Image segmentation method and system based on multiplicative bias field modeling

Abstract

The invention relates to the technical field of image processing, and discloses an image segmentation method and system based on multiplicative bias field modeling, comprising: filtering an image to be segmented by using nonlinear bilateral guided filtering to obtain an optimized image; constructing a basis function group based on multiple preset two-dimensional Gaussian basis functions and a pseudo-bias field basis function of the optimized image, and multiplying the group with a coefficient column vector to obtain an initial bias field component of the optimized image; dividing the optimized image into sub-image domains, presetting an initial contour line represented by a pseudo-level set method for each sub-image domain, and obtaining an initial regional mean constant of each sub-image domain; iteratively updating the initial bias field component, the pseudo-level set function of each sub-image domain and the initial regional mean constant by using an energy function difference until a preset requirement is met, and obtaining a target pseudo-level set function of each sub-image domain as a target contour line; and segmenting the image to be segmented based on the target contour line of each sub-image domain to obtain an image segmentation result.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image segmentation method and system based on multiplicative bias field modeling. Background Art

[0002] Image segmentation is an important research direction in the field of computer vision and image processing. Its goal is to divide an image into several non-overlapping connected regions and extract the region of interest (ROI). Over the years, scholars at home and abroad have conducted extensive research on image segmentation and proposed a variety of segmentation algorithms. Among them, the active contour model has attracted widespread attention in recent years due to its flexible topological transformation and strong mathematical theoretical support. Many scholars have conducted research on this and derived a variety of different algorithms. The active contour model can be combined with a variety of prior knowledge to construct an energy function, which can obtain sub-pixel accuracy and smooth, closed contour curves, and can handle complex topological changes. To date, the active contour model has become an important method in image segmentation.

[0003] However, in the actual imaging process, due to the influence of lighting conditions, shooting angles or transmission equipment, uneven grayscale often occurs, resulting in similar grayscale values ​​at the boundaries of different regions, making the boundaries unclear. For such images, in the traditional active contour model based on fuzzy theory, the average grayscale value inside and outside the contour line is difficult to truly reflect the grayscale distribution characteristics, which easily leads to the problem of mis-segmentation of non-target areas in grayscale uneven images. In the traditional active contour model, the evolution of the contour line is achieved by updating the level set function; in order to make the evolution curve smooth and fit the target edge, it is usually necessary to add a length term to the energy function and adjust the length term synchronously during the iterative update. However, this method not only increases the iteration time, but also requires setting and optimizing more parameters for the length term.

[0004] In summary, the existing image segmentation methods based on active contour models have the problem of mis-segmentation and poor segmentation accuracy due to the uneven grayscale distribution of the image and insufficient reflection of the grayscale distribution characteristics; and the increase of the length term during the evolution of the contour line increases the iteration time and reduces the image segmentation efficiency. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the problems in the prior art of insufficient reflection of grayscale distribution characteristics of grayscale-uneven images and complex length term iteration, which lead to poor image segmentation accuracy and low computational efficiency.

[0006] In order to solve the above technical problems, the present invention provides an image segmentation method based on multiplicative bias field modeling, comprising:

[0007] Obtain an image to be segmented, filter the image to be segmented using a nonlinear bilateral guided filter, and obtain a corresponding optimized image;

[0008] Based on a plurality of preset two-dimensional Gaussian basis functions and the quasi-bias field basis functions of the optimized image, a basis function group of the optimized image is constructed; based on the product of the basis function group of the optimized image and a preset initial coefficient column vector, an initial bias field component corresponding to the optimized image is obtained;

[0009] The optimized image is divided into sub-image domains, and an initial contour line is preset for each sub-image domain. The initial contour line of each sub-image domain is represented by a pseudo level set method, and the initial regional mean constant corresponding to each sub-image domain is obtained.

[0010] The initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image are iteratively updated using the energy function difference until the maximum absolute value of the difference between two adjacent iteratively updated regional mean constants corresponding to all sub-image domains is less than a preset threshold, the iteration is stopped, and the target pseudo level set function corresponding to each sub-image domain is obtained as the target contour line of each sub-image domain;

[0011] The image to be segmented is segmented based on the target contour line of each sub-image domain in the optimized image to obtain the image segmentation result.

