Intelligent Analysis Method for Benzimidazolone Pigment Particle Contour Based on Graph Search Algorithm

Abstract

This invention relates to the field of image processing technology, and more specifically, to an intelligent contour analysis method for benzimidazolone pigment particles based on a graph search algorithm. The method includes: acquiring a grayscale image using an electron microscope and performing binarization segmentation; constructing a particle distance map based on Euclidean distance transformation; analyzing the curvature features of pixels in the particle distance map using the Hessian matrix to obtain the probability that a pixel in the particle distance map belongs to the core skeleton of the particle; extracting contour depression points based on the closed contour of the particle region; calculating the adhesion penalty intensity of each pixel in the particle distance map; accurately locating seed points by combining the skeleton probability and penalty intensity; and achieving precise cutting of adhered particles by constructing a graph model containing specific edge weights, using the seed points and background as constraints. This invention solves the problem of over-segmentation or under-segmentation caused by severe adhesion of pigment particles, improving the accuracy of contour recognition.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to an intelligent analysis method for the contours of benzimidazolone pigment particles based on a graph search algorithm. Background Technology

[0002] Benzimidazolone pigments, as high-performance organic pigments, exhibit physical properties such as hiding power, tinting strength, and dispersion stability that are closely related to their particle size, aspect ratio, and crystal integrity. In the quality control phase of industrial production, scanning electron microscopy or transmission electron microscopy is typically used to acquire microscopic images of the particles. Otsu's method is then employed for initial binarization to separate the particle regions. Subsequently, a graph-based minimum cut / maximum flow algorithm is used to construct a flow network containing source vertices (foreground seeds) and sink vertices (background seeds) to find the globally optimal segmentation boundary, thereby achieving precise separation of individual particles.

[0003] The segmentation accuracy of graph theory-based minimum cut / maximum flow algorithms is highly dependent on the correctness of seed point labeling. Therefore, existing techniques often use distance transformation combined with local maximum detection to obtain seed points. The aim is to calculate the distance from each pixel in the binarized particle region to the background, and the point with the largest distance value is the potential particle center. However, in actual production imaging, since benzimidazolone crystals are mostly long and thin rod-shaped or needle-shaped, the particles are very prone to close side-by-side adhesion or cross-over. This can lead to false distance maxima at the adhesion points, mistakenly identifying the adhesion points as particle centers. Alternatively, in closely packed particle clusters, only one seed point can be extracted. Once seed points are missing or the number is incorrect, the graph theory-based minimum cut / maximum flow algorithm will be unable to accurately identify the physical boundaries of the adhered particles and effectively separate the adhered particle clusters. Ultimately, this leads to distorted particle size statistics, directly affecting the accurate feedback and optimization adjustment of the production process. Summary of the Invention

[0004] To address the technical problem of missing or incorrectly numbered seed points when using distance transformation combined with local maximum detection to obtain seed points in existing technologies, due to the tendency of benzimidazolone crystals to tightly adhere or overlap, this invention proposes an intelligent analysis method for benzimidazolone pigment particle contours based on a graph search algorithm. This method includes the following steps:

[0005] Obtain the grayscale image of benzimidazolone pigment, perform binarization segmentation to obtain a binarized image, and determine the set of particle regions and the set of background regions; A particle distance map is constructed based on the distance values ​​from each pixel in the particle region set to the nearest background region set pixels. The curvature features of the pixels in the particle distance map are analyzed using the Hessian matrix to obtain the probability that a pixel in the particle distance map belongs to the core skeleton of the particle. Contour depression points are extracted based on the closed contours of each particle region in the particle distance map. The adhesion penalty intensity of the pixels in the particle distance map is obtained based on the distance from the pixel to the contour depression point and the gradient consistency coefficient. By combining the distance values ​​of pixels in the particle distance map, the probability that a pixel belongs to the particle core skeleton, and the adhesion penalty intensity, the probability that a pixel belongs to a seed point is obtained, and then the seed point is determined. Based on the seed points as constraints and the edge weights between adjacent pixels, a graph model is constructed. The graph model is then segmented using the minimum cut / maximum flow algorithm to obtain independent particle regions. The contours of each independent particle region were extracted to calculate the morphological parameters of the benzimidazolone particles.

