An Image Processing Method for Foamed Concrete Based on Sub-Pixel Edge Reconstruction

Abstract

The present invention discloses a method for processing images of foamed concrete based on sub-pixel edge reconstruction, specifically relating to the field of image processing of foamed concrete, including acquiring an image of foamed concrete and extracting a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information; performing edge structure classification on the image region based on the combined threshold judgment of the set of boundary response features; for the image region determined to be a locally directionally convergent mutation region, identifying a set of cross-mutation points by fitting the convergence trend of the direction tensor, and performing image processing of directional guided erosion and dilation. By constructing a structure determination variable based on the image boundary response features and guiding the double strategies of the erosion and dilation path and the vein backtracking path to perform sub-pixel-level aperture boundary separation, the image structure processing and contour reconstruction of the porous adhesion region are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of foamed concrete image processing. More specifically, the present invention relates to a foamed concrete image processing method based on sub-pixel edge reconstruction. Background Art

[0002] In the image processing of the surface of foamed concrete, the conventional pore size recognition method based on pixel-level binaryzation and connected component extraction is difficult to deal with the problem of hole adhesion caused by blurred pore boundaries, close fitting of pore walls or local structural collapse.

[0003] Especially at the microscale, multiple actual independent pores often present as a continuous high-gray-scale area, resulting in the traditional algorithm misidentifying them as a single large pore, destroying the true statistical structure of the pore size distribution, and further affecting the accurate evaluation of porosity, mechanical properties and material looseness.

[0004] Therefore, the current problem in the image processing of the surface of foamed concrete is: how to separate the porous boundary regions mismerged due to structural blurring at the sub-pixel scale. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a foamed concrete image processing method based on sub-pixel edge reconstruction. By constructing a structural determination variable based on the image boundary response characteristics, and guiding the dual strategies of the erosion and dilation path and the vein backtracking path to perform sub-pixel-level pore size boundary separation, so as to realize the image structure processing and contour reconstruction of the porous adhesion region.

[0006] To achieve the above object, the present invention provides the following technical solution: A foamed concrete image processing method based on sub-pixel edge reconstruction, comprising:

[0007] Obtain an image of foamed concrete, and extract a set of boundary response characteristics including edge direction information, edge gradient change information and local intensity contrast information; perform edge structure classification on the image region based on the combined threshold judgment of the set of boundary response characteristics;

[0008] For the image region determined to be a locally directionally convergent mutation region, identify the set of cross mutation points by fitting the convergence trend of the direction tensor, and perform image processing of directionally guided erosion and dilation;

[0009] For the image region determined to be a continuous closed high-consistency region, construct a structural potential field and perform a combined judgment of path tension and boundary coincidence degree, identify the true fracture path and perform topological division accordingly;

[0010] For the pore size structure maps respectively output by the image processing of the corrosion expansion path and the vein backtracking path, perform the cross-matching judgment of the three consistency scoring items in the boundary matching scoring matrix, and selectively integrate or retain the structure difference regions according to the fusion conditions;

[0011] Construct a structure determination mechanism through the fusion path and the scoring results, perform path update and callback correction on the boundaries of the regions marked as fuzzy regions in the image, and realize the identification of the pore size structure and the optimization of the image processing process.

[0012] In a preferred embodiment, obtain the original image data of the foamed concrete image, perform standardized grayscale processing and multi-scale edge enhancement operations on the original image data to obtain a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information;

[0013] Based on the set of boundary response features, extract three structure determination variables, where the structure determination variables include the boundary direction consistency metric value B1, the boundary strength gradient variation value B2, and the regional edge closure ratio B3;

[0014] If B1 is less than the boundary direction consistency metric threshold T1 and B2 is greater than the boundary strength gradient variation threshold T2, the image region is marked as a local direction convergence mutation region; if B1 is greater than or equal to T1 and B3 is greater than the regional edge closure ratio threshold T3, the image region is marked as a continuously closed high-consistency region; if the three are in a critical cross state, the image region is marked as a boundary feature determination fuzzy region;

[0015] Temporarily store the image blocks marked as boundary feature determination fuzzy regions as fuzzy region masks, and jointly incorporate them into the fusion judgment process with the image blocks marked as local direction convergence mutation regions and the image blocks marked as continuously closed high-consistency regions to perform the unified determination of the path attribution of the three types of regions.

[0016] In a preferred embodiment, input all local direction convergence mutation regions into the structure direction tensor extraction process, construct a tensor map of the corresponding main axis through the set of boundary response features, perform the convergence trend fitting of the main axis direction on the tensor map, generate a direction tensor convergence map, and identify the aggregated boundary region;

[0017] Based on the image analysis, identify the set of direction gradient mutation points in the aggregated boundary region as the fracture candidate subset, and perform tensor cross-angle analysis on each candidate point in the fracture candidate subset; if the cross angle is greater than the preset fracture criterion, the point is identified as a valid fracture point and is used to generate a fracture-induced layer;

[0018] Perform a direction tensor weighted erosion operation based on the fracture-induced layer, and then perform a finite dilation operation that maintains the main axis direction to preserve the boundary continuity and spatial consistency formed by the fractures; perform a structural contour extraction operation on the processed area, form a contour set by combining all the extracted boundary contours, and record the processing path labels corresponding to each contour at the same time;

[0019] Output the obtained contour set as the initial pore size structure map one under the erosion-dilation path.

[0020] In a preferred embodiment, input all continuous closed high-consistency regions into the structural potential energy construction process, generate a sub-pixel level structural potential energy field based on the edge strength and closed stability, perform a guided path search in the structural potential energy field to obtain a set of potential fracture paths, and calculate the path tension value and the structural boundary overlap degree for each path in turn;

[0021] If the path tension value is greater than the path tension threshold T4 and the structural boundary overlap degree is less than the structural boundary overlap degree threshold T5, then this path is confirmed as a real fracture path and used as the boundary cleavage trigger path;

[0022] Use the confirmed real fracture paths to perform a topological structure reconstruction operation on the original image and divide it into multiple logically pore sub-blocks; perform path label marking on the pore structures of the divided pore sub-blocks and output them as the initial pore size structure map two under the vein backtracking path.

