Method and system for safely and efficiently examining different types of contents by single system

Abstract

The invention relates to the technical field of security review, in particular to an efficient review method and system for security of different types of contents by a single system, and the method comprises the following steps: based on image input contents, carrying out size normalization processing, analyzing global color distribution, extracting histogram data, calling edge texture parameters to analyze local edge change and texture distribution, and carrying out feature extraction; and performing global and local feature vectorization coding, and performing fusion operation to generate a color and texture basic feature vector set. According to the method, local feature capture and detail analysis are enhanced through edge texture analysis, accurate modeling of a target area is realized through rule segmentation and spatial geometrical relationship analysis, feature matching and classification analysis are optimized through a category distribution dynamic score matrix, and a high-density area distribution center and a coverage area are accurately identified through density aggregation calculation. Layered analysis refines color, form and detail feature classification, and the judgment depth and review precision of complex content are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of security review technology, and in particular to a method and system for efficiently reviewing the security of different types of content by a single system. Background Art

[0002] The field of security review technology includes various methods and technologies for protecting and managing data, information and content. The core content of this technical field is to ensure the legality, compliance and security of content through the collection, analysis, classification and filtering of information. The overall technical field covers multiple sub-fields including information security, network security, data encryption, and content filtering, with the focus on reviewing and supervising different forms of digital content (such as text, images, audio and video, program code) to prevent the spread of illegal, bad or harmful content. This technical field usually relies on the coordinated application of computer information processing technology, network transmission technology and data storage technology.

[0003] Among them, a method for efficiently reviewing the security of different types of content by a single system refers to a technical method that can implement security reviews for multiple types of digital content within a single system. The subject matter of this patent covers technical matters such as the reception, decoding, parsing, analysis, and review of different types of content. Specifically, different types of content (such as text, images, audio and video files, etc.) are classified and key features are extracted through a predefined set of rules or a set of algorithms, and then the legality of the content is judged and marked according to the matching rules, and corresponding measures are taken according to the judgment results to complete the security review. This method usually relies on a data processing module to implement the parsing of different types of content, and completes efficient review of multiple contents by matching rules.

[0004] When dealing with multi-type digital content, existing technologies lack the ability to extract consistent features of multi-resolution content, which makes it difficult to balance global and local features, and there is a disconnect between detail capture and overall analysis. In the geometric feature modeling of the target area, existing technologies usually rely only on simple segmentation and morphological extraction, and fail to fully analyze the complexity of boundary connectivity and spatial relationships, resulting in inaccurate geometric morphological features. In feature matching analysis, existing technologies lack the flexibility of dynamic score generation and are difficult to effectively adapt to the complexity and dynamic changes of multi-category feature distribution. When dealing with high-density feature areas, existing technologies fail to fully identify the distribution center and coverage, resulting in blind spots in the analysis of dense feature areas. This deficiency affects the ability to comprehensively judge high-risk content. Lack of support for hierarchical analysis and classification refinement makes it difficult to effectively distinguish the differences in detail features, and the accurate classification of complex target content is limited. This deficiency may lead to insufficient accuracy in content classification and risk review, thereby affecting the reliability and depth of the review. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a method and system for efficiently reviewing the security of different types of content by a single system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for efficiently reviewing the security of different types of content by a single system, comprising the following steps:

[0007] S1: Based on the image input content, perform size normalization, analyze the global color distribution, extract histogram data, call edge texture parameters to analyze local edge changes and texture distribution, perform global and local feature vectorization encoding, and perform fusion operations to generate a set of color and texture basic feature vectors;

[0008] S2: Based on the color and texture basic feature vector set, the target area is segmented according to preset rules, the spatial geometric relationship is analyzed and modeled, the boundary connectivity of the segmented area is analyzed, and the geometric structure is aggregated and calculated according to the feature quantification to generate a spatial geometric feature set;

[0009] S3: Based on the spatial geometric morphological feature set, extract target features, perform feature hierarchical analysis and semantic vector matching, calculate category distribution relationships and dynamic scores through vector matching rules, and generate a dynamic matching score matrix based on the distribution results;

[0010] S4: Based on the dynamic matching score matrix, a risk threshold is set, target areas with risk value parameters meeting the conditions are screened, characteristic distribution center point parameters are called, density aggregation operations are performed, characteristic distribution coverage and overlap rate of the target area are calculated, target area distribution centers and coverage parameters are summarized, and a high-density characteristic distribution parameter set is generated;

[0011] S5: Based on the high-density feature distribution parameter set, the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area are analyzed hierarchically, the difference value distribution is calculated, and the target content is classified, and the hierarchical feature difference distribution judgment result is generated in combination with the classification system.

[0012] As a further solution of the present invention, the color and texture basic feature vector set includes a global color distribution vector, a local edge feature vector and a texture feature vector, the spatial geometric morphology feature set includes segmentation area boundary connectivity features, geometric morphology features and spatial geometric relationship features, the dynamic matching score matrix includes category distribution parameters, dynamic score parameters and vector matching parameters, the high-density feature distribution parameter set includes distribution center point parameters, coverage range parameters and overlap rate parameters, and the hierarchical feature difference distribution judgment results include color difference features, geometric morphology difference features, detail pattern difference features and classification feature content.