[0012] Preferably, the image to be segmented is filtered using a nonlinear bilateral guided filter to obtain a corresponding optimized image, which is expressed as:

[0013] ;

[0014] in, Indicates optimized image. Represents the vector pixel points in the image to be segmented, Indicates Centered at The square window, Display Window The vector pixels within for the summation symbol;

[0015] represents the normalized nonlinear bilateral weight, expressed as ; represents the nonlinear bilateral weight, expressed as ; represents the exponential function, represents the Euclidean function, represents the spatial variance, represents the grayscale variance, Represents the grayscale value of the vector pixel in the image to be segmented, Display Window Gray value of the vector pixel within;

[0016] represents the Gaussian filtered image, and the expression is ; Indicates is the Gaussian filter of the convolution kernel template, It is the convolution symbol;

[0017] and is the guided filter coefficient, and the expressions are:

[0018] ; ;

[0019] in, Represents the vector pixel point in the Gaussian filter image The gray value at ; Indicates that the Gaussian filtered image is in the window The weighted grayscale mean within is expressed as ; Indicates that the image to be segmented is in the window The weighted grayscale mean within is expressed as ; represents the regularization parameter.

[0020] Preferably, based on a plurality of preset two-dimensional Gaussian basis functions and the quasi-biased field basis functions of the optimized image, a basis function group of the optimized image is constructed, which is expressed as:

[0021] ;

[0022] Among them, the basis function The dimension is , and Respectively represent the width and height of the optimized image; represents the transpose symbol; Indicated by Basis Functions The basis function group composed of The dimension is ;

[0023] when hour, represents the two-dimensional Gaussian basis function, expressed as ; and Respectively represent Mean and variance in a two-dimensional Gaussian basis function;

[0024] when hour, represents the quasi-bias field basis function, and the expression is ; represents the minimum value function, represents the maximum value function; It means that the intensity distribution features of the optimized image are extracted by dilation and corrosion operations, and then mean filtering is performed. The expression is , Indicates the window size is The mean filter, Indicates size The rectangular structural element, represents the dilation operation, Represents the erosion operation.

[0025] Preferably, based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, the initial bias field component corresponding to the optimized image is obtained, which is expressed as:

[0026] ;

[0027] in, represents the initial bias field component, The dimension is The coefficient column vector of , represents the first The preset weight coefficients corresponding to the basis functions.

[0028] Preferably, the optimized image is divided into sub-image domains, and an initial contour line is preset for each sub-image domain. The initial contour line of each sub-image domain is represented by a pseudo level set method, including:

[0029] The optimized image is divided into sub-image domains. In each sub-image domain, when the vector pixel point x is inside or on the initial contour line, the pseudo level set function The value of is 1; when the vector pixel point x is outside the initial contour line, the pseudo level set function The value of is 0;

[0030] No. The initial pseudo level set corresponding to the sub-image domain is expressed as:

[0031] ;

[0032] in, represents the image domain of the optimized image, The image domain represents the optimized image. sub-image domain, , Indicates the total number of sub-image regions.

[0033] Preferably, the initial region mean constant corresponding to each sub-image domain is obtained, which is expressed as:

[0034] ;

[0035] in, Indicates The initial region mean constant corresponding to the sub-image domain, represents the average value calculation function, represents the element-wise product symbol, Indicates the optimized image Sub-image area The average gray value of the vector pixels.

[0036] Preferably, the energy function difference is used to iteratively update the initial bias field component of the optimized image, the pseudo level set function corresponding to each sub-image domain in the optimized image, and the initial regional mean constant until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, and the iteration is stopped to obtain the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain, including:

[0037] Iteration-based The pseudo level set function, regional mean constant and bias field component after the iteration are constructed The energy function after the second;

[0038] Based on the energy function after two adjacent iterations, the iteration is calculated The energy function difference after the second time;

[0039] The initial bias field component in the optimized image is iteratively updated using the energy function difference;

[0040] The energy function difference is used to iteratively update the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image;

[0041] Until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterations of updating corresponding to all sub-image domains is less than the preset threshold, the level set function after the current iteration of each sub-image domain is obtained as the target pseudo level set function corresponding to each sub-image domain;

[0042] The target pseudo level set function corresponding to each sub-image domain is used as the target contour line of each sub-image domain.

[0043] Preferably, obtaining the energy function difference includes:

[0044] Iteration-based The pseudo level set function, regional mean constant and bias field component after the iteration are constructed The energy function after , expressed as:

[0045] ;

[0046] Based on the energy function after two adjacent iterations, the iteration is calculated Energy function difference after , expressed as:

[0047] ;

[0048] in, represents the number of iterations, Represents iteration The pseudo level set function set after this time is expressed as , Indicates The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after the second; Represents iteration The set of regional mean constants after the times, , Indicates The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second; represents the initial bias field component iteration The bias field component after the second; Indicates in the area Upper pair function To perform integration, is a constant.