[0023] In a preferred embodiment, merge the sets of the initial pore size structure map one and the initial pore size structure map two of the erosion-dilation path and the vein backtracking path, and construct a candidate boundary set for the boundary pairs in the two maps; for each pair of candidate boundaries in the candidate boundary set, calculate the boundary coincidence degree R1, the direction alignment degree R2, and the structural center offset amount R3, and form a boundary matching score matrix;

[0024] In the boundary matching score matrix, if R1 in a certain boundary pair is greater than the boundary coincidence degree threshold T6, and R2 and R3 are less than the direction alignment degree threshold T7 and the structural center offset amount threshold T8, then this boundary pair is determined as a fused boundary, perform layer merging and assign a unified number;

[0025] If the boundary matching score matrix does not meet the fused boundary determination criteria, then this image area is retained as a boundary feature determination fuzzy area; perform sub-pixel curve fitting on the boundaries of all determined image areas, and output a vector boundary set and the corresponding path label information.

[0026] In a preferred embodiment, path labels, scoring results of a boundary matching scoring matrix, and a structure status table are established with each boundary structure in the contour set as the main index. The structure status table is expanded based on the scoring results of the boundary matching scoring matrix by introducing a path execution status and a region fusion determination result. In the structure status table, the original policy path and the scoring matching value corresponding to each boundary structure are recorded;

[0027] For all image regions marked as regions with ambiguous boundary feature determination, evaluate the path determination confidence and the scoring consistency difference, and determine whether there is aperture recognition uncertainty;

[0028] If the path confidence is less than the rejudgment threshold T9 and the scoring consistency difference is greater than the conflict tolerance threshold T 10 , then re - perform boundary structure determination based on three structure determination variables to complete the missing or misclassified aperture boundaries;

[0029] According to the new round of structure determination results, re - specify the image structure path, and perform the corresponding image processing method to obtain the updated aperture boundary extraction result; the image structure path includes an erosion - dilation path or a vein backtracking path; replace the corresponding region in the original boundary layer with the updated image structure boundary, and synchronously update its corresponding path identification information.

[0030] In a preferred embodiment, define to represent the joint classification result of three structure determination variables corresponding to the pixel point in the image; define to represent the boundary direction consistency metric value B1 of the pixel point ; define to represent the boundary strength gradient variation value B2 of the pixel point ; define to represent the regional edge closure ratio B3 of the point in the image, which is used to measure whether the contour is closed and complete;

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] where represents the neighborhood set of the pixel point ; Represents the pixel in the neighborhood The main direction angle of the edge; Indicates the current pixel The main direction of the edge; For point Directional divergence suppression factor calculated in the local tensor field; is a multi-scale image pyramid set; Indicated in Pixels in the scaled image The gradient intensity modulus of ; is the anisotropy adjustment factor; Indicates that the boundary tracing algorithm is used to trace from pixel points Starting from, extract the pixel set on all boundary paths; Point on the path The local connectivity score of For point A set of multiple closed boundary candidate paths formed by the area; is a closed cost function; is a very small positive number; is the joint classification function.

[0037] In a preferred embodiment, for the image region structure in the local direction convergence mutation region, a nonlinear path mutation metric value is constructed by integrating the tensor direction torsion rate, local tension gradient variation and disturbance diffusion response. , construct the fracture induction function Determine potential boundary breakpoints in the image and use them to guide subsequent erosion and dilation processes to form a structural contour path;

[0038] ;

[0039] ;

[0040] The fracture induction function Represents the pixel in the image Whether it is marked as a structural fracture induction point, the output is Boolean value of Respectively represent the main direction angles extracted from the tensor map between the current pixel and its neighboring pixels; Represents the directed angle difference between the main directions; is the tension function; are the gradients of structural tension in the horizontal and vertical directions, respectively; is the direction consistency function; is the mixed second-order partial derivative with direction consistency; is the directional mutation intensity factor; is the local response rate of the direction diffusion tensor; additionally, in the formula represents the non-linear expansion term for constructing the propagation intensity of the structural perturbation; is the combined determination threshold for structural mutations;

[0041] Input all continuous closed high-consistency regions into the structural potential energy construction process to generate the structural potential energy trajectory function and the reference guiding path function , and combine with the closed perturbation function to construct the cleavage scoring function ; if the score is higher than the preset threshold , then this point is marked as a true fracture path point, and the cleavage path set is output for topological division and pore reconstruction;

[0042] ;

[0043] ;

[0044] At in the formula represents the sub-pixel coordinates of the currently processed pixel in the image; the structural potential energy trajectory function represents the potential energy-driven path within the closed region; , , and represent the first-order derivative along ; , is the second-order partial derivative of , is used to calculate the local structural curvature change; the reference guiding path function is used as the comparison reference path for the potential energy path ; , is the derivative of in two directions; the closed perturbation function represents the sub-pixel perturbation tension residual function of the path points in the closed region; the cleavage scoring function is a scoring function that fuses three factors and serves as the basis for judging whether a fracture path is formed; the preset threshold is the path cleavage scoring determination threshold; is the binary determination function for the fracture path, and the output of

[0045] In a preferred embodiment, candidate boundaries are constructed for the initial pore size structure map one and the initial pore size structure map two extracted from the corrosion expansion path and the vein backtracking path respectively, and a boundary matching score matrix is constructed based on three types of scoring quantities, namely the boundary coincidence degree R1, the direction alignment degree R2, and the structure center offset R3, to determine the boundary fusion condition;

[0046] The is used to define the score of the boundary coincidence degree R1:

[0047] ;

[0048] The is used to define the score of the direction alignment degree R2:

[0049] ;

[0050] The is used to define the score of the structure center offset R3:

[0051] ;

[0052] In , , in the formula, represents the pair of candidate boundaries; is the length of the overlapping part of the pair of candidate boundaries in the initial pore size structure map one and the initial pore size structure map two; is the total length of the union of the pair of candidate boundaries in the initial pore size structure map one and the initial pore size structure map two; is the difference in the change of the image edge gradient in the overlapping region of the pair of candidate boundaries ; is the topological fracture complexity index corresponding to the pair of candidate boundaries ; , is the main direction angle of the pair of candidate boundaries in the corrosion expansion path map and the vein backtracking path map; is the average curvature of the pair of candidate boundaries ; , is the horizontal and vertical offset of the geometric center of the pair of candidate boundaries ; is the structural symmetry coefficient of the pair of candidate boundaries .