[0013] As a further solution of the present invention, based on the image input content, size normalization processing is performed, global color distribution is analyzed, and histogram data is extracted. The edge texture parameters are called to analyze local edge changes and texture distribution, and global and local feature vector quantization encoding is performed, and fusion operations are performed to generate a color and texture basic feature vector set. The specific steps are:

[0014] S101: Based on the input image, perform size normalization processing, adjust the image aspect ratio and pixel density, analyze the overall color distribution, count the number of pixels of each color, calculate the color proportion value, and analyze the distribution gradient change to generate a global color distribution matrix;

[0015] S102: Based on the global color distribution matrix, the spatial position and adjacent color distribution of each pixel region in the image are analyzed, the edge direction change of each region is extracted, the texture features are vectorized and encoded through the position coordinates, the texture pattern of the local region is analyzed in combination with the pixel intensity distribution, and a local edge texture matrix is ​​generated;

[0016] S103: Based on the global color distribution matrix and the local edge texture matrix, the global color features and the local texture features are vectorized and the two features are combined according to corresponding dimensions to generate a color and texture basic feature vector set.

[0017] As a further solution of the present invention, based on the color and texture basic feature vector set, the target area is segmented according to preset rules, the spatial geometric relationship is analyzed and modeled, the boundary connectivity of the segmented area is analyzed, and the geometric structure is aggregated and calculated according to the feature quantification to generate the spatial geometric feature set. The specific steps are:

[0018] S201: Based on the color and texture basic feature vector set, extract image pixel data according to the preset, assign each pixel to a corresponding area according to its color and texture value, analyze the assignment result, locate the boundary range of each area, establish the center position coordinates of the area, and generate a set of spatial segmentation areas;

[0019] S202: Based on the set of spatial segmented regions, the boundary lines of the regions and their adjacent relationships are analyzed, a geometric description of the region boundaries is constructed according to the coordinate positions and connection relationships of the boundary vertices, boundary intersections and connection paths are analyzed, the connectivity mode of each boundary and the relationship between regions are calculated, and a spatial boundary connectivity matrix is ​​generated;

[0020] S203: Based on the spatial boundary connectivity matrix, the geometric morphological features of each region are calculated, the area, shape ratio and boundary complexity of the region are analyzed, the features are combined to form a geometric morphological model of the entire region, the morphology is summarized according to the parameter data of the segmented region, and a set of spatial geometric morphological features is generated.

[0021] As a further solution of the present invention, the complexity parameter calculation formula is specifically:

[0022]

[0023] Among them, C represents the region boundary complexity parameter, n represents the total number of regions involved in the calculation, and B i represents the boundary length of the ith region, A i represents the area of ​​the ith region, D i represents the boundary feature difference value of the i-th region, Represents the mean of the difference values ​​of all regional boundary characteristics.

[0024] As a further solution of the present invention, based on the spatial geometric morphological feature set, target features are extracted, feature hierarchical parsing and semantic vector matching are performed, and the category distribution relationship and dynamic score are calculated by vector matching rules. The specific steps of generating a dynamic matching score matrix according to the distribution result are as follows:

[0025] S301: Based on the spatial geometric morphology feature set, the vector data of each feature is parsed to extract the boundary range and spatial distribution of the feature, and the similarity value is calculated according to the difference between the features. The core feature value is extracted by summarizing the similarity to generate a target feature vector set;

[0026] S302: Based on the target feature vector set, construct a feature hierarchy structure, analyze the semantic attributes of each layer of feature vectors and the association between the upper and lower layers, calculate the association scores between feature vectors according to the semantic attribute relationship, and integrate the calculation results to form a semantic matching score matrix;

[0027] S303: Based on the semantic matching score matrix, analyze the distribution of vector scores in each category, calculate weight scores according to the vector category distribution ratio, integrate the category distribution and weight scores into the matrix, and generate a dynamic matching score matrix.

[0028] As a further solution of the present invention, the correlation score calculation formula is specifically:

[0029]

[0030] Among them, S represents the correlation score of the feature vector, W represents the dimension of the feature vector, and u k and v k They represent the value of the feature vector in the kth dimension, w k Represents the weight factor of the kth dimension, which is obtained by normalizing the importance of the dimension.

[0031] As a further solution of the present invention, based on the dynamic matching score matrix, a risk threshold is set, target areas with risk value parameters meeting the conditions are screened, characteristic distribution center point parameters are called, density aggregation operation is performed, characteristic distribution coverage and overlap rate of the target area are calculated, target area distribution center and coverage parameters are summarized, and specific steps of generating a high-density characteristic distribution parameter set are as follows:

[0032] S401: Based on the dynamic matching score matrix, analyze the risk value of each target area, screen the regional risk parameters according to the preset risk threshold, extract the boundary parameters and characteristic center points of the screened area, and integrate the boundary data and the center point position to form a risk screening area set;

[0033] S402: Based on the risk screening area set, calling the spatial coordinates of the center points of regional features, analyzing the distribution density of the center points, calculating the distance matrix between the center points, identifying the feature overlap relationship between adjacent areas, and generating regional distribution coverage parameters;

[0034] S403: Based on the regional distribution coverage parameters, analyze the coverage range and feature overlap rate of each target region, calculate the feature distribution data inside and outside the region, summarize the feature density, coverage range and center distribution of each region, and generate a high-density feature distribution parameter set.