[0006] The innovation of this invention lies in introducing the Hessian matrix to analyze the curvature characteristics of pixels, considering the ridge morphology of rod-shaped benzimidazolone particles in the distance map, and obtaining the probability that a pixel belongs to the particle core skeleton. Then, combining the gradient consistency of the contour depression points generated by particle overlap, the adhesion penalty intensity of the pixels in the particle distance map is obtained. Based on the probability of the particle core skeleton and the adhesion penalty intensity, the probability of belonging to a seed point is obtained, resulting in several various sub-points. This invention considers the geometric characteristics of the particles themselves and the physical connection characteristics at the adhesion points, effectively solving the problem of missing or excessive seed points caused by particle agglomeration, and can effectively separate adhered particle clusters.

[0007] Preferably, the construction of the particle distance map includes: For any pixel within the set of particle regions in the binarized image, obtain the Euclidean distance between the pixel and all background pixels in the binarized image, and denote the background pixel corresponding to the minimum value of the obtained Euclidean distance as the target background pixel of the pixel; use the Euclidean distance between the pixel and the target background pixel as the distance value of the pixel; construct a particle distance map based on the distance values ​​of all pixels in the set of particle regions.

[0008] Preferably, obtaining the probability that a pixel in the particle distance map belongs to the particle core skeleton includes:

[0009] In the formula, This represents the probability that the p-th pixel in the particle distance map belongs to the particle's core skeleton. The principal curvature value of the Hessian matrix representing the p-th pixel in the particle distance map; The second curvature value of the Hessian matrix at the p-th pixel in the particle distance map represents the value of the second curvature; exp() represents an exponential function with the natural constant as the base. The adjustment factor representing the principal curvature value.

[0010] Preferably, the extraction of contour depression points based on the closed contours of each particle region in the particle distance map includes: Based on the topology contour tracking algorithm, the closed contour of the particle region to which the p-th pixel belongs in the particle distance map is extracted, and the curvature value of each point on the closed contour is obtained. If the midpoint between the left and right neighboring points of the k-th point on the closed contour is outside the particle distance map, the k-th point on the closed contour is taken as a suspected contour depression point. All adjacent suspected contour depression points on the closed contour are connected to obtain several depression line segments. The point with the largest absolute value of curvature value on each depression line segment of the closed contour is taken as the contour depression point of the p-th pixel in the particle distance map.

[0011] Preferably, obtaining the adhesion penalty intensity of pixels in the particle distance map includes: ; In the formula, This represents the adhesion penalty intensity of the p-th pixel in the particle distance map; This represents the number of contour depressions at the p-th pixel in the particle distance map; This represents the distance between the p-th pixel and its q-th contour depression in the image. The gradient consistency coefficient represents the distance between the p-th pixel in the particle distance map and the q-th contour depression point. This represents the preset penalty intensity coefficient.

[0012] Preferably, obtaining the gradient consistency coefficient includes: Obtain the unit direction vector pointing from the qth contour depression point of the p-th pixel in the particle distance image to the p-th pixel; obtain the gradient vector of the qth contour depression point of the p-th pixel in the binarized image, and invert the direction of the gradient vector to obtain the normal vector of the qth contour depression point of the p-th pixel; use vector dot product operation to obtain the cosine similarity between the unit direction vector and the normal vector, and take the maximum value between the cosine similarity and 0 as the gradient consistency coefficient between the p-th pixel and the qth contour depression point in the particle distance image.

[0013] Preferably, the step of obtaining the probability that a pixel belongs to a seed point, and then determining the seed point, includes: ; In the formula, This represents the probability that the p-th pixel in the particle distance map belongs to the seed point. This represents the probability that the p-th pixel in the particle distance map belongs to the particle's core skeleton. This represents the adhesion penalty intensity of the p-th pixel in the particle distance map; This represents the distance value of the p-th pixel in the particle distance map; local maxima extraction is performed on the probability of all pixels in the particle distance map belonging to seed points to obtain several seed points.

[0014] By quantifying the geometric morphological features of rod-shaped particles using the Hessian matrix and combining this with contour concavity information to apply strong suppression to the adhesion region, effective seed points can be obtained.