[0053] In a preferred embodiment, for the image patches marked as the regions with ambiguous boundary feature determination, a path confidence index and a scoring consistency difference index indexed by the boundary structure are established, and a dynamic callback mechanism is constructed based on the two to update the image path and boundary judgment; the path confidence is defined as ; the scoring consistency difference is defined as ;

[0054] ;

[0055] ;

[0056] where is the boundary layer tension response extracted at the pixel point ; is the difference value of the image processing strategies of the pixel point under two paths; is the score of the pixel point under the erosion and dilation path; is the score of the pixel point under the vein backtracking path.

[0057] The technical effects and advantages of the present invention:

[0058] Through three types of structure determination variables, namely the boundary direction consistency metric, the boundary strength gradient variation, and the regional edge closure ratio, it is possible to classify the locally directionally convergent mutation regions and the closed high-consistency regions, achieve the structural correction of the misjudgment of adjacent pore adhesions, thereby breaking through the problem of the traditional pixel-level processing mismerging multiple pore diameters, and ensuring the granularity and authenticity of pore diameter recognition;

[0059] Adopt structure tensor fitting to construct the directionally convergent trend, form a fracture-induced layer, and perform directional erosion and dilation operations under the guidance of the main axis; at the same time, guide the path search through the structure potential field, calculate the path tension value and the structural boundary overlap degree to identify the true cleavage path, and realize the dual-path extraction mechanism of the image structure contour, enhancing the physical rationality and direction adaptability of path construction;

[0060] Fuse the pore diameter maps output by the erosion and dilation path and the vein backtracking path, form a matching scoring matrix by constructing three scoring factors, namely the boundary coincidence degree, the direction alignment degree, and the structural center offset amount, and systematically judge the fusion conditions of the candidate boundaries, solving the problems of boundary inconsistency or redundant overlap under traditional multi-path processing;

[0061] For the structurally uncertain region, by recording the boundary path score and execution status, combining the path confidence and the scoring consistency difference index to construct a dynamic callback mechanism, and re-specifying the processing path according to the structure determination variable, finally realizing the path callback and boundary reconstruction of the fuzzy region, and improving the flexibility of the overall image processing;

[0062] By integrating multi-scale boundary response features, tensor principal axis trends, structural potential energy functions, and scoring matrix mechanisms, a multi-dimensional cognitive model of the image structure from pixel level to sub-pixel level, from geometric features to path topology is realized, improving the characterization ability of complex pore morphology, and applicable to the image structure analysis and pore size evaluation of various types of foamed concrete materials. Brief Description of the Drawings

[0063] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0065] Referring to the attached drawings of the specification Figure 1 , a method for processing foamed concrete images based on sub-pixel edge reconstruction according to an embodiment of the present invention includes:

[0066] Obtain an image of foamed concrete, and extract a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information; perform edge structure classification on the image region based on the combined threshold judgment of the set of boundary response features;

[0067] For the image region determined to be a locally directionally convergent mutation region, identify the set of cross-mutation points by fitting the direction tensor convergence trend, and perform image processing of directional guided erosion and dilation, so as to achieve fracture splitting and contour extraction while maintaining structural continuity;

[0068] For the image region determined to be a continuous closed high-consistency region, construct a structural potential field and perform joint determination of path tension and boundary coincidence degree, identify the true fracture path and perform topological division accordingly, and finally complete the contour reconstruction of the veined pores;

[0069] For the pore size structure maps respectively output by the image processing of the erosion and dilation path and the vein backtracking path, perform cross-matching judgment on the three consistency scoring items in the boundary matching scoring matrix, and selectively integrate or retain the structural difference regions according to the fusion conditions;

[0070] Construct a structure determination mechanism by fusing paths and scoring results, perform path updates and callback corrections on the boundaries of the regions marked as blurred areas in the image, and achieve the recognition of the aperture structure and the optimization of the image processing process.

[0071] Obtain the original image data of the foamed concrete image, perform standardized gray processing and multi-scale edge enhancement operations on the original image data to obtain a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information;

[0072] Based on the set of boundary response features, extract three structure determination variables, including the boundary direction consistency metric value B1, the boundary strength gradient variation value B2, and the regional edge closure ratio B3;

[0073] If B1 is less than the boundary direction consistency metric threshold T1 and B2 is greater than the boundary strength gradient variation threshold T2, the image region is marked as a locally directionally convergent mutation region; if B1 is greater than or equal to T1 and B3 is greater than the regional edge closure ratio threshold T3, the image region is marked as a continuously enclosed high-consistency region; if the three are in a critical cross state, the image region is marked as a boundary feature determination blurred region;

[0074] Temporarily store the image blocks marked as boundary feature determination blurred regions as a blurred region mask, and jointly include them in the fusion judgment process with the image blocks marked as locally directionally convergent mutation regions and the image blocks marked as continuously enclosed high-consistency regions to perform unified determination of the path attribution of the three types of regions.