[0035] As a further solution of the present invention, based on the high-density feature distribution parameter set, the color distribution characteristics, geometric features and detail pattern characteristics of the target area are analyzed hierarchically, the difference value distribution is calculated, and the target content is classified. The specific steps of generating the hierarchical feature difference distribution determination result in combination with the classification system are as follows:

[0036] S501: Based on the high-density feature distribution parameter set, analyze the color distribution characteristics of the target area, count the spatial distribution law of the color value, calculate the distribution ratio data of the color characteristics in the differentiated area, identify the change gradient, analyze the contrast relationship between colors according to the difference data, and generate a color feature distribution analysis result;

[0037] S502: Based on the color feature distribution analysis result, the geometric morphological features of the target area are analyzed, the boundary length and area ratio of the morphological parameters are calculated, the difference between the morphological features is determined, the geometric features are classified into types, and a geometric morphological classification analysis result is generated;

[0038] S503: Based on the geometric morphology classification analysis result, the pattern features of the target area are analyzed, the boundary lines and texture distribution of the pattern features are analyzed, the contrast ratio of the pattern features between regions is calculated, the multi-level feature data of the region is integrated, and the hierarchical feature difference distribution determination result is generated.

[0039] A single system for efficient review of different types of content security, including:

[0040] The normalization module performs size normalization based on the image input content, analyzes the global color distribution, extracts histogram data, calls edge texture parameters to analyze local edge changes and texture distribution, performs global and local feature vectorization encoding, and generates a set of color and texture basic feature vectors;

[0041] The segmentation modeling module segments the target area according to preset rules based on the color and texture basic feature vector set, performs spatial geometric relationship analysis and modeling, analyzes the boundary connectivity of the segmented area, aggregates and calculates the geometric morphological structure, and generates a spatial geometric morphological feature set;

[0042] The vector matching module performs feature hierarchical analysis and semantic vector matching based on the spatial geometric morphological feature set, calculates the category distribution relationship and dynamic score through the vector matching rule, and generates a dynamic matching score matrix according to the distribution result;

[0043] The risk screening module sets the risk threshold based on the dynamic matching score matrix, screens the target area with the risk value parameter meeting the conditions, calls the characteristic distribution center point parameter, performs density aggregation operation, calculates the characteristic distribution coverage and overlap rate of the target area, and generates a high-density characteristic distribution parameter set;

[0044] The hierarchical analysis module hierarchically analyzes the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area based on the high-density feature distribution parameter set, calculates the difference value distribution, and generates a hierarchical feature difference distribution determination result.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, local feature capture and detail analysis are enhanced through edge texture analysis, accurate modeling of the target area is achieved through rule segmentation and spatial geometric relationship analysis, feature matching and classification analysis are optimized through category distribution dynamic score matrix, density aggregation calculation accurately identifies the distribution center and coverage of high-density areas, and layered analysis refines the classification of color, shape and detail features, which significantly improves the judgment depth and review accuracy of complex content. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1It is a schematic diagram of the steps of the present invention;

[0049] Figure 2 is a flow chart of the steps of S1 of the present invention;

[0050] Figure 3 is a flow chart of the steps of S2 of the present invention;

[0051] Figure 4 is a flow chart of the steps of S3 of the present invention;

[0052] Figure 5 is a flow chart of the steps of S4 of the present invention;

[0053] Figure 6 is a flow chart of the steps of S5 of the present invention;

[0054] Figure 7 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0056] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0057] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0058] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be written in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0059] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0060] See also Figure 1 , a method for efficiently reviewing the security of different types of content with a single system, including the following steps:

[0061] S1: Based on the image input content, perform size normalization, analyze the global color distribution, extract histogram data, call edge texture parameters to analyze local edge changes and texture distribution, perform global and local feature vectorization encoding, and perform fusion operations to generate a set of color and texture basic feature vectors;

[0062] S2: Based on the color and texture basic feature vector set, the target area is segmented according to preset rules, the spatial geometric relationship is analyzed and modeled, the boundary connectivity of the segmented area is analyzed, and the geometric structure is aggregated and calculated according to the feature quantification to generate a spatial geometric feature set;

[0063] S3: Based on the spatial geometric morphological feature set, extract the target features, perform feature hierarchical analysis and semantic vector matching, calculate the category distribution relationship and dynamic score through vector matching rules, and generate a dynamic matching score matrix based on the distribution results;

[0064] S4: Based on the dynamic matching score matrix, set the risk threshold, screen the target areas that meet the risk value parameters, call the characteristic distribution center point parameters, perform density aggregation operations, calculate the characteristic distribution coverage and overlap rate of the target area, summarize the distribution center and coverage parameters of the target area, and generate a high-density characteristic distribution parameter set;

[0065] S5: Based on a set of high-density feature distribution parameters, the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area are analyzed hierarchically, the difference value distribution is calculated, and the target content is classified. The hierarchical feature difference distribution judgment result is generated in combination with the classification system.