[0015] Preferably, the step of constructing a graph model based on the seed point as a constraint and based on the edge weights between adjacent pixels includes: Pixels in the benzimidazolone pigment grayscale image other than seed points and background area pixels are taken as target pixels; , This represents the edge weight between the i-th target pixel and its j-th neighboring pixels; This represents the grayscale value of the i-th target pixel. represents the grayscale value of the j-th neighboring pixel of the i-th target pixel; exp() represents an exponential function with the natural constant as the base; each pixel in the benzimidazolone pigment grayscale image is mapped as a graph node, and a graph model is constructed based on the infinite connection weight between the seed point and the background pixel in the image and the edge weight.

[0016] The seed points in the image are assigned infinite connection weights to prevent them from being cut off.

[0017] Preferably, the step of extracting the contour of each independent particle region and calculating the morphological parameters of the benzimidazolone particles includes: Based on the topology contour tracking algorithm, the closed contour of each independent particle region in the grayscale image of benzimidazolone pigment is extracted, and the equivalent particle size, aspect ratio and roundness of each benzimidazolone particle are calculated based on the closed contour of each independent particle region.

[0018] By extracting the contours of the segmented particles and calculating their morphological parameters, precise quantitative indicators can be provided for pigment performance evaluation.

[0019] Preferably, the step of acquiring the benzimidazolone pigment grayscale image, performing binarization segmentation to obtain a binarized image, and determining the set of particle regions and the set of background regions includes: Grayscale images of benzimidazolone pigments were acquired using an electron microscope. Illumination compensation and noise suppression were performed on the grayscale images using the sliding window mean method. The Otsu method was used to segment the grayscale images of the benzimidazolone pigments to obtain binarized images. Regions with a pixel grayscale value of 1 in the binarized images were designated as granular regions, and regions with a pixel grayscale value of 1 were designated as background regions. Connectivity analysis was performed on the granular regions, and the set of all granular regions was denoted as the granular region set, and the set of background regions was denoted as the background region set.

[0020] By performing preliminary segmentation on the original image, the target region to be analyzed was extracted, laying a high-quality data foundation for subsequent refined processing.

[0021] The present invention has the following beneficial effects: The purpose of the present invention is to obtain the probability of a pixel belonging to the core skeleton of the particle by introducing the Hessian matrix to analyze the curvature characteristics of the pixel, and to quantify the ridge morphology characteristics of the rod-shaped benzimidazolone particles in the distance map; then, by combining the gradient consistency of the contour depression points generated by particle overlap, the adhesion penalty intensity of the pixel in the particle distance map is obtained, and based on the probability of the core skeleton of the particle and the adhesion penalty intensity, the probability of belonging to the seed point is obtained, resulting in a number of various sub-points. It takes into account the geometric characteristics of the particles themselves and the physical connection characteristics of the adhesion points, effectively solving the problem of missing or excessive seed points caused by particle agglomeration, and can effectively separate the agglomerates of adhered particles. Attached Figure Description

[0022] Figure 1 This is a flowchart of the steps of the intelligent analysis method for benzimidazolone pigment particle contour based on graph search algorithm in an embodiment of the present invention; Figure 2 It is a distribution map of the probability that pixels at different positions belong to the core skeleton. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent analysis of benzimidazolone pigment particle contours based on a graph search algorithm, according to an embodiment of the present invention. The method includes the following steps: S001. Obtain the grayscale image of benzimidazolone pigment, perform illumination compensation and noise suppression processing on it, and determine the set of particle regions and the set of background regions through binarization segmentation.

[0025] In this embodiment of the invention, a grayscale image of benzimidazolone pigment is acquired using an electron microscope. Illumination compensation and noise suppression are performed on the grayscale image using the sliding window mean method. The Otsu method is used to segment the grayscale image of benzimidazolone pigment to obtain a binarized image. Regions with a pixel grayscale value of 1 in the binarized image are designated as granular regions, and regions with a pixel grayscale value of 1 are designated as background regions. Connectivity analysis is performed on the granular regions, and the set of all granular regions is denoted as the granular region set, and the set of background regions is denoted as the background region set.

[0026] It should be noted that the sliding window averaging method is a well-known technique used to eliminate uneven illumination and suppress noise during electron microscope imaging. In this embodiment of the invention, it will not be described in detail.