[0049] Preferably, the iterative updating of the initial bias field component of the optimized image, the pseudo level set function corresponding to each sub-image domain in the optimized image and the initial regional mean constant by using the energy function difference comprises:

[0050] Iterate the initial bias field component The subsequent bias field component is expressed as: ;

[0051] Represents iteration The coefficient column vector after the second is expressed as ;

[0052] For The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second is expressed as: ;

[0053] Based on The bias field component after the iteration and the initial regional mean constant are obtained to obtain the The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after this time is expressed as: ; Indicates the window size is The median filter.

[0054] This embodiment also provides a system based on the above-mentioned image segmentation method based on multiplicative bias field modeling, comprising:

[0055] A filtering optimization module is used to obtain an image to be segmented, filter the image to be segmented using a nonlinear bilateral guided filter, and obtain a corresponding optimized image;

[0056] The bias field modeling module is used to construct a basis function group of the optimized image based on multiple preset two-dimensional Gaussian basis functions and the quasi-bias field basis functions of the optimized image; based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, obtain the initial bias field component corresponding to the optimized image;

[0057] An initial input module is used to divide the optimized image into sub-image domains, preset an initial contour line for each sub-image domain, express the initial contour line of each sub-image domain using a pseudo level set method, and obtain an initial regional mean constant corresponding to each sub-image domain;

[0058] An iterative update module is used to iteratively update the initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image using the energy function difference, until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, and then the iteration is stopped to obtain the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain;

[0059] The image segmentation module is used to segment the image to be segmented based on the target contour line of each sub-image domain in the optimized image to obtain the image segmentation result.

[0060] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0061] The image segmentation method based on multiplicative bias field modeling of the present invention optimizes the image to be segmented by using nonlinear bilateral guided filtering, while removing noise interference, retaining edge details, improving image quality, and laying a foundation for subsequent segmentation. The present invention adopts a bias field modeling method based on the product of a basis function group and a coefficient column vector, mathematically represents the bias field component through the basis function group, optimizes the coefficient column vector, and realizes accurate estimation of the bias field; at the same time, by dividing the optimized image into sub-image domains, using a pseudo-level set function to represent the initial contour line, and obtaining the initial regional mean constant; based on the pseudo-level set function and the regional mean constant corresponding to each sub-image domain, iteratively updating using the energy function difference; the present invention models each sub-image domain of the image to be segmented as the product of the regional mean constant and the bias field component, and uses the energy function minimization to synchronously optimize the bias field component, and realizes bias field estimation and image segmentation at the same time, thereby improving the efficiency and accuracy of image segmentation, and realizing multi-phase segmentation of the image to be segmented, which can effectively cope with multi-target scenes, and improves the smoothness and accuracy of the segmentation boundary by optimizing the level set function, and further improves the efficiency and accuracy of segmenting weak edges and grayscale uneven images. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0063] Figure 1 It is a flowchart of the steps of the image segmentation method based on multiplicative bias field modeling provided by the present invention;

[0064] Figure 2 A schematic diagram showing a comparison of the segmentation results of the present invention and the MICO model for the first image;

[0065] Figure 3 A schematic diagram showing a comparison of the segmentation results of the second image between the present invention and the MICO model;

[0066] Figure 4 A schematic diagram showing a comparison of the segmentation results of the present invention and the MICO model for the third image;

[0067] Figure 5 A schematic diagram showing a comparison of the segmentation results of the fourth image between the present invention and the MICO model;

[0068] Figure 6 A schematic diagram showing a comparison of the segmentation results of the fifth image between the present invention and the MICO model;

[0069] Figure 7 A schematic diagram showing a comparison of the segmentation results of the sixth image between the present invention and the MICO model. DETAILED DESCRIPTION

[0070] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0071] Reference Figure 1 As shown, the step flow chart of the image segmentation method based on multiplicative bias field modeling provided by the present invention includes the following steps:

[0072] S101: Obtain an image to be segmented, and filter the image to be segmented using a nonlinear bilateral guided filter to obtain a corresponding optimized image;

[0073] S102: constructing a basis function group of the optimized image based on a plurality of preset two-dimensional Gaussian basis functions and a quasi-biased field basis function of the optimized image;

[0074] S103: obtaining an initial bias field component corresponding to the optimized image based on the product of the basis function group of the optimized image and the preset initial coefficient column vector;

[0075] S104: Divide the optimized image into sub-image domains, preset an initial contour line for each sub-image domain, express the initial contour line of each sub-image domain using a pseudo level set method, and obtain an initial region mean constant corresponding to each sub-image domain;

[0076] S105: using the energy function difference to iteratively update the initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image, until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, the iteration is stopped, and the target pseudo level set function corresponding to each sub-image domain is obtained as the target contour line of each sub-image domain;

[0077] S106: Segment the image to be segmented based on the target contour line of each sub-image domain in the optimized image to obtain an image segmentation result.