[0066] The color and texture basic feature vector set includes global color distribution vector, local edge feature vector and texture feature vector; the spatial geometric morphology feature set includes segmentation area boundary connectivity features, geometric morphology features and spatial geometric relationship features; the dynamic matching score matrix includes category distribution parameters, dynamic score parameters and vector matching parameters; the high-density feature distribution parameter set includes distribution center point parameters, coverage range parameters and overlap rate parameters; the hierarchical feature difference distribution judgment results include color difference features, geometric morphology difference features, detail pattern difference features and classification feature content.

[0067] See also Figure 2 , the specific steps of S1 are:

[0068] S101: Based on the input image, perform size normalization processing, adjust the image aspect ratio and pixel density, analyze the overall color distribution, count the number of pixels of each color, calculate the color proportion value, and analyze the distribution gradient change to generate a global color distribution matrix;

[0069] The initial aspect ratio and resolution are obtained by parsing the metadata of the input image, and the image interpolation algorithm is used to adjust the distribution of pixels to match the specified target resolution while ensuring that the aspect ratio remains consistent. The pixel intensity analysis technology is used to perform statistical processing on the overall color distribution of the image. The color quantization model is used to divide the color types and the number of pixels of each color is counted one by one. The normalization operation is used to calculate the pixel ratio of each color and generate a proportional histogram of the color distribution. The gradient calculation method is used to further analyze the color gradient changes to obtain the global color distribution matrix.

[0070] S102: Based on the global color distribution matrix, the spatial position and adjacent color distribution of each pixel region in the image are analyzed, the edge direction change of each region is extracted, the texture features are vectorized and encoded through the position coordinates, the texture pattern of the local region is analyzed in combination with the pixel intensity distribution, and a local edge texture matrix is ​​generated;

[0071] The spatial position of each pixel area in the image is extracted, the boundary lines between pixel areas are identified by the boundary detection algorithm and the corresponding edge map is generated. The directional changes of the edges between adjacent areas are extracted by combining the piecewise fitting method. The texture features of each area are encoded and represented by the geometric vectorization method. The local distribution statistics method of pixel intensity is used to calculate the change pattern of pixel values ​​in a specific area and generate a statistical distribution map of the texture pattern. Finally, the directional change information of the edge of the area and the feature description of the texture pattern are combined to generate a local edge texture matrix to represent the texture feature relationship between the interior of the area and between areas.

[0072] S103: Based on the global color distribution matrix and the local edge texture matrix, the global color features and the local texture features are vectorized and the two features are combined according to corresponding dimensions to generate a color and texture basic feature vector set;

[0073] The global color features and local texture features are vectorized and described according to the formula

[0074]

[0075] Compute the total eigenvalue of a set of color and texture basis eigenvectors.

[0076] Where F represents the total eigenvalue of the basic eigenvector, C i represents the global color feature value of the i-th dimension, T i represents the local texture eigenvalue of the i-th dimension, and n represents the dimension of the feature vector.

[0077] Global color eigenvalue C i Through color distribution matrix extraction, the calculation formula is:

[0078]

[0079] Among them, N c The number of pixels representing the color category, N t Indicates the total number of pixels in the image.

[0080] Local texture eigenvalue T i Through edge texture matrix extraction, the calculation formula is:

[0081]

[0082] Among them, S l represents the sum of the local texture change strength, L b Indicates the length of the region's boundary.

[0083] Specific calculation example:

[0084] Assume that the image distribution matrix has three-dimensional features and the color eigenvalue is C 1 =0.2,C 2 =0.3,C 3 = 0.1, the texture feature value is T 1 =0.4,T 2 =0.5,T 3 =0.6, the total eigenvalue F is calculated as follows:

[0085]

[0086] Substitute the specific values ​​into:

[0087]

[0088] The result shows that the total eigenvalue F reflects the comprehensive characteristics of the color and texture features of the image and can be used for further image classification or feature comparison analysis.

[0089] See also Figure 3 , the specific steps of S2 are:

[0090] S201: Based on the color and texture basic feature vector set, extract the image pixel data according to the preset, assign each pixel to the corresponding area according to its color and texture value, analyze the assignment result, locate the boundary range of each area, establish the center position coordinates of the area, and generate a set of spatial segmentation areas;

[0091] Analyze the color and texture value range of image pixel data, classify each pixel into the corresponding spatial area according to its color and texture characteristics by constructing allocation rules, calculate the boundary range of each area based on the pixel coordinate data, and use the geometric center of gravity formula to aggregate the pixel position of each area, determine the center point position of the area and mark the area number, merge all area data and generate a set of spatial segmentation areas.

[0092] S202: Based on the set of spatially segmented regions, the boundary lines of the regions and their adjacent relationships are analyzed, a geometric description of the region boundaries is constructed according to the coordinate positions and connection relationships of the boundary vertices, boundary intersections and connection paths are analyzed, the connectivity mode of each boundary and the relationship between regions are calculated, and a spatial boundary connectivity matrix is ​​generated;

[0093] The vertex coordinates of the regional boundary are extracted, and the starting and ending points of the boundary are identified using boundary detection and point-to-point connection methods. The connection relationship of the boundary is recorded and geometric vectorization is performed. The connectivity relationship between the boundary intersections between regions is analyzed, and the connection paths of the intersections are classified based on topological sorting and a connectivity pattern data table between regions is generated. The spatial boundary connectivity matrix is ​​constructed based on the geometric relationship between the boundary and the intersection.