[0075] Input all locally directionally convergent mutation regions into the structure direction tensor extraction process, construct a tensor atlas of the corresponding main axis through the set of boundary response features, perform a convergence trend fitting of the main axis direction on the tensor atlas, generate a direction tensor convergence map, and identify the aggregated boundary region;

[0076] Based on image analysis, identify the set of direction gradient mutation points in the aggregated boundary region as a fracture candidate subset, and perform tensor cross-angle analysis on each candidate point in the fracture candidate subset; if the cross angle is greater than the preset fracture criterion, the point is identified as a valid fracture point and used to generate a fracture-induced layer;

[0077] Perform a direction tensor weighted erosion operation based on the fracture-induced layer, and then perform a limited dilation operation that maintains the main axis direction to preserve the boundary continuity and spatial consistency formed by the fracture; perform a structure contour extraction operation on the processed region, form a contour set with all the extracted boundary contours, and record the corresponding processing path labels for each contour at the same time;

[0078] Output the obtained contour set as the initial pore size structure map one under the erosion and dilation path.

[0079] Input all continuous closed high-consistency regions into the structural potential energy construction process, generate a sub-pixel-level structural potential energy field based on the edge strength and closed stability, perform a guided path search in the structural potential energy field to obtain a set of potential fracture paths, and calculate the path tension value and the structural boundary overlap degree for each path in turn;

[0080] If the path tension value is greater than the path tension threshold T4 and the structural boundary overlap degree is less than the structural boundary overlap degree threshold T5, then this path is confirmed as a real fracture path and used as a boundary cleavage trigger path;

[0081] Use the confirmed real fracture paths to perform topological structure reconstruction operations on the original image and divide it into multiple logically pore sub-blocks; perform path label marking on the pore structures of the divided pore sub-blocks and output them as the initial pore size structure map two under the vein backtracking path; in addition, the calculation of the path tension value is based on the pixel point tension response intensity on each potential fracture path in the structural potential energy field and the consistency of its gradient direction along the path direction, and is obtained by weighted accumulation in combination with the continuity of the guiding directions of each point on the path; the calculation of the structural boundary overlap degree is based on the ratio between the length of the geometric coincidence region between this potential fracture path and the current image boundary contour and the total length of the path, specifically evaluating whether the path effectively covers the original boundary structure, thus serving as an important basis for determining real fracture paths.

[0082] Merge the sets of the initial pore size structure map one and the initial pore size structure map two of the erosion and dilation path and the vein backtracking path, and construct a candidate boundary set for the boundary pairs in the two maps; for each pair of candidate boundaries in the candidate boundary set, calculate the boundary coincidence degree R1, the direction alignment degree R2, and the structural center offset amount R3, and form a boundary matching score matrix;

[0083] In the boundary matching score matrix, if R1 in a certain boundary pair is greater than the boundary coincidence degree threshold T6, and R2 and R3 are less than the direction alignment degree threshold T7 and the structural center offset amount threshold T8, then this boundary pair is determined as a fusion boundary, perform layer merging and assign a unified number;

[0084] If the boundary matching score matrix does not meet the fusion boundary determination criteria, then this image region is retained as a boundary feature determination fuzzy region; perform sub-pixel curve fitting on the boundaries of all determined image regions, and output a vector boundary set and corresponding path label information.

[0085] Establish a path label, a scoring result of a boundary matching scoring matrix, and a structure status table with each boundary structure in the contour set as the main index. The structure status table is extended based on the scoring result of the boundary matching scoring matrix by introducing a path execution status and a region fusion determination result. Record the original policy path and the scoring matching value corresponding to each boundary structure in the structure status table;

[0086] For all image regions marked as regions with ambiguous boundary feature determination, evaluate the path determination confidence and the scoring consistency difference, and determine whether there is aperture recognition uncertainty;

[0087] If the path confidence is less than the rejudgment threshold T9 and the scoring consistency difference is greater than the conflict tolerance threshold T 10 , then re - perform the boundary structure determination based on three structure determination variables to complement the missing or misclassified aperture boundaries;

[0088] According to the new round of structure determination results, re - specify the image structure path, and execute the corresponding image processing method to obtain the updated aperture boundary extraction result; the image structure path includes an erosion - dilation path or a vein backtracking path; replace the corresponding region in the original boundary layer with the updated image structure boundary, and synchronously update its corresponding path identification information, finally implementing an adaptive closed - loop aperture recognition method composed of five stages: image region structure determination, policy path judgment, path execution, boundary fusion, and callback correction.

[0089] It should be noted that in the formula structure involved in this solution, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions, so they do not change or confuse the overall unit system of the expression; such combinations of "dimensionless terms and terms with units" can be understood as the composite structure expression forms commonly used in mathematical - physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis;

[0090] Secondly, in the formula structure of this solution, if there are multiple variable terms with different physical units, including but not limited to time - type, mass - type, or energy - type variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can form a unified structure through function mapping, ratio combination, or normalization adjustment, with clear units and clear meanings, and the overall expression conforms to the principle of dimensional consistency and the common norms of engineering modeling;

[0091] In this solution, if constants, weights, adjustment factors, threshold parameters, proportionality coefficients, etc. are designed, they all belong to adjustable control parameters for different application environments. Their values depend on the target device configuration, data input characteristics, and performance optimization goals, and converge and are set within a reasonable range through model verification, performance constraints, or engineering calibration during the implementation stage. Although these parameters do not have a preset unique value, they have a clear adjustment logic and calculation path, belonging to a deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution has both general adaptability and reproducibility and operability, without affecting its technical clarity and implementability;

[0092] Define to represent the joint classification result of the three structural decision variables corresponding to the pixel point in the image, which is used to determine its structural region category in the image; Define to represent the joint classification result of the three structural decision variables corresponding to the pixel point in the image, which is used to determine its structural region category in the image; Define to represent the boundary direction consistency measurement value B1 of the pixel point of, which is used to measure whether the edge directions in its local neighborhood have the characteristic of main direction aggregation, and the higher the value, the stronger the boundary direction consistency; Define to represent the boundary strength gradient variation value B2 of the pixel point of, which represents whether there is an obvious intensity jump of this point in the multi-scale image and is used to reflect the severity of the edge structure; Define to represent the regional edge closure ratio B3 of the point in the image, which is used to measure whether the contour is closed and complete;