[0027] S002. Construct a particle distance map based on the Euclidean distance from each pixel in the particle region set to the nearest background pixel. Analyze the curvature features of the pixels in the particle distance map using the Hessian matrix to obtain the probability that a pixel in the particle distance map belongs to the core skeleton of the particle. Extract contour depression points based on the closed contour of the particle region and calculate the adhesion penalty intensity of each pixel in the particle distance map.

[0028] It should be noted that in the binarized particle region, the geometric center of the particle is the position farthest from the background. In the distance map generated by the Euclidean distance transformation, the distance value at this position is the largest. Therefore, this invention locates the seed point based on the Euclidean distance transformation. Since benzimidazolone pigment particles usually have a slender rod-shaped or needle-shaped structure, when two particles are side by side or cross-adhesive, the traditional Euclidean distance transformation will cause the distance value of the pixel at the adhesion point to be very close to the value of the particle center, so it is impossible to distinguish between the two by a simple threshold or maximum value. Geometric analysis reveals that the distance map of a single rod-shaped particle exhibits a unidirectional convex ridge shape, while the distance map at the particle adhesion point resembles a saddle surface. Based on this morphological difference, this invention introduces the Hessian matrix to perform second-order partial derivative analysis on the distance map, solving for the principal and secondary curvatures of each pixel within the distance map. This quantitatively distinguishes the morphological characteristics of the central axis of the rod-shaped particle and the pixels at the adhesion point. The specific analysis is as follows: For a single rod-shaped particle, the distance value distribution follows the basic rule of high at the center and low at the edges. Therefore, the rod-shaped particle has a continuous high distance value distribution along the long axis, and the gradient change is very gentle. Thus, the smaller the secondary curvature (curvature along the long axis of the rod-shaped particle) calculated by the Hessian matrix, the smaller the distance value. On the other hand, along the direction perpendicular to the long axis, the distance value shows a rapid decreasing trend, and the gradient change is steeper. Therefore, the larger the principal curvature (curvature along the direction perpendicular to the long axis of the rod-shaped particle) calculated by the Hessian matrix, that is, the distance map of a single rod-shaped particle has a unidirectional convex ridge shape. When rod-shaped particles adhere together, the distance value of pixels closer to the adhesion point increases along the direction of the particle connection (the surface along the particle connection is convex upwards). Therefore, the principal curvature calculated by the Hessian matrix is ​​larger and positive. Furthermore, when two rod-shaped particles adhere incompletely, a gap without particle material is formed between the adhered particles, which belongs to the background area. When a pixel moves from the center of adhesion to both sides perpendicular to the particle connection, it quickly enters this gap, causing the distance from the pixel to the nearest background to shorten sharply, and the distance value decreases rapidly. Therefore, the absolute value of the secondary curvature (the surface is concave downwards along the direction perpendicular to the particle connection) calculated by the Hessian matrix is ​​larger and the sign of the secondary curvature is negative. That is, the distance map of the stick-shaped particle adhesion is convex in one direction and concave in the other direction perpendicular to it, resembling a saddle surface. In this embodiment of the invention, for any pixel within the set of particle regions in a binarized image, the Euclidean distance between the pixel and all background pixels in the binarized image is obtained, and the background pixel corresponding to the minimum value of the obtained Euclidean distance is recorded as the target background pixel of the pixel; the Euclidean distance between the pixel and the target background pixel is used as the distance value of the pixel; and a particle distance map is constructed based on the distance values ​​of all pixels in the set of particle regions. The Hessian matrix of the p-th pixel in the particle distance map is obtained. The principal curvature value and the secondary curvature value of the Hessian matrix of the p-th pixel in the particle distance map are obtained, and the principal curvature value is greater than the secondary curvature value. It should be noted that the acquisition of the Hessian matrix and the principal and secondary curvature values ​​are well-known technologies, and will not be described in detail in this embodiment of the invention. Obtain the probability that each pixel in the particle distance map belongs to the particle core skeleton:

[0029] In the formula, This represents the probability that the p-th pixel in the particle distance map belongs to the particle's core skeleton. The principal curvature value of the Hessian matrix representing the p-th pixel in the particle distance map; The second curvature value of the Hessian matrix at the p-th pixel in the particle distance map represents the value of the second curvature; exp() represents an exponential function with the natural constant as the base. The adjustment factor representing the principal curvature value; In this embodiment of the invention, the adjustment factor for the preset principal curvature value is... ,when Taking a smaller value makes the ratio extremely sensitive to changes in the secondary curvature. If a pixel deviates slightly from the center of the grain's major axis, causing the secondary curvature to be non-zero, the ratio will be severely affected. The value of will rapidly decay towards 0, thus greatly reducing the probability and ensuring that only pixels in the long axis region of the grain with extremely low sub-curvature can obtain a high skeleton probability. The smaller the secondary curvature value of any pixel and the larger the principal curvature value of that pixel, the flatter the surface is in the secondary direction and the more convex it is in the principal direction, which better matches the ridge morphology of the rod-shaped particle region, thus leading to... The indicator increased; The larger the square of the secondary curvature value and the principal curvature value of any pixel, the more it conforms to the saddle-shaped surface of the adhesion region, leading to... The indicator decreased rapidly.

[0030] For the center of the rod-shaped particle, its distance field exhibits a unidirectional ridge distribution, satisfying the condition that the smaller the secondary curvature value and the larger the principal curvature value, the better. and The values ​​all approach 1, leading to The index increases; however, in the saddle-shaped adhesion region, due to the larger square of the curvature in each direction, The value rapidly decays to 0, causing The indicator decreased rapidly.

[0031] It should be noted that even if the skeletal reinforcement process is completed for the particle region, the core regions of adjacent particles in deep aggregation will still overlap and overlap, making the central regions of two particles more likely to be close. Reinforcing the skeletal process alone cannot distinguish their respective belonging ranges, which can easily lead to undersegmentation problems. To address this deficiency, this step utilizes the characteristic of the high curvature contour depression boundary that inevitably exists at the adhesion point of benzimidazole pigment particles to obtain the contour depression point. If any pixel is in the adhesion region, the closer it is to the contour depression point, and the more consistent the gradient of the pixel is with the direction pointing to the contour depression feature point, the greater the adhesion penalty weight of the pixel, and the more likely the pixel is to belong to the adhesion region.

[0032] It should be further noted that the particle distance map contains both single particle regions and particle cluster regions, and each pixel in the particle distance map belongs to a certain particle region.

[0033] In this embodiment of the invention, based on the topological contour tracking algorithm, the closed contour of the particle region to which the p-th pixel belongs in the particle distance map is extracted, and the curvature value of each point on the closed contour is obtained; if the midpoint between the left and right neighboring points of the k-th point on the closed contour is outside the particle distance map, the k-th point on the closed contour is taken as a suspected contour depression point; all adjacent suspected contour depression points on the closed contour are connected to obtain several depression line segments, and the point with the largest absolute value of curvature value on each depression line segment of the closed contour is taken as the contour depression point of the p-th pixel in the particle distance map.

[0034] Obtain the unit direction vector pointing from the qth contour depression point of the p-th pixel in the particle distance image to the p-th pixel; obtain the gradient vector of the qth contour depression point of the p-th pixel in the binarized image, and invert the direction of the gradient vector to obtain the normal vector of the qth contour depression point of the p-th pixel; use vector dot product operation to obtain the cosine similarity between the unit direction vector and the normal vector, and take the maximum value between the cosine similarity and 0 as the gradient consistency coefficient between the p-th pixel and the qth contour depression point in the particle distance image; Obtain the adhesion penalty strength of the p-th pixel in the particle distance map:

[0035] In the formula, This represents the adhesion penalty intensity of the p-th pixel in the particle distance map; This represents the number of contour depressions at the p-th pixel in the particle distance map; This represents the distance between the p-th pixel and its q-th contour depression in the image. The gradient consistency coefficient represents the distance between the p-th pixel in the particle distance map and the q-th contour depression point. This represents a preset penalty strength coefficient, used to adjust the contribution strength of the gradient consistency coefficient; in this embodiment of the invention, the preset... In other embodiments, implementers may pre-set according to specific implementation conditions. The value; When pixel p is located in the adhesion region, its distance to the qth contour depression point is significantly reduced, resulting in an exponential increase in the adhesion penalty intensity. When pixel p is located in the adhesion region, the unit direction vector pointing from the q-th contour depression point of the p-th pixel to the p-th pixel is aligned with the direction of the normal vector of the q-th contour depression point of the p-th pixel. Approaching 1.