[0094] S203: Based on the spatial boundary connectivity matrix, the geometric morphological feature quantity of each region is calculated, the area, shape ratio and boundary complexity of the region are analyzed, the feature quantities are combined to form a geometric morphological model of the entire region, the morphology is summarized according to the parameter data of the segmented region, and a spatial geometric morphological feature set is generated;

[0095] The calculation formula of complexity parameter is as follows:

[0096]

[0097] Among them, C represents the region boundary complexity parameter, n represents the total number of regions involved in the calculation, and B i represents the boundary length of the ith region, A i represents the area of ​​the ith region, D i represents the boundary feature difference value of the i-th region, Represents the mean of the difference values ​​of all regional boundary characteristics.

[0098] Parameter C represents the regional boundary complexity parameter, which is a comprehensive quantitative index of regional boundary characteristics and area. n represents the total number of regions involved in the calculation, and B i represents the boundary length of the ith region, which is obtained by accumulating the length of the boundary line segments. i represents the area of ​​the ith region, which is calculated by the pixel area. irepresents the boundary feature difference value of the ith region, which is calculated by comparing the difference between the region boundary and the standard feature. D￣ represents the mean of the boundary feature difference values ​​of all regions, which is calculated by D i Find the arithmetic mean.

[0099] Given the actual data as follows:

[0100] The total number of regions involved in the calculation is n = 3, and the boundary length of each region is B 1 =12,B 2 =15,B 3 =10 (unit: pixel length), area of ​​each region A 1 =30,A 2 =35,A 3 = 25 (unit: pixel area), the boundary feature difference value D of each area 1 =5,D 2 =8,D 3 =6 (dimensionless).

[0101] Calculate the mean

[0102]

[0103] Calculate the complexity term for each region

[0104] For the first region:

[0105]

[0106] For the second region:

[0107]

[0108] For the third region:

[0109]

[0110] Calculate the total complexity parameter C:

[0111]

[0112] The results show that the complexity parameter C is 2.606, which is used to quantitatively describe the relationship between the complexity of regional boundaries and area ratios. It can comprehensively reflect the complex characteristics of boundaries in spatial geometric morphological models and provide reliable parameter support for the subsequent induction of morphological models and generation of feature sets.

[0113] See also Figure 4 , the specific steps of S3 are:

[0114] S301: Based on the spatial geometric morphology feature set, the vector data of each feature is parsed to extract the boundary range and spatial distribution of the feature, and the similarity value is calculated according to the difference between the features. The core feature value is extracted by summarizing the similarity to generate the target feature vector set;

[0115] Extract the vector data of each feature, analyze the boundary range of the feature, calculate the difference between features through the statistical value of pixel distribution in the segmentation area, project multidimensional features such as color value and texture change into a unified vector space, calculate the similarity value using Euclidean distance or cosine similarity, summarize the similarity data and extract the core feature value through clustering algorithm, and finally integrate the core features to form a target feature vector set.

[0116] S302: Based on the target feature vector set, construct a feature hierarchy structure, analyze the semantic attributes of each layer of feature vectors and the association between the upper and lower layers, calculate the association scores between feature vectors according to the semantic attribute relationship, and integrate the calculation results to form a semantic matching score matrix;

[0117] The calculation formula of the correlation score is as follows:

[0118]

[0119] Among them, S represents the correlation score of the feature vector, W represents the dimension of the feature vector, and u k and v k They represent the value of the feature vector in the kth dimension, w k Represents the weight factor of the kth dimension, which is obtained by normalizing the importance of the dimension.

[0120] The parameter S represents the correlation score of the feature vector, W represents the dimension of the feature vector, and u k and v k They represent the value of the feature vector in the kth dimension respectively. The acquisition method is the numerical value of the feature component, which is extracted through the target feature vector set. k Represents the weight factor of the kth dimension, which is normalized and allocated according to the influence of each dimensional feature on the overall similarity.

[0121] Specific parameter values:

[0122] Assume W = 3, the eigenvector components of the three dimensions are u 1 =4,u 2 =7,u 3 =3 and v 1 =5,v 2 =6,v 3 =2. The dimension weight factors are w 1 =0.2,w 2 =0.5,w 3 =0.3.

[0123] Calculation process:

[0124] Calculate the value of each dimension

[0125] For the first dimension:

[0126]

[0127] For the second dimension:

[0128]

[0129] For the third dimension:

[0130]

[0131] Calculate the numerator:

[0132]

[0133] Calculate the denominator:

[0134]

[0135] Calculate the correlation score S:

[0136]

[0137] Result analysis:

[0138] The result shows that the association score S is 0.168, which is used to measure the overall similarity of two feature vectors in three-dimensional space. Combined with the influence of the weight factor, it can be further used for the calculation and result analysis of the semantic matching score matrix.

[0139] S303: Based on the semantic matching score matrix, analyze the distribution of the vector score in each category, calculate the weight score according to the vector category distribution ratio, integrate the category distribution and the weight score into the matrix, and generate a dynamic matching score matrix;

[0140] Analyze the distribution of vector scores in each category, calculate the weight score of the category by counting the frequency and mean of the category distribution, normalize the frequency value of the category into a weight coefficient, integrate the weight coefficient and the score in the distribution matrix for dynamic adjustment, and map the updated score back to the corresponding category to generate a dynamic matching score matrix.