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] Among them to represent the neighborhood set of the pixel point of, and in practical applications includes the square region window for which is used to calculate the direction consistency; to represent the edge main direction angle of the pixel point in the neighborhood, and in practical applications the value of is calculated through the image edge gradient direction and the unit is radians; Represents the current pixel point The edge main direction of which is compared with its neighborhood direction to judge the direction consistency; Is the point The direction divergence suppression factor calculated in the local tensor field, which is used to enhance the weighting of the direction difference in the high direction fluctuation region and reflects the local instability of the direction; Is the multi-scale image pyramid set, which includes multiple resolution versions such as the original image, the reduced image and the enlarged image, and is used to extract the cross-scale consistent gradient jump behavior; Represents at the scale image, the gradient intensity modulus value of the pixel point ; Is the anisotropy adjustment factor, which is constructed according to the gradient ratio of the main direction and its orthogonal direction at the scale , and its expression is:

[0099] ;

[0100] Where Represents the orthogonal direction of the gradient, The addition of which is used to prevent the denominator from being zero as a very small positive number, and the anisotropy adjustment factor is used to enhance the response of those regions where the gradient direction jumps violently; Represents the set of pixel points on all boundary paths extracted starting from the pixel point by the boundary tracking algorithm, and the boundary tracking algorithm includes edge connection, Canny edge growth, etc.; Is the local connection score of the point on the path, and the local connection score includes three weighted dimensions: boundary direction consistency, local gray contrast, local gradient jump amplitude, and is fused to form a single-point connection quality value; Is the set of multiple closed boundary candidate paths formed by the region where the point is located, and the closed boundary candidate path set is used to analyze its boundary closure degree; Is the closed cost function, which represents the minimum structural cost value among multiple candidate closed paths, including multiple sub-items such as closed side length, contour curvature, direction jump penalty, etc., and the closed cost function is used to measure the cost of the "easiest closed path"; Is a very small positive number, which is used to avoid the mathematical calculation instability caused by the closed cost being zero; Is the joint classification function, and after inputting three structure determination variables in the joint classification function, it judges the structure category of the region according to the threshold combination logic and outputs the structure label.

[0101] For the image region structure in the locally directionally convergent mutation region, by constructing a non-linear path mutation metric value that fuses the tensor direction torsion rate, local tension gradient variation, and perturbation diffusion response , construct a fracture induction function to determine potential boundary fracture points in the image, and accordingly guide the subsequent erosion and dilation processes to form a structural contour path;

[0102] ;

[0103] ;

[0104] where the fracture induction function represents whether the pixel point in the image is marked as a structural fracture induction point, and the output is a Boolean value of , 0 indicating that the pixel point is not determined as a structural fracture point, and 1 indicating that it is determined as a fracture point; respectively represent the principal direction angles extracted from the current pixel point and its neighboring pixel points in the tensor atlas, used to capture the sharp rotation in the direction field; represents the directed angular difference between the principal directions (i.e., the angular torsion rate), non-linearly describing the rotation direction and amplitude of the principal axis change; is a tension function, and the tension function represents the contraction or expansion trend of the image structure boundary, generated by the density change in the structural path; are respectively the gradients of the structural tension in the horizontal and vertical directions, and the two are used to reflect the local structural tension perturbation; is a direction consistency function, and the direction consistency function is used to characterize the continuity of the principal axis direction of the tensor in the image boundary structure; is the mixed second-order partial derivative of the direction consistency, and the mixed second-order partial derivative of the direction consistency represents the asymmetric perturbation of the consistency jump in the two-way structural change; is a direction mutation intensity factor, and the direction mutation intensity factor is used to extract the gradient jump amplitude information from the boundary response set, measuring the sharp turning behavior of the edge; is the local response rate of the direction diffusion tensor, and the local response rate of the direction diffusion tensor is used to characterize the expansion or contraction effect of the boundary information on the principal axis of the direction tensor; additionally, in the formula represents a non-linear expansion term for constructing the structural perturbation propagation intensity, used to enhance the selective discrimination ability of the boundary mutation points; is a structural mutation joint determination threshold, and the structural mutation joint determination threshold is used to comprehensively evaluate the lower limit of the response of fracture sensitivity;

[0105] Input all continuously closed high-consistency regions into the structural potential energy construction process to generate a structural potential energy trajectory function and a reference guidance path function , and in combination with a closed perturbation function Construct a cleavage scoring function ; if the score is higher than a preset threshold , then this point is marked as a true fracture path point, and the cleavage path set is output for topological division and pore reconstruction;

[0106] ;

[0107] ;

[0108] In in the formula represents the sub-pixel coordinates of the currently processed pixel in the image; the structural potential energy trajectory function represents the potential energy-driven path within the closed region; , , and represent the first-order derivative along ; , is the second-order partial derivative of , , used to calculate the change in local structural curvature; the reference guiding path function is used as a comparison reference path for the potential energy path ; , is the derivative of in two directions; the closed perturbation function is a scoring function that fuses three factors and is used as the basis for judging whether a fracture path is formed; the preset threshold is the path cleavage scoring determination threshold; is the binary determination function for the fracture path, the output of which is 1 indicating that it is recognized as a pore cleavage point, otherwise 0; in addition, in the cleavage scoring function represents the structural curvature mutation term; in the cleavage scoring function represents the path guiding offset term; in the cleavage scoring function represents the closed perturbation tension term; among them, the sub-pixel coordinates do not refer to the original pixel points in the image, but are continuous positions estimated inside the pixel grid through interpolation or fitting, etc., and are used for higher-precision positioning of image structural features such as edges, corners or paths, so as to achieve fine-grained extraction and determination of key structures in image analysis;

[0109] In addition, in this solution , The construction is a further solution to the fracture path determination logic constructed based on the two structural determination variables of "path tension value" and "structural boundary overlap degree". The constructed unified cracking scoring function integrates the above two structural determination variables and the structural potential energy response variable, and transforms them into a single scoring mechanism for path recognition. Thus, on the basis of keeping the path validity judgment logic in the original solution unchanged, it realizes the structural optimization expression of variable integration and continuous determination process.