[0078] Specifically, in step S101, the image to be segmented is filtered using a nonlinear bilateral guided filter to obtain a corresponding optimized image, including:

[0079] S101-1: Calculate Gaussian filter image , expressed as:

[0080] ;

[0081] Represents the vector pixel points in the image to be segmented, Indicates Centered at The square window, Indicates is the Gaussian filter of the convolution kernel template, It is the convolution symbol; Represents the vector pixel points in the image to be segmented Gray value of Set to 7;

[0082] S101-2: Calculate normalized nonlinear bilateral weights , expressed as:

[0083] ;

[0084] represents the nonlinear bilateral weight, expressed as ; represents the exponential function, represents the Euclidean function; represents the spatial variance, ; represents the grayscale variance, ; Display Window The vector pixels within For the summation symbol, Represents the grayscale value of the vector pixel in the image to be segmented, Display Window Gray value of the vector pixel within;

[0085] S101-3: Calculate the guided filter coefficient and , expressed as:

[0086] ;

[0087] ;

[0088] Represents the vector pixel point in the Gaussian filter image The gray value at ; Indicates that the Gaussian filtered image is in the window The weighted grayscale mean within is expressed as ; Indicates that the image to be segmented is in the window The weighted grayscale mean within is expressed as ; represents the regularization parameter, ;

[0089] S101-4: Calculate the optimized image based on Gaussian filter image, normalized nonlinear bilateral weights and guided filter coefficients , expressed as: .

[0090] The nonlinear bilateral guided filtering of the embodiment of the present invention combines the nonlinear bilateral mechanism with the spatial weight and grayscale weight of the local area to further enhance the sensitivity to the edge, and can better retain complex edges and details while smoothing noise; by introducing more contextual information through the guide image, it can more accurately distinguish between noise and real image content, thereby improving the robustness of the filter. The nonlinear bilateral guided filtering balances the smoothing effect and detail retention in a larger range through a nonlinear weighted method and combines the information of the guide image, significantly improving the image quality and laying the foundation for subsequent segmentation.

[0091] Specifically, in step S102, a basis function group of the optimized image is constructed based on a plurality of preset two-dimensional Gaussian basis functions and the quasi-biased field basis functions of the optimized image, including:

[0092] S102-1: When When , construct the corresponding two-dimensional Gaussian basis function , the expression is ; and Respectively represent Mean and variance in a two-dimensional Gaussian basis function;

[0093] S102-2: When When , construct the quasi-bias field basis function , the expression is ; represents the minimum value function, represents the maximum value function; It means that the intensity distribution features of the optimized image are extracted by dilation and corrosion operations, and then mean filtering is performed. The expression is , Indicates the window size is The mean filter, Indicates size The rectangular structural element, represents the dilation operation, Represents the erosion operation.

[0094] S102-3: By Basis function group consisting of basis functions , expressed as:

[0095] ;

[0096] Basis Functions The dimension is , and Respectively represent the width and height of the optimized image; represents the transpose symbol; Indicated by Basis Functions The basis function group composed of The dimension is .

[0097] In this embodiment, when setting When , the mean and variance in the Gaussian basis function are set to be expressed as and ; Set the window size of the mean filter The value range is , the size of the rectangular structure element .

[0098] Specifically, in step S103, based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, the initial bias field component corresponding to the optimized image is obtained, which is expressed as: ;in, represents the initial bias field component, The dimension is The coefficient column vector of , represents the first The preset weight coefficients corresponding to the basis functions.

[0099] The embodiment of the present invention performs a method for bias field modeling based on the product of a preset number of basis functions and the corresponding coefficient column vector, and mathematically represents the bias field components through basis functions; at the same time, in the process of constructing the basis functions, Gaussian basis functions are processed, and image morphological processing is also used to construct quasi-bias field basis vectors, which can adapt to image bias field distributions of different complexities and further improve the robustness of image segmentation.