[0141] See also Figure 5 , the specific steps of S4 are:

[0142] S401: Based on the dynamic matching score matrix, the risk value of each target area is analyzed, the regional risk parameters are screened according to the preset risk threshold, the boundary parameters and characteristic center points of the screened area are extracted, and the boundary data and the center point positions are integrated to form a risk screening area set;

[0143] Calculate the difference between the risk index value and the risk threshold of each area, and use the difference to select the areas exceeding the threshold as screening targets. Extract the position coordinates of the boundary points and feature center points of these target areas, integrate the boundary data through geometric methods and mark the coordinates of the center points to generate a set of risk screening areas for subsequent analysis.

[0144] S402: Based on the risk screening area set, the spatial coordinates of the center points of the regional features are called, the distribution density of the center points is analyzed, the distance matrix between the center points is calculated, the feature overlap relationship between adjacent areas is identified, and the regional distribution coverage parameters are generated;

[0145] The distribution density of the center points in the two-dimensional plane is counted, and the distance matrix is ​​calculated using the coordinate value difference between the center points. Based on the result of the distance matrix, the adjacent center point pairs are marked and their connection relationship is analyzed. The feature overlap of the region is identified by the distance range, and the feature overlap amount and regional distribution coverage are calculated, and finally the regional distribution coverage parameters are formed.

[0146] S403: Based on the regional distribution coverage parameters, the coverage range and feature overlap rate of each target area are analyzed, the feature distribution data inside and outside the area are calculated, the feature density, coverage range and center distribution of each area are summarized, and a high-density feature distribution parameter set is generated;

[0147] Based on the regional distribution coverage parameters, according to the formula

[0148]

[0149] Computes the combined eigenvalue of a set of high-density eigendistribution parameters.

[0150] In the formula, D represents the comprehensive characteristic value, V j represents the feature density of the jth region, C j represents the coverage of the jth region, O j represents the overlapping feature quantity of the jth region, R j represents the radius of the jth region, and q represents the total number of regions.

[0151] Feature density V j Represents the number of features in a specific area, and the calculation formula is:

[0152]

[0153] Among them, N jrepresents the total number of features in the region, A j Indicates the area of ​​a region.

[0154] Coverage C j It is obtained by integrating the boundary data of the region, and the calculation formula is:

[0155]

[0156] in, Represents the boundary line of region j.

[0157] Assume the density V of the three regions 1 =5,V 2 =8,V 3 =10, coverage area C 1 =20,C 2 =25,C 3 =30, overlap O 1 =4,O 2 =5,O 3 =6, radius R 1 =3,R 2 =4,R 3 =5, the total number of regions q = 3. The comprehensive eigenvalue D is calculated as follows:

[0158]

[0159] Molecular calculation:

[0160]

[0161] Numerator sum:

[0162] 1.583+1.57+1.533=4.686;

[0163] Comprehensive eigenvalue:

[0164]

[0165] The results show that the comprehensive eigenvalue D quantitatively describes the density, coverage and feature overlap relationship of each region in the high-density feature distribution parameter set, and can be used for subsequent regional feature distribution optimization and risk management analysis.

[0166] See also Figure 6 , the specific steps of S5 are:

[0167] S501: Based on the high-density feature distribution parameter set, analyze the color distribution characteristics of the target area, count the spatial distribution law of the color value, calculate the distribution ratio data of the color characteristics in the differentiated area, and identify the change gradient, analyze the contrast relationship between colors according to the difference data, and generate the color feature distribution analysis result;

[0168] The color values ​​are divided into different regions and statistically analyzed through the quantitative model of color space. The spatial distribution law of color values ​​is analyzed, and the distribution ratio of color features in different regions is calculated. The distribution ratio is converted into a standardized gradient change value. The contrast characteristics between regional colors are further analyzed through the difference relationship between gradient changes between regions. The calculation results are integrated to generate color feature distribution analysis results.

[0169] S502: Based on the color feature distribution analysis result, the geometric morphological features of the target area are analyzed, the boundary length and area ratio of the morphological parameters are calculated, the difference between the morphological features is determined, the geometric features are classified into types, and the geometric morphological classification analysis result is generated;

[0170] The geometric features of the target area are extracted, the ratio of the boundary length to the area of ​​each area is calculated, the boundary morphology of the area is identified using the boundary detection method, and the size of the area is obtained through the area calculation formula. The morphological features are analyzed based on the ratio of length to area, and the geometric morphology is divided into multiple types through the feature matching model, and classified into geometric morphology classification analysis results.

[0171] S503: Based on the geometric morphology classification analysis results, the pattern features of the target area are analyzed, the boundary lines and texture distribution of the pattern features are analyzed, the contrast ratio of the pattern features between the regions is calculated, the multi-level feature data of the region is integrated, and the hierarchical feature difference distribution determination result is generated;

[0172] Based on the results of geometric classification analysis, according to the formula

[0173]

[0174] Compute contrast ratios of differential distributions of stratified features.