[0036] S003. Combining the distance value of the particle distance map, the probability that a pixel belongs to the particle core skeleton, and the adhesion penalty intensity, the probability that a pixel belongs to a seed point is obtained, and then the seed point is determined; based on the seed point and the background region as constraints, and obtaining the edge weights of the target pixel and each of its neighboring pixels, a graph model is constructed, and the minimum cut path is solved based on the minimum cut / maximum flow algorithm to cut the adhered particles and obtain each independent particle region.

[0037] It should be noted that the probability of a pixel belonging to a seed point is obtained based on the probability that the pixel belongs to the core skeleton of the particle and the adhesion penalty strength of the pixel.

[0038] In this embodiment of the invention, the probability that the p-th pixel in the particle distance map belongs to the seed point is obtained as follows:

[0039] In the formula, This represents the probability that the p-th pixel in the particle distance map belongs to the seed point. This represents the probability that the p-th pixel in the particle distance map belongs to the particle's core skeleton. This represents the adhesion penalty intensity of the p-th pixel in the particle distance map; This represents the distance value of the p-th pixel in the particle distance map; The larger the value, the greater the distance between the particle and the p-th pixel in the image, the greater the probability that the particle belongs to the core skeleton and the greater the probability that the p-th pixel belongs to the seed point. The purpose is to reduce the probability that pixels in the adhesion region belong to seed points, and to prevent pixels in the adhesion region from being mistakenly selected as seed points, thereby solving the undersegmentation problem caused by overlapping seed points in traditional methods.

[0040] In this embodiment of the invention, a local maximum extraction operation is performed on the probability of all pixels in the particle distance map belonging to seed points to obtain several seed points.

[0041] It should be noted that traditional graph search methods are prone to cutting deviations in low-gradient, contiguous regions, mainly due to the lack of seed point constraints and reasonable edge weight allocation between adjacent pixels. Therefore, this invention uses seed points as constraints on the graph model. Under the constraints of seed points, the algorithm can automatically locate the path with the lowest cutting cost in the contiguous region. Furthermore, it utilizes the sharp grayscale jumps at the particle boundaries to obtain the edge weights between adjacent pixels to guide the cutting direction of the algorithm. This enables separation along the physical contact surface of crystal growth even when benzimidazolone pigment particles overlap, ensuring the integrity of the crystal morphology while achieving accurate reconstruction of the benzimidazolone crystal contour.

[0042] In this embodiment of the invention, seed points in the benzimidazole pigment grayscale image are assigned unweighted large connection weights, and background region pixels in the benzimidazole pigment grayscale image are assigned infinite weights; the pixels with infinite weights are less likely to be segmented. Pixels in the benzimidazolone pigment grayscale image other than seed points and background area pixels are taken as target pixels; Obtain the edge weights between the target pixel and each of its neighboring pixels:

[0043] In the formula, This represents the edge weight between the i-th target pixel and its j-th neighboring pixels; This represents the grayscale value of the i-th target pixel. represents the gray value of the j-th neighboring pixel of the i-th target pixel; exp() represents an exponential function with the natural constant as the base. The larger the value, the greater the difference in grayscale value between the i-th target pixel and its j-th neighboring pixel, indicating that it is located at the edge of the grain region and the background region. In this case, the i-th target pixel and its j-th neighboring pixel should be cut off. The smaller the value, the closer the i-th target pixel is to its j-th neighboring pixel. In this case, the i-th target pixel and its j-th neighboring pixel should be cut. The larger the value, the smaller the difference in grayscale value between the i-th target pixel and its j-th neighboring pixel, indicating that the pixel belongs to the particle region. In this case, the i-th target pixel and its j-th neighboring pixel should not be cut.

[0044] Each pixel in the benzimidazolone pigment grayscale image is mapped as a graph node. Based on the connection weights between seed points and background pixels and the edge weights between target pixels and their neighboring pixels, a graph model is constructed. The minimum cut / maximum flow algorithm is used to segment the graph model and solve for the minimum cut path that separates the foreground regions corresponding to different seed points, thereby cutting the adhered particles and obtaining each independent particle region in the benzimidazolone pigment grayscale image.

[0045] It should be noted that, under the constraint of the seed point, the algorithm can automatically locate the path with the lowest cutting cost in the adhesion region, thereby achieving accurate reconstruction of the crystal outline of benzimidazolone while ensuring the integrity of the crystal morphology.