[0100] Specifically, in step S104, the optimized image is divided into sub-image domains, and the corresponding pseudo level set function and the initial regional mean constant are obtained, including:

[0101] S104-1: Divide the optimized image into sub-image regions;

[0102] S104-2: The initial contour corresponding to each sub-image area is represented by a pseudo level set method. When the vector pixel point x is located inside or on the initial contour, the pseudo level set function The value of is 1; when the vector pixel point x is outside the initial contour line, the pseudo level set function The value of is 0; The initial pseudo level set corresponding to the sub-image domain , expressed as:

[0103] ;

[0104] in, represents the image domain of the optimized image, The image domain represents the optimized image. sub-image domain, , Indicates the total number of sub-image regions.

[0105] S104-3: Obtain the initial region mean constant corresponding to each sub-image domain , expressed as:

[0106] ;

[0107] in, Indicates The initial region mean constant corresponding to the sub-image domain, represents the average value calculation function, represents the element-wise product symbol, Indicates the optimized image Sub-image area The average gray value of the vector pixels.

[0108] Specifically, in step S105, the energy function difference is used to perform iterative updating, including:

[0109] S105-1: Iteration-based The pseudo level set function, regional mean constant and bias field component after the iteration are constructed The energy function after , expressed as:

[0110] ;

[0111] S105-2: Calculate the iteration based on the energy function after two adjacent iterations Energy function difference after , expressed as:

[0112] ;

[0113] in, represents the number of iterations, Represents iteration The pseudo level set function set after this time is expressed as , Indicates The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after the second; Represents iteration The set of regional mean constants after the times, , Indicates The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second; represents the initial bias field component iteration The bias field component after the second; Indicates in the area Upper pair function To perform integration, is a constant, usually taking the value of 2.

[0114] S105-3: Iteratively updating the initial bias field component of the optimized image using the energy function difference;

[0115] Iterate the initial bias field component The subsequent bias field component is expressed as: ;

[0116] Represents iteration The coefficient column vector after the second is expressed as ;

[0117] S105-4: iteratively updating the pseudo level set function and the initial region mean constant corresponding to each sub-image domain in the optimized image using the energy function difference;

[0118] For The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second is expressed as: ;

[0119] Based on The bias field component after the iteration and the initial regional mean constant are obtained to obtain the The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after this time is expressed as: ; Indicates the window size is The median filter, The value range is .

[0120] S105-5: until the maximum value of the difference absolute values ​​of the regional mean constants updated after two adjacent iterations corresponding to all sub-image domains is less than a preset threshold, the level set function after the current iteration of each sub-image domain is obtained as the target pseudo level set function corresponding to each sub-image domain;

[0121] That is, when the energy function difference is used for iterative update, the termination condition is met. When , the pseudo level set function stops iterative updating, and the last updated pseudo level set function is used as the final contour line. is the preset threshold value, which is taken as .

[0122] S105-6: Using the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain.

[0123] The energy function of the embodiment of the present invention models the image as the product of the regional mean constant and the bias field component, and optimizes the two components synchronously by minimizing the energy function, so as to realize bias field estimation and image segmentation, thereby improving segmentation efficiency and accuracy. While optimizing the coefficient column vector, the bias field modeling is made more accurate, the accuracy of the segmented target area is improved, and the mis-segmentation caused by the influence of the bias field is avoided. In addition, the present embodiment is a multi-pseudo level set active contour model, which can realize multi-phase segmentation, effectively cope with multi-target scenes, improve the smoothness and accuracy of the segmentation boundary by optimizing the level set function, and is suitable for image segmentation with weak edges and uneven grayscale.

[0124] Based on the above embodiments, in this embodiment, the image segmentation method based on multiplicative bias field modeling and the MICO (Multiplicative Intrinsic Component Optimization) model described in this application are used to perform image segmentation simulation experiments, and the experimental results are compared.

[0125] All experiments in this example were implemented in MatlabR2021b on a 1.8 GHz Intel Core i5 personal computer. The images used in the comparative experiment were from the BrainWeb library. The brain MRI images were segmented. The areas to be segmented included white matter areas, gray matter areas, and cerebrospinal fluid areas. Therefore, the total number of image domains was ; The remaining parameters are set to: , , , , , , , , , , , , and set the maximum number of iterations and initial contours.