[0110] Construct candidate boundaries for the initial pore size structure map 1 and the initial pore size structure map 2 extracted from the corrosion expansion path and the vein backtracking path respectively, and construct a boundary matching scoring matrix based on three types of scoring quantities of the boundary coincidence degree R1, the direction alignment degree R2, and the structural center offset amount R3 to judge the boundary fusion condition;

[0111] Through to define the scoring of the boundary coincidence degree R1:

[0112] ;

[0113] Through to define the scoring of the direction alignment degree R2:

[0114] ;

[0115] Through to define the scoring of the structural center offset amount R3:

[0116] ;

[0117] In , , in the formula represents the th pair of candidate boundaries; is the length of the overlapping part of the pair of candidate boundaries in the initial pore size structure map 1 and the initial pore size structure map 2; is the total length of the union of the pair of candidate boundaries in the initial pore size structure map 1 and the initial pore size structure map 2; is the difference in the image edge gradient change in the overlapping region of the pair of candidate boundaries ; is the corresponding topological fracture complexity index of the pair of candidate boundaries ; , is the main direction angle of the pair of candidate boundaries in the corrosion expansion path map and the vein backtracking path map; is the candidate boundary pair 's average curvature; , is the candidate boundary pair 's geometric center horizontal and vertical offset; is the candidate boundary pair 's structural symmetry coefficient.

[0118] For the image block marked as the boundary feature determination fuzzy area, establish a path confidence index and a scoring consistency difference index indexed by the boundary structure, and build a dynamic callback mechanism based on the two to update the image path and boundary judgment; define the path confidence as ; define the scoring consistency difference as ;

[0119] ;

[0120] ;

[0121] where is the boundary layer tension response extracted at the pixel point ; is the difference value of the image processing strategy at the pixel point under two paths; is the score of the pixel point under the erosion and dilation path; is the score of the pixel point under the vein backtracking path; where the scores under the path include three dimensions: boundary coincidence degree, direction alignment degree, and structural center offset, corresponding to R1, R2, and R3 respectively. These three indicators together constitute the matching evaluation basis for each candidate boundary pair. Therefore, there are three scores for each path respectively; the scoring consistency difference is used to dynamically judge path conflicts; the path confidence is used to judge whether the path update is reasonable.

[0122] It should be noted for the whole that the purpose of this solution is to solve the problem of pore adhesion caused by boundary blur, tight pore wall fitting, or local collapse in the image processing of the foam concrete surface; traditional pixel-level pore size recognition methods are prone to misidentifying multiple actually independent pores as a whole when processing high-density continuous gray areas, resulting in distorted judgment of the material porosity and structural looseness; for this reason, the present invention proposes a fine pore size recognition method with structural recognition ability and adaptive correction ability by establishing a multi-stage image processing process with sub-pixel edge structure as the core;

[0123] The entire solution starts with the "structural determination of boundary response features" and constructs a region classification mechanism driven by structural determination variables such as boundary direction consistency, intensity jump variation, and edge closure ratio. Starting from the image boundary level, three types of structural regions are identified: locally directionally convergent mutation regions, continuously enclosed highly consistent regions, and regions with ambiguous boundary feature determination. Then, for the first two types of regions, boundary contour extraction paths driven by structure tensors and potential energy are designed respectively, and two sets of initial aperture maps are formed.

[0124] On this basis, to deal with the differences between the two maps, a fusion judgment mechanism is further constructed. A boundary matching score matrix is constructed through three scoring indicators: boundary coincidence degree, direction alignment degree, and structural center offset amount, to achieve the disambiguation and integration of redundant or ambiguous boundaries in the maps. Finally, for the remaining regions with ambiguous determination, a structure status table and a path score callback mechanism are constructed, and the paths and boundaries are corrected again according to the results of the determination variables, realizing the closed-loop optimization of structure recognition.

[0125] The core of the solution lies in the collaborative use of multi-path joint judgment and structural closed-loop correction mechanisms. First, in image processing, three structural determination variables are extracted through boundary response features. These three variables analyze the essence of the image structure from three dimensions: direction consistency, intensity mutation, and geometric closure, and establish a preliminary structural classification of the image region.

[0126] In the regions with clear structures, for the locally directionally convergent mutation regions, by fitting the main axis change trend of the direction tensor field, the local direction torsion rate and abnormal points of the structural tension gradient are extracted, and erosion and finite dilation operations in the main axis direction are performed. Finally, a contour set guided by structural fractures is obtained.

[0127] For the continuously enclosed highly consistent regions, by constructing a structural potential energy field and guiding path search, path tension evaluation and boundary overlap degree judgment are performed on each potential path, so as to screen out the real fracture paths, and then the pore structure is divided through topological reconstruction.

[0128] Subsequently, the above two types of maps are input into the fusion judgment module, a boundary matching score matrix is constructed, and based on the three scoring indicators - coincidence degree, alignment degree, and offset amount, operations such as fusion, retention, or re-judgment are performed on the boundaries.

[0129] For all regions with ambiguous determination, the established structure status table is used to record the scoring paths and execution status corresponding to the boundary structures. When there is a conflict in scoring consistency or the path confidence level is lower than the threshold, the system will automatically call the determination variables to re-identify the structure type of the boundary and specify a new image processing path, finally realizing path update, layer replacement, and closed-loop correction.

[0130] The design of this solution not only breaks the dependence on pixel-level edges in traditional aperture recognition, but also proposes an image processing logic oriented to the "synergistic expression of structural continuity and fracture mutation". Its core parts include:

[0131] Explicitly classify image regions using structure determination variables to enhance the ability to recognize image structures;

[0132] Respectively guide the boundary extraction paths of the two types of regions through structure tensors and potential energy functions to form a highly targeted heterogeneous graph construction mechanism;

[0133] Introduce a matching score matrix to fuse the two types of graphs to enhance the ability to judge the synergy and structural consistency between paths;

[0134] Construct a structure state table and a path score callback mechanism to form an image processing flow with closed-loop feedback ability, which can dynamically adapt to the uncertainty of fuzzy region recognition.