[0046] S004. Extract the contours of each independent particle region and calculate the morphological parameters of the benzimidazolone particles.

[0047] In this embodiment of the invention, based on the topological contour tracking algorithm, the closed contour of each independent particle region in the grayscale image of benzimidazolone pigment is extracted, and the equivalent particle size, aspect ratio and roundness of each benzimidazolone particle are calculated based on the closed contour of each independent particle region.

[0048] It should be noted that the extraction of the closed contour and morphological parameter calculation of the benzimidazolone pigment particle region adopts conventional techniques known in the art, and will not be described in detail in the embodiments of the present invention.

[0049] Figure 2This is a distribution chart showing the probability of pixels at different locations belonging to the core skeleton. It illustrates the response curves of skeleton probability as pixel curvature changes with principal and secondary curvature. The horizontal axis represents the principal curvature value, and the shape of the curve is determined by the secondary curvature value. When a pixel is located on the central axis of a rod-shaped particle like benzimidazolone, the greater its principal curvature and the closer its secondary curvature is to 0, the greater the probability that the pixel belongs to the particle's core skeleton. Figure 2 This manifests as a rapid rise and maintenance at a high level, achieving enhancement of the core skeleton pixels; when a pixel is located in the adhesion boundary region, the greater its principal curvature and the larger the absolute value of its secondary curvature, the greater the probability that the pixel belongs to the particle core skeleton, and the smaller the probability that the pixel belongs to the particle core skeleton. Figure 2 In this design, the dotted line always stays close to the bottom, thus suppressing the adhesion of pixels.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent analysis of the profile of a benzimidazolone pigment particle based on a graph search algorithm, characterized in that, include: Obtain the grayscale image of benzimidazolone pigment, perform binarization segmentation to obtain a binarized image, and determine the set of particle regions and the set of background regions; A particle distance map is constructed based on the distance values ​​from each pixel in the particle region set to the nearest background region set pixels. The curvature features of the pixels in the particle distance map are analyzed using the Hessian matrix to obtain the probability that a pixel in the particle distance map belongs to the core skeleton of the particle. Contour depression points are extracted based on the closed contours of each particle region in the particle distance map. The adhesion penalty intensity of the pixels in the particle distance map is obtained based on the distance from the pixel to the contour depression point and the gradient consistency coefficient. By combining the distance values ​​of pixels in the particle distance map, the probability that a pixel belongs to the particle core skeleton, and the adhesion penalty intensity, the probability that a pixel belongs to a seed point is obtained, and then the seed point is determined. Based on the seed points as constraints and the edge weights between adjacent pixels, a graph model is constructed. The graph model is then segmented using the minimum cut / maximum flow algorithm to obtain independent particle regions. The contours of each independent particle region were extracted to calculate the morphological parameters of the benzimidazolone particles.

2. The graph search algorithm based intelligent analysis method of the benzoimidazolone pigment particle profile according to claim 1, characterized in that, The construction of the particle distance map includes: For any pixel within the set of particle regions in the binarized image, obtain the Euclidean distance between the pixel and all background pixels in the binarized image, and denote the background pixel corresponding to the minimum value of the obtained Euclidean distance as the target background pixel of the pixel; use the Euclidean distance between the pixel and the target background pixel as the distance value of the pixel; construct a particle distance map based on the distance values ​​of all pixels in the set of particle regions.

3. The method of claim 1, wherein the method is characterized by, The process of obtaining the probability that a pixel in the particle distance map belongs to the particle core skeleton includes: wherein, represents the probability that the pth pixel in the particle distance map belongs to the particle core skeleton; represents the principal curvature value of the Hessian matrix of the pth pixel in the particle distance map; represents the secondary curvature value of the Hessian matrix of the pth pixel in the particle distance map; exp() represents the exponential function with the natural constant as the base number; represents the adjustment factor of the principal curvature value.

4. The graph search algorithm based intelligent analysis method of the benzoimidazolone pigment particle profile according to claim 1, characterized in that, The extraction of contour depression points based on the closed contours of each particle region in the particle distance map includes: Based on the topology contour tracking algorithm, the closed contour of the particle region to which the p-th pixel belongs in the particle distance map is extracted, and the curvature value of each point on the closed contour is obtained. If the midpoint between the left and right neighboring points of the k-th point on the closed contour is outside the particle distance map, the k-th point on the closed contour is taken as a suspected contour depression point. All adjacent suspected contour depression points on the closed contour are connected to obtain several depression line segments. The point with the largest absolute value of curvature value on each depression line segment of the closed contour is taken as the contour depression point of the p-th pixel in the particle distance map.