[0126] First, six brain MRI images with grayscale unevenness of 60% were selected from the BrainWeb library, and the six images (numbered a to f) were segmented using the MICO model and the image segmentation method based on multiplicative bias field modeling of the present invention, where Poisson noise was artificially added to images d to f. The segmentation results and correction results of the MICO model and the present invention are shown in Figure 2. Figures 2 to 7 As shown; Figure 2 A schematic diagram showing a comparison of the segmentation results of the present invention and the MICO model for the first image; Figure 3 A schematic diagram showing a comparison of the segmentation results of the second image between the present invention and the MICO model; Figure 4 A schematic diagram showing a comparison of the segmentation results of the present invention and the MICO model for the third image; Figure 5 A schematic diagram showing a comparison of the segmentation results of the fourth image between the present invention and the MICO model; Figure 6 A schematic diagram showing a comparison of the segmentation results of the fifth image between the present invention and the MICO model; Figure 7 Schematic diagram showing the comparison of the segmentation results of the present invention and the MICO model for the sixth image. The first row of each sub-image is the result of the MICO model, the second row is the result of the present invention, the second to fourth columns are the segmented brain white matter area, gray matter area and cerebrospinal fluid area, the fifth column is the model predicted bias field, and the sixth column is the corrected image. Figures 2 to 7 It can be seen that the present invention can obtain accurate segmentation results and bias field correction results for both noise-free grayscale uneven images and noisy grayscale uneven images. From the overall image segmentation results, the segmentation results of the method of the present invention are better than those of the MICO model.

[0127] In order to further compare the performance of the MICO model and the present invention, the running time and segmentation accuracy are compared. The shorter the running time, the higher the segmentation efficiency. In terms of segmentation accuracy, the DSC (Dice Similarity Coefficient) and IoU (Intersection of Union) are used to measure the segmentation accuracy of each model. DSC is defined as , IoU is defined as ,in, is a standard target area provided in the BrainWeb library. is the target area obtained by the experiment. The closer the DSC and IOU values ​​are to 1, the higher the segmentation accuracy is. Table 1 is a comparison table of the running time and segmentation accuracy of the MICO model and the model of the present invention;

[0128] Table 1 Comparison of running time (seconds) and segmentation accuracy (DSC / IoU) of two image segmentation models

[0129]

[0130] According to Table 1, compared with the MICO model, the image segmentation method of the present invention can quickly and accurately segment noise-free grayscale uneven images and noisy grayscale uneven images, and has certain advantages in segmentation efficiency and segmentation accuracy.

[0131] Based on the above embodiment, the embodiment of the present invention further provides a system based on the image segmentation method based on multiplicative bias field modeling as described above, comprising:

[0132] A filtering optimization module is used to obtain an image to be segmented, filter the image to be segmented using a nonlinear bilateral guided filter, and obtain a corresponding optimized image;

[0133] The bias field modeling module is used to construct a basis function group of the optimized image based on multiple preset two-dimensional Gaussian basis functions and the quasi-bias field basis functions of the optimized image; based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, obtain the initial bias field component corresponding to the optimized image;

[0134] An initial input module is used to divide the optimized image into sub-image domains, preset an initial contour line for each sub-image domain, express the initial contour line of each sub-image domain using a pseudo level set method, and obtain an initial regional mean constant corresponding to each sub-image domain;

[0135] An iterative update module is used to iteratively update the initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image using the energy function difference, until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, and then the iteration is stopped to obtain the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain;

[0136] The image segmentation module is used to segment the image to be segmented based on the target contour line of each sub-image domain in the optimized image to obtain the image segmentation result.

[0137] The image segmentation device based on multiplicative bias field modeling of this embodiment is used to implement the aforementioned image segmentation method based on multiplicative bias field modeling. Therefore, the specific implementation method of the image segmentation device based on multiplicative bias field modeling can be seen in the embodiment part of the image segmentation method based on multiplicative bias field modeling in the previous text, for example, the filtering optimization module is used to implement step S101 in the above-mentioned image segmentation method based on multiplicative bias field modeling; the bias field modeling module is used to implement steps S102 and S103 in the above-mentioned image segmentation method based on multiplicative bias field modeling; the initial input module, the iterative update module, and the image segmentation module are respectively used to implement steps S104, S105 and S106 in the above-mentioned image segmentation method based on multiplicative bias field modeling. Therefore, its specific implementation method can refer to the description of the corresponding each part of the embodiment, which will not be repeated here.