[0135] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An image processing method for foamed concrete based on sub-pixel edge reconstruction, characterized in that, Including: Obtain an image of foamed concrete, and extract a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information; Based on the set of boundary response features, perform edge structure classification on the image region by combining threshold judgment; For the image region determined to be a locally directionally convergent mutation region, identify the set of cross-mutation points by fitting the convergence trend of the direction tensor, and perform image processing of directional guided erosion and dilation; For the image region determined to be a continuously closed high-consistency region, construct a structural potential field and jointly determine the path tension and boundary coincidence degree, identify the true fracture path, and perform topological division accordingly; For the pore size structure maps respectively output by the image processing of the erosion and dilation path and the vein backtracking path, perform cross-matching judgment on the three consistency scoring items in the boundary matching scoring matrix, and selectively integrate or retain the structure difference regions according to the fusion conditions; Construct a structure determination mechanism by fusing the path and the scoring result, perform path update and callback correction on the boundaries marked as fuzzy regions in the image, and realize the identification of the pore size structure and the optimization of the image processing flow.

2. The method for processing an image of foamed concrete based on sub-pixel edge reconstruction according to claim 1, wherein: Obtain the original image data of the foamed concrete image, perform standardized gray processing and multi-scale edge enhancement operations on the original image data to obtain a set of boundary response features including edge direction information, edge gradient change information, and local intensity contrast information; Based on the set of boundary response features, extract three structure determination variables, and the structure determination variables include the boundary direction consistency measurement value B1, the boundary intensity gradient variation value B2, and the regional edge closure ratio B3; If B1 is less than the boundary direction consistency measurement threshold T1 and B2 is greater than the boundary intensity gradient variation threshold T2, the image region is marked as a locally directionally convergent mutation region; if B1 is greater than or equal to T1 and B3 is greater than the regional edge closure ratio threshold T3, the image region is marked as a continuously closed high-consistency region; if the three are in a critical cross state, the image region is marked as a boundary feature determination fuzzy region; Temporarily store the image blocks marked as boundary feature determination fuzzy regions as fuzzy region masks, and jointly include them in the fusion judgment process with the image blocks marked as locally directionally convergent mutation regions and the image blocks marked as continuously closed high-consistency regions to perform unified determination of the path attribution of the three types of regions.

3. The method for processing an image of foamed concrete based on sub-pixel edge reconstruction according to claim 2, wherein: Input all locally directionally convergent mutation regions into the structural direction tensor extraction process, construct a tensor map of the corresponding main axis through the set of boundary response features, perform convergence trend fitting of the main axis direction on the tensor map, generate a direction tensor convergence map, and identify the aggregated boundary region; Based on image analysis, identify the set of points with abrupt changes in the direction gradient in the aggregation boundary region as the fracture candidate subset, and perform tensor cross-angle analysis on each candidate point in the fracture candidate subset; if the cross angle is greater than the preset fracture criterion, then this point is identified as a valid fracture point and is used to generate a fracture-induced layer; Perform a direction tensor weighted erosion operation based on the fracture-induced layer, and then perform a limited dilation operation that maintains the principal axis direction to preserve the boundary continuity and spatial consistency formed by the fracture; perform a structural contour extraction operation on the processed region, form a contour set by all the extracted boundary contours, and record the processing path label corresponding to each contour at the same time; Output the obtained contour set as the initial pore size structure map one under the erosion-dilation path.

4. The method for processing foam concrete images based on sub-pixel edge reconstruction according to claim 3, wherein: Input all continuous closed high-consistency regions into the structural potential construction process, generate a sub-pixel level structural potential field according to the edge strength and the closing stability, perform a guided path search in the structural potential field, obtain a set of potential fracture paths, and calculate the path tension value and the structural boundary overlap degree for each path in turn; If the path tension value is greater than the path tension threshold T4 and the structural boundary overlap degree is less than the structural boundary overlap degree threshold T5, then this path is confirmed as a real fracture path and serves as the boundary cracking trigger path; Perform a topological structure reconstruction operation on the original image using the confirmed real fracture path, and divide it into multiple logically pore sub-blocks; Mark the pore structures of the divided pore sub-blocks with path labels and output them as the initial pore size structure map two under the vein backtracking path.

5. The method for processing foam concrete images based on sub-pixel edge reconstruction according to claim 4, wherein: Merge the sets of the initial pore size structure map one and the initial pore size structure map two of the erosion-dilation path and the vein backtracking path, and construct a candidate boundary set for the boundary pairs in the two maps; for each pair of candidate boundaries in the candidate boundary set, calculate the boundary coincidence degree R1, the direction alignment degree R2, and the structural center offset amount R3, and form a boundary matching score matrix; In the boundary matching score matrix, if R1 in a certain boundary pair is greater than the boundary coincidence degree threshold T6, and R2 and R3 are less than the direction alignment degree threshold T7 and the structural center offset amount threshold T8, then this boundary pair is determined as a fusion boundary, perform layer merging and assign a unified number; If the boundary matching score matrix does not meet the fusion boundary determination criterion, then this image region is retained as a boundary feature determination fuzzy region; perform sub-pixel curve fitting on the boundaries of all determined image regions, and output a vector boundary set and the corresponding path label information.