5. The method of claim 1, wherein the method is characterized by, The step of obtaining the adhesion penalty intensity of pixels in the particle distance map includes: ； In the formula, represents the adhesion penalty intensity of the pth pixel point in the particle distance map; represents the number of contour concave points of the pth pixel point in the particle distance map; represents the distance between the pth pixel point and the qth contour concave point thereof in the particle distance map; represents the gradient consistency coefficient between the pth pixel point and the qth contour concave point thereof in the particle distance map; represents the preset penalty intensity coefficient.

6. The method of intelligent analysis of the benzoimidazolone pigment particle profile based on the graph search algorithm according to claim 4 or 5, characterized in that, The acquisition of the gradient consistency coefficient includes: Obtain the unit direction vector pointing from the qth contour depression point of the p-th pixel in the particle distance image to the p-th pixel; obtain the gradient vector of the qth contour depression point of the p-th pixel in the binarized image, and invert the direction of the gradient vector to obtain the normal vector of the qth contour depression point of the p-th pixel; use vector dot product operation to obtain the cosine similarity between the unit direction vector and the normal vector, and take the maximum value between the cosine similarity and 0 as the gradient consistency coefficient between the p-th pixel and the qth contour depression point in the particle distance image.

7. The intelligent analysis method for benzimidazolone pigment particle contours based on graph search algorithm according to claim 1, characterized in that, The process of obtaining the probability that a pixel belongs to a seed point, and then determining the seed point, includes: ； In the formula, This represents the probability that the p-th pixel in the particle distance map belongs to the seed point. This represents the probability that the p-th pixel in the particle distance map belongs to the particle's core skeleton. This represents the adhesion penalty intensity of the p-th pixel in the particle distance map; This represents the distance value of the p-th pixel in the particle distance map; local maxima extraction is performed on the probability of all pixels in the particle distance map belonging to seed points to obtain several seed points.

8. The intelligent analysis method for benzimidazolone pigment particle contours based on graph search algorithm according to claim 1, characterized in that, The construction of a graph model based on the seed point as a constraint and based on the edge weights between adjacent pixels includes: Pixels in the benzimidazolone pigment grayscale image other than seed points and background area pixels are taken as target pixels; , This represents the edge weight between the i-th target pixel and its j-th neighboring pixels; This represents the grayscale value of the i-th target pixel. represents the grayscale value of the j-th neighboring pixel of the i-th target pixel; exp() represents an exponential function with the natural constant as the base; each pixel in the benzimidazolone pigment grayscale image is mapped as a graph node, and a graph model is constructed based on the infinite connection weight between the seed point and the background pixel in the image and the edge weight.

9. The intelligent analysis method for benzimidazolone pigment particle contours based on graph search algorithm according to claim 1, characterized in that, The extraction of contours from each independent particle region and the calculation of morphological parameters for benzimidazolone particles include: Based on the topology contour tracking algorithm, the closed contour of each independent particle region in the grayscale image of benzimidazolone pigment is extracted, and the equivalent particle size, aspect ratio and roundness of each benzimidazolone particle are calculated based on the closed contour of each independent particle region.

10. The intelligent analysis method for benzimidazolone pigment particle contours based on graph search algorithm according to claim 1, characterized in that, The process of acquiring a grayscale image of the benzimidazolone pigment, performing binarization segmentation to obtain a binarized image, and determining the set of particle regions and the set of background regions includes: Grayscale images of benzimidazolone pigments were acquired using an electron microscope. Illumination compensation and noise suppression were performed on the grayscale images using the sliding window mean method. The Otsu method was used to segment the grayscale images of the benzimidazolone pigments to obtain binarized images. Regions with a pixel grayscale value of 1 in the binarized images were designated as granular regions, and regions with a pixel grayscale value of 1 were designated as background regions. Connectivity analysis was performed on the granular regions, and the set of all granular regions was denoted as the granular region set, and the set of background regions was denoted as the background region set.

Images