[0138] The image segmentation method based on multiplicative bias field modeling of the present invention optimizes the image to be segmented by using nonlinear bilateral guided filtering, while removing noise interference, retaining edge details, improving image quality, and laying a foundation for subsequent segmentation. The present invention adopts a bias field modeling method based on the product of a basis function group and a coefficient column vector, mathematically represents the bias field component through the basis function group, optimizes the coefficient column vector, and realizes accurate estimation of the bias field; at the same time, by dividing the optimized image into sub-image domains, using a pseudo-level set function to represent the initial contour line, and obtaining the initial regional mean constant; based on the pseudo-level set function and the regional mean constant corresponding to each sub-image domain, iteratively updating using the energy function difference; the present invention models each sub-image domain of the image to be segmented as the product of the regional mean constant and the bias field component, and uses the energy function minimization to synchronously optimize the bias field component, and realizes bias field estimation and image segmentation at the same time, thereby improving the efficiency and accuracy of image segmentation, and realizing multi-phase segmentation of the image to be segmented, which can effectively cope with multi-target scenes, and improves the smoothness and accuracy of the segmentation boundary by optimizing the level set function, and further improves the efficiency and accuracy of segmenting weak edges and grayscale uneven images.

[0139] 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 combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0140] The present application 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 box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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.

[0141] 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 comprising 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.

[0142] 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 in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. An image segmentation method based on multiplicative bias field modeling, characterized in that: include: Obtain an image to be segmented, filter the image to be segmented using a nonlinear bilateral guided filter, and obtain a corresponding optimized image; Based on a plurality of preset two-dimensional Gaussian basis functions and the quasi-biased field basis functions of the optimized image, a basis function group of the optimized image is constructed; Based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, an initial bias field component corresponding to the optimized image is obtained; The optimized image is divided into sub-image domains, and an initial contour line is preset for each sub-image domain. The initial contour line of each sub-image domain is represented by a pseudo level set method, and the initial regional mean constant corresponding to each sub-image domain is obtained. The initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image are iteratively updated using the energy function difference until the maximum absolute value of the difference between two adjacent iteratively updated regional mean constants corresponding to all sub-image domains is less than a preset threshold, the iteration is stopped, and the target pseudo level set function corresponding to each sub-image domain is obtained as the target contour line of each sub-image domain; The image to be segmented is segmented based on the target contour line of each sub-image domain in the optimized image to obtain the image segmentation result.

2. The image segmentation method based on multiplicative bias field modeling according to claim 1, characterized in that: The nonlinear bilateral guided filter is used to filter the segmented image to obtain the corresponding optimized image, which is expressed as: ； in, Indicates optimized image. Represents the vector pixel points in the image to be segmented, Indicates Centered at The square window, Display Window The vector pixels within for the summation symbol; represents the normalized nonlinear bilateral weight, expressed as ; represents the nonlinear bilateral weight, expressed as ; represents the exponential function, represents the Euclidean function, represents the spatial variance, represents the grayscale variance, Represents the grayscale value of the vector pixel in the image to be segmented, Display Window Gray value of the vector pixel within; represents the Gaussian filtered image, and the expression is ; Indicates is the Gaussian filter of the convolution kernel template, It is the convolution symbol; and is the guided filter coefficient, and the expressions are: ； ； in, Represents the vector pixel point in the Gaussian filter image The gray value at ; Indicates that the Gaussian filtered image is in the window The weighted grayscale mean within is expressed as ; Indicates that the image to be segmented is in the window The weighted grayscale mean within is expressed as ; represents the regularization parameter.

3. The image segmentation method based on multiplicative bias field modeling according to claim 2, characterized in that: Based on multiple preset two-dimensional Gaussian basis functions and the quasi-biased field basis functions of the optimized image, a basis function group of the optimized image is constructed, which is expressed as: ； Among them, the basis function The dimension is , and Respectively represent the width and height of the optimized image; represents the transpose symbol; Indicated by Basis Functions The basis function group composed of The dimension is ; when hour, represents the two-dimensional Gaussian basis function, expressed as ; and Respectively represent Mean and variance in a two-dimensional Gaussian basis function; when hour, represents the quasi-bias field basis function, and the expression is ; represents the minimum value function, It represents the maximum value function; It means that the intensity distribution features of the optimized image are extracted by dilation and corrosion operations, and then mean filtering is performed. The expression is , Indicates the window size is The mean filter, Indicates size The rectangular structural element, represents the dilation operation, Represents the erosion operation.

4. The image segmentation method based on multiplicative bias field modeling according to claim 3 is characterized in that: Based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, the initial bias field component corresponding to the optimized image is obtained, which is expressed as: ； in, represents the initial bias field component, The dimension is The coefficient column vector of , represents the first The preset weight coefficients corresponding to the basis functions.

