradient; Define the local grayscale mean inside and outside the contour, respectively: f1(x) represents the local mean inside the curve contour, and f2(x) represents the local mean outside the curve contour; H represents a smooth unit step (Heaviside) function, and H is expressed as: K σ represents the Gaussian kernel function; Then the local energy term is expressed as: E L (C,f1(x),f2(x))=λ1∫ in K σ (x-y)(I(y)-f3(x)) 2 dy+λ2∫ 0ut K σ (x-y)(I(y)-f4(x)) 2 dy+b(λ3(σ1(x)-σ(x))+λ4(σ2(x)-σ(x))) λ1, λ2, λ3, and λ4 represent various coefficients.

3. The hybrid level set image segmentation method based on distance regularization term and local feature enhancement according to claim 1, characterized in that: The weight function is constructed using the global grayscale mean and the local grayscale mean, and the weight function is applied to the global energy term and the local energy term respectively, including: Generate weight function Where p is the parameter controlling the weight, t1(x) and t2(x) represent the mean of the global information inside and outside the contour, and l1(x) and l2(x) represent the mean of the local information inside and outside the contour: t1=mean{I(x)φ(x)>0,x∈Ω} t2=mean{I(x)|0(x)<0,x∈Ω} l1(x)=mean({I(x)|φ(x)>0}∩W k (x),x∈Ω) l2(x)=mean({I(x)|φ(x)<0}∩W k (x),x∈Ω) Where W k It is a Gaussian rectangular window with a standard deviation of 3 and a scale of 13*13; The weight function is applied to the global energy term and the local energy term respectively, and the weight function is interacted with the grayscale inhomogeneous image in the global energy term and the grayscale mean inside and outside the evolution curve respectively; q(x)=ω(x)·I(x) m1(x) and m2(x) are the inner and outer fitting information of the region respectively: Where c1 and c2 are represented by H represents a smooth unit step (Heaviside) function, and H is expressed as: The global term is then expressed as At the same time, the weight function is applied to the local energy term (1-ω)E L , adjust the gray value of the local area; Then the global energy term plus the local energy is expressed as: Among them, λ1, λ2, λ3, λ4, λ5, and λ6 are coefficients of each item.

4. The hybrid level set image segmentation method based on distance regularization term and local feature enhancement according to claim 1, characterized in that: The edge stop function is constructed by using the result of the canny operator and the image convolution. The reasonable order is selected by adaptive method, and the high-order boundary operator is applied to the boundary term, including: The canny operator is used to convolve the image to enhance edge detection; the edge stop function is constructed using the convolution result. in is the gradient operator, G σ represents the canny operator, ρ controls the speed of curve evolution. The larger the ρ, the faster the curve evolves. α is the sensitivity of the edge index to noise.

5. The hybrid level set image segmentation method based on distance regularization term and local feature enhancement according to claim 4, characterized in that: Select an adaptive method to select a reasonable order α. By setting parameters, the speed of curve evolution can be increased while reducing the sensitivity to noise, including: Define the adaptive selection order: α is an integer, and the adaptive selection indicates that the order of the edge stop function starts from 1 and reaches α. When (G α+1 -G α +G α +G α-1 +…+G2-G1) / N is less than the threshold τ, α is taken as the order of the edge stop function, N is the number of pixels, and τ is a threshold; Higher-order boundary operators are applied to the boundary energy terms:

6. The hybrid level set image segmentation algorithm with logarithmic and polynomial distance regularization terms and local variance improvement according to claim 1, characterized in that: Generate a distance regularization term and introduce the edge indicator function into the regularization term to avoid repeated initialization of the level set, including: Combining potential function penalties using logarithmic, polynomial, and constant represents the gradient; The Gaussian kernel function is used to convolve with the image, and the result is an edge indication function; G σ represents the Gaussian kernel function; Introducing the edge indicator function into the energy penalty term can smooth the image and reduce the impact of noise. The distance regularization term is expressed as:

7. The hybrid level set image segmentation method based on distance regularization term and local feature enhancement according to claim 1, characterized in that: Generate level set energy terms, use gradient descent method to calculate, minimize the energy function, and achieve image segmentation, including: Combine global terms, local terms, boundary terms, and distance regularization terms; Among them, λ1, λ2, λ3, λ4, λ5, λ6, μ, ν, and p are the parameters of each item; The energy function is calculated using the gradient flow descent method to minimize the energy function and obtain the segmentation curve; Take λ1=λ2=λ3=λ4=λ5=λ6=1 Let I(t)-f3(x)=e1, I(y)-f4(x)=e2 Where div is the divergence and δ is the Dirac function, which can be obtained by taking the derivative of H(φ):

Images