6. The method for processing foam concrete images based on sub-pixel edge reconstruction according to claim 5, wherein: Establish a path label, a scoring result of a boundary matching scoring matrix, and a structure status table with each boundary structure in the contour set as the main index. The structure status table is extended based on the scoring result of the boundary matching scoring matrix by introducing a path execution status and a region fusion determination result. Record the original policy path and the scoring matching value corresponding to each boundary structure in the structure status table; For all image regions marked as regions with ambiguous boundary feature determination, evaluate the path determination confidence and the scoring consistency difference, and determine whether there is aperture recognition uncertainty; If the path confidence is less than the rejudgment threshold T9 and the scoring consistency difference is greater than the conflict tolerance threshold T 10 , then re-perform the boundary structure determination based on the three structural determination variables to complete the missing or misclassified aperture boundaries; According to the new round of structure determination result, re-specify the image structure path, and execute the corresponding image processing method to obtain the updated aperture boundary extraction result; the image structure path includes an erosion and dilation path or a vein backtracking path; replace the corresponding region in the original boundary layer with the updated image structure boundary, and synchronously update its corresponding path identification information.

7. The method for processing images of foamed concrete based on sub-pixel edge reconstruction according to claim 6, wherein: Definition Represents the combined classification result of three structural determination variables corresponding to a pixel point in the image; define Represents the combined classification result of three structural determination variables corresponding to a pixel point in the image; define Represents a pixel point The boundary direction consistency measurement value B1 of; define Represents a pixel point The boundary strength gradient variation value B2 of; define Represents the point in the image The region edge closure ratio B3 to which it belongs, Used to measure whether the contour is closed and complete; ； ； ； ； ； in Represents pixel The neighborhood set of ; Represents the pixel in the neighborhood The main direction angle of the edge; Indicates the current pixel The main direction of the edge; For point Directional divergence suppression factor calculated in the local tensor field; is a multi-scale image pyramid set; Indicated in Pixels in the scaled image The gradient intensity modulus of ; is the anisotropy adjustment factor; Indicates that the boundary tracing algorithm is used to trace from pixel points Starting from, extract the pixel set on all boundary paths; Point on the path The local connectivity score of For point A set of multiple closed boundary candidate paths formed by the area; is a closed cost function; is a very small positive number; is the joint classification function.

8. The method for processing images of foamed concrete based on sub-pixel edge reconstruction according to claim 7, wherein: For the image region structure in the locally directionally convergent mutation region, a non-linear path mutation metric value that fuses the tensor direction torsion rate, local tension gradient variation, and perturbation diffusion response is constructed , and a fracture induction function is constructed to determine potential boundary fracture points in the image, and based on this, guide the subsequent erosion and dilation processes to form a structural contour path; ； ； Among them, the fracture induction function indicates whether the pixel point in the image is marked as a structural fracture induction point, and the output is a Boolean value; respectively represent the principal direction angles extracted from the current pixel point and its neighboring pixel points in the tensor map; represents the directed angular difference between the principal directions; is the tension function; are the gradients of the structural tension in the horizontal and vertical directions respectively; is the direction consistency function; is the mixed second-order partial derivative of the direction consistency; is the direction mutation intensity factor; is the local response rate of the direction diffusion tensor; additionally, in the formula represents the non-linear expansion term for constructing the structural perturbation propagation intensity; is the structural mutation joint determination threshold; Input all continuous closed high-consistency regions into the structural potential construction process to generate a structural potential trajectory function and the reference guiding path function , and combine with the closed perturbation function to construct a cleavage scoring function ; If the score is higher than the preset threshold , then this point is marked as a true fracture path point, and the cleavage path set is output for topological division and pore reconstruction; ； ； In in the formula represents the sub-pixel coordinates of the currently processed pixel in the image; the structural potential energy trajectory function represents the potential energy-driven path within the closed region; , , and represent along the first derivative; , is the second partial derivative of , used to calculate the change in local structural curvature; the reference guiding path function is used as the comparison reference path for the potential energy path ; , is the derivative in two directions; the closed perturbation function represents the sub-pixel perturbation tension residual function of the path points in the closed region; the cleavage scoring function is a scoring function that fuses three factors and is used as the basis for judging whether a fracture path is formed; the preset threshold is the path cleavage scoring determination threshold; is the binary determination function for the fracture path, and the output of being 1 indicates that it is recognized as a pore cleavage point, otherwise it is 0.

9. The method for processing images of foamed concrete based on sub-pixel edge reconstruction according to claim 8, wherein: Construct candidate boundaries for the initial aperture structure map one and the initial aperture structure map two extracted from the erosion and dilation path and the vein backtracking path respectively, and construct a boundary matching scoring matrix based on three types of scoring quantities: the boundary coincidence degree R1, the direction alignment degree R2, and the structure center offset amount R3, to judge the boundary fusion condition; By to define the scoring of the boundary coincidence degree R1: ； By to define the scoring of the direction alignment degree R2: ； By to define the scoring of the structural center offset R3: ； In , , in the formula, represents the th pair of candidate boundaries; is the pair of candidate boundaries the length of the overlapping part between the first initial aperture structure map and the second initial aperture structure map; is the pair of candidate boundaries the total length of the union of the first initial aperture structure map and the second initial aperture structure map; is the pair of candidate boundaries the difference in the change of the image edge gradient in the overlapping region; is the pair of candidate boundaries corresponding to the topological fracture complexity index; , is the pair of candidate boundaries the main direction angle in the corrosion and dilation path map and the vein backtracking path map; is the pair of candidate boundaries the average curvature of; , is the pair of candidate boundaries the horizontal and vertical offset of the geometric center of; is the pair of candidate boundaries the structural symmetry coefficient of.

10. The method for processing images of foamed concrete based on sub-pixel edge reconstruction according to claim 9, wherein: For the image blocks marked as the regions with fuzzy boundary feature determination, establish a path confidence index and a scoring consistency difference index indexed by the boundary structure, and construct a dynamic callback mechanism based on the two to update the image path and boundary judgment; define the path confidence as ; Define the difference in scoring consistency as ; ； ； wherein is the boundary layer tension response extracted at the pixel point ; is the difference value of the image processing strategies at the pixel point under two paths; is the score at the pixel point under the erosion and dilation path; is the score at the pixel point under the vein backtracking path.

Images