5. The image segmentation method based on multiplicative bias field modeling according to claim 4, characterized in that: The optimized image is divided into sub-image domains, and an initial contour line is preset for each sub-image domain. The initial contour line of each sub-image domain is represented by a pseudo level set method, including: The optimized image is divided into sub-image domains. In each sub-image domain, when the vector pixel point x is inside or on the initial contour line, the pseudo level set function The value of is 1; when the vector pixel point x is outside the initial contour line, the pseudo level set function The value of is 0; No. The initial pseudo level set function corresponding to the sub-image domain is expressed as: ； in, represents the image domain of the optimized image, The image domain represents the optimized image. sub-image domain, , Indicates the total number of sub-image regions.

6. The image segmentation method based on multiplicative bias field modeling according to claim 5, characterized in that: Get the initial region mean constant corresponding to each sub-image domain, expressed as: ； in, Indicates The initial region mean constant corresponding to the sub-image domain, represents the average value calculation function, represents the element-wise product symbol, Indicates the optimized image Sub-image area The average gray value of the vector pixels.

7. The image segmentation method based on multiplicative bias field modeling according to claim 6, characterized in that: The method uses the energy function difference to iteratively update the initial bias field component of the optimized image, the pseudo level set function corresponding to each sub-image domain in the optimized image, and the initial regional mean constant until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, stops iteration, and obtains the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain, including: Iteration-based The pseudo level set function, regional mean constant and bias field component after the iteration are constructed The energy function after the second; Based on the energy function after two adjacent iterations, the iteration is calculated The energy function difference after the second time; The initial bias field component in the optimized image is iteratively updated using the energy function difference; The energy function difference is used to iteratively update the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image; Until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterations of updating corresponding to all sub-image domains is less than the preset threshold, the level set function after the current iteration of each sub-image domain is obtained as the target pseudo level set function corresponding to each sub-image domain; The target pseudo level set function corresponding to each sub-image domain is used as the target contour line of each sub-image domain.

8. The image segmentation method based on multiplicative bias field modeling according to claim 7, characterized in that: The energy function difference is obtained, including: Iteration-based The pseudo level set function, regional mean constant and bias field component after the iteration are constructed The energy function after , expressed as: ； Based on the energy function after two adjacent iterations, the iteration is calculated Energy function difference after , expressed as: ； in, represents the number of iterations, Represents iteration The pseudo level set function set after this time is expressed as , Indicates The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after the second; Represents iteration The set of regional mean constants after the times, , Indicates The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second; represents the initial bias field component iteration The bias field component after the second; Indicates in the area Upper pair function To perform integration, is a constant.

9. The image segmentation method based on multiplicative bias field modeling according to claim 8, characterized in that: The method of iteratively updating the initial bias field component of the optimized image, the pseudo level set function corresponding to each sub-image domain in the optimized image, and the initial regional mean constant by using the energy function difference comprises: Iterate the initial bias field component The subsequent bias field component is expressed as: ; Represents iteration The coefficient column vector after the second is expressed as ； For The initial region mean constant iteration corresponding to the sub-image domain The regional mean constant after the second is expressed as: ； Based on The bias field component after the iteration and the initial regional mean constant are obtained to obtain the The initial pseudo level set function corresponding to the sub-image domain is iterated The pseudo level set function after this time is expressed as: ; Indicates the window size is The median filter.

10. A system based on the image segmentation method based on multiplicative bias field modeling as claimed in any one of claims 1 to 9, characterized in that: include: A filtering optimization module is used to obtain an image to be segmented, filter the image to be segmented using a nonlinear bilateral guided filter, and obtain a corresponding optimized image; A bias field modeling module, used for constructing a basis function group of an optimized image based on a plurality of preset two-dimensional Gaussian basis functions and a quasi-bias field basis function of the optimized image; Based on the product of the basis function group of the optimized image and the preset initial coefficient column vector, an initial bias field component corresponding to the optimized image is obtained; An initial input module is used to divide the optimized image into sub-image domains, preset an initial contour line for each sub-image domain, express the initial contour line of each sub-image domain using a pseudo level set method, and obtain an initial regional mean constant corresponding to each sub-image domain; An iterative update module is used to iteratively update the initial bias field component of the optimized image, as well as the pseudo level set function and the initial regional mean constant corresponding to each sub-image domain in the optimized image using the energy function difference, until the maximum value of the difference absolute value of the regional mean constant after two adjacent iterative updates corresponding to all sub-image domains is less than a preset threshold, and then the iteration is stopped to obtain the target pseudo level set function corresponding to each sub-image domain as the target contour line of each sub-image domain; The image segmentation module is used to segment the image to be segmented based on the target contour line of each sub-image domain in the optimized image to obtain the image segmentation result.

Images