[0175] Where P represents the comprehensive contrast ratio of the difference distribution of stratified features, B i represents the boundary length of the ith region, T i represents the area of ​​the ith region, D i represents the boundary difference of the i-th region, L i represents the texture complexity of the i-th region, and m represents the total number of regions.

[0176] Boundary length B i It is obtained by accumulating the length of the boundary segments. The formula is:

[0177]

[0178] Among them, k i represents the number of boundary segments of region i, b j Represents the length of the j-th line segment.

[0179] Area T i It is obtained by summing the pixel areas, and the formula is:

[0180] T i =N i ·A p ;

[0181] Among them, N i represents the total number of pixels in region i, A p Represents the area of ​​a single pixel.

[0182] Assume that the boundary lengths of the three regions are B 1 =12,B 2 =15,B 3 =10, the areas are T 1 =30,T 2 =35,T 3 =25, the boundary differences are D 1 =5,D 2 =8,D 3 =6, the texture complexity is L 1 =3,L 2 =4,L 3 =2, the total number of regions is m=3. The comprehensive contrast ratio P is calculated as follows:

[0183]

[0184] Molecular calculation:

[0185]

[0186] Numerator sum:

[0187] 2.067+2.429+3.4=7.896;

[0188] Comprehensive contrast ratio:

[0189]

[0190] The results show that the comprehensive contrast ratio P reflects the overall characteristics of the differential distribution of stratified features in each region and provides a quantitative reference for the morphological and textural characteristics between regions.

[0191] See also Figure 7 , a single system for efficient review of different types of content security, including:

[0192] The normalization module performs size normalization based on the image input content, analyzes the global color distribution, extracts histogram data, calls edge texture parameters to analyze local edge changes and texture distribution, performs global and local feature vectorization encoding, and generates a set of color and texture basic feature vectors;

[0193] The segmentation modeling module segments the target area according to preset rules based on the color and texture basic feature vector set, performs spatial geometric relationship analysis and modeling, analyzes the boundary connectivity of the segmented area, aggregates and calculates the geometric morphological structure, and generates a spatial geometric morphological feature set;

[0194] The vector matching module performs feature hierarchical analysis and semantic vector matching based on the spatial geometric morphological feature set, calculates the category distribution relationship and dynamic score through the vector matching rules, and generates a dynamic matching score matrix based on the distribution results;

[0195] The risk screening module sets the risk threshold based on the dynamic matching score matrix, screens the target areas with qualified risk value parameters, calls the characteristic distribution center point parameters, performs density aggregation operations, calculates the characteristic distribution coverage and overlap rate of the target area, and generates a high-density characteristic distribution parameter set;

[0196] The hierarchical analysis module is based on a set of high-density feature distribution parameters, hierarchically analyzes the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area, calculates the difference value distribution, and generates a hierarchical feature difference distribution judgment result.

[0197] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An efficient method for reviewing the security of different types of content using a single system, characterized by: The following steps are involved: S1: Based on the image input content, perform size normalization, analyze the global color distribution, extract histogram data, call edge texture parameters to analyze local edge changes and texture distribution, perform global and local feature vectorization encoding, and perform fusion operations to generate a set of color and texture basic feature vectors; S2: Based on the color and texture basic feature vector set, the target area is segmented according to preset rules, the spatial geometric relationship is analyzed and modeled, the boundary connectivity of the segmented area is analyzed, and the geometric structure is aggregated and calculated according to the feature quantification to generate a spatial geometric feature set; S3: Based on the spatial geometric morphological feature set, extract target features, perform feature hierarchical analysis and semantic vector matching, calculate category distribution relationships and dynamic scores through vector matching rules, and generate a dynamic matching score matrix based on the distribution results; S4: Based on the dynamic matching score matrix, a risk threshold is set, target areas with risk value parameters meeting the conditions are screened, characteristic distribution center point parameters are called, density aggregation operations are performed, characteristic distribution coverage and overlap rate of the target area are calculated, target area distribution centers and coverage parameters are summarized, and a high-density characteristic distribution parameter set is generated; S5: Based on the high-density feature distribution parameter set, the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area are analyzed hierarchically, the difference value distribution is calculated, and the target content is classified, and the hierarchical feature difference distribution judgment result is generated in combination with the classification system.

2. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: The color and texture basic feature vector set includes a global color distribution vector, a local edge feature vector and a texture feature vector; the spatial geometric morphology feature set includes segmentation region boundary connectivity features, geometric morphology features and spatial geometric relationship features; the dynamic matching score matrix includes category distribution parameters, dynamic score parameters and vector matching parameters; the high-density feature distribution parameter set includes distribution center point parameters, coverage range parameters and overlap rate parameters; the hierarchical feature difference distribution determination result includes color difference features, geometric morphology difference features, detail pattern difference features and classification feature content.

3. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: Based on the image input content, size normalization is performed, global color distribution is analyzed, and histogram data is extracted. Edge texture parameters are called to analyze local edge changes and texture distribution. Global and local feature vector quantization encoding is performed, and fusion operations are performed. The specific steps to generate a set of color and texture basic feature vectors are as follows: S101: Based on the input image, perform size normalization processing, adjust the image aspect ratio and pixel density, analyze the overall color distribution, count the number of pixels of each color, calculate the color proportion value, and analyze the distribution gradient change to generate a global color distribution matrix; S102: Based on the global color distribution matrix, the spatial position and adjacent color distribution of each pixel region in the image are analyzed, the edge direction change of each region is extracted, the texture features are vectorized and encoded through the position coordinates, the texture pattern of the local region is analyzed in combination with the pixel intensity distribution, and a local edge texture matrix is ​​generated; S103: Based on the global color distribution matrix and the local edge texture matrix, the global color features and the local texture features are vectorized and the two features are combined according to corresponding dimensions to generate a color and texture basic feature vector set.

4. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: Based on the color and texture basic feature vector set, the target area is segmented according to preset rules, the spatial geometric relationship is analyzed and modeled, the boundary connectivity of the segmented area is analyzed, and the geometric structure is aggregated and calculated according to the feature quantification to generate the spatial geometric feature set. The specific steps are as follows: S201: Based on the color and texture basic feature vector set, extract image pixel data according to the preset, assign each pixel to a corresponding area according to its color and texture value, analyze the assignment result, locate the boundary range of each area, establish the center position coordinates of the area, and generate a set of spatial segmentation areas; S202: Based on the set of spatial segmented regions, the boundary lines of the regions and their adjacent relationships are analyzed, a geometric description of the region boundaries is constructed according to the coordinate positions and connection relationships of the boundary vertices, boundary intersections and connection paths are analyzed, the connectivity mode of each boundary and the relationship between regions are calculated, and a spatial boundary connectivity matrix is ​​generated; S203: Based on the spatial boundary connectivity matrix, the geometric morphological features of each region are calculated, the area, shape ratio and boundary complexity of the region are analyzed, the features are combined to form a geometric morphological model of the entire region, the morphology is summarized according to the parameter data of the segmented region, and a set of spatial geometric morphological features is generated.

5. The method for efficiently reviewing the security of different types of content by a single system according to claim 4, characterized in that: The complexity parameter calculation formula is specifically: Among them, C represents the region boundary complexity parameter, n represents the total number of regions involved in the calculation, and B i represents the boundary length of the ith region, A i represents the area of ​​the ith region, D i represents the boundary feature difference value of the i-th region, Represents the mean of the difference values ​​of all regional boundary characteristics.

6. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: Based on the spatial geometric morphological feature set, target features are extracted, feature hierarchical analysis and semantic vector matching are performed, and the category distribution relationship and dynamic score are calculated through vector matching rules. The specific steps of generating a dynamic matching score matrix according to the distribution results are as follows: S301: Based on the spatial geometric morphology feature set, the vector data of each feature is parsed to extract the boundary range and spatial distribution of the feature, and the similarity value is calculated according to the difference between the features. The core feature value is extracted by summarizing the similarity to generate a target feature vector set; S302: Based on the target feature vector set, construct a feature hierarchy structure, analyze the semantic attributes of each layer of feature vectors and the association between the upper and lower layers, calculate the association scores between feature vectors according to the semantic attribute relationship, and integrate the calculation results to form a semantic matching score matrix; S303: Based on the semantic matching score matrix, analyze the distribution of vector scores in each category, calculate weight scores according to the vector category distribution ratio, integrate the category distribution and weight scores into the matrix, and generate a dynamic matching score matrix.

7. The method for efficiently reviewing the security of different types of content using a single system according to claim 6, characterized in that: The calculation formula of the correlation score is specifically: Among them, S represents the correlation score of the feature vector, W represents the dimension of the feature vector, and u k and v k They represent the value of the feature vector in the kth dimension, w k Represents the weight factor of the kth dimension, which is obtained by normalizing the importance of the dimension.

8. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: Based on the dynamic matching score matrix, the risk threshold is set, the target area with the risk value parameter meeting the condition is screened, the characteristic distribution center point parameter is called, the density aggregation operation is performed, the characteristic distribution coverage and overlap rate of the target area are calculated, the distribution center and coverage parameters of the target area are summarized, and the specific steps of generating a high-density characteristic distribution parameter set are as follows: S401: Based on the dynamic matching score matrix, analyze the risk value of each target area, screen the regional risk parameters according to the preset risk threshold, extract the boundary parameters and characteristic center points of the screened area, and integrate the boundary data and the center point position to form a risk screening area set; S402: Based on the risk screening area set, calling the spatial coordinates of the center points of regional features, analyzing the distribution density of the center points, calculating the distance matrix between the center points, identifying the feature overlap relationship between adjacent areas, and generating regional distribution coverage parameters; S403: Based on the regional distribution coverage parameters, analyze the coverage range and feature overlap rate of each target region, calculate the feature distribution data inside and outside the region, summarize the feature density, coverage range and center distribution of each region, and generate a high-density feature distribution parameter set.

9. The method for efficiently reviewing the security of different types of content by a single system according to claim 1, characterized in that: Based on the high-density feature distribution parameter set, the color distribution characteristics, geometric features and detail pattern characteristics of the target area are analyzed hierarchically, the difference value distribution is calculated, and the target content is classified. The specific steps of generating the hierarchical feature difference distribution determination result in combination with the classification system are as follows: S501: Based on the high-density feature distribution parameter set, analyze the color distribution characteristics of the target area, count the spatial distribution law of the color value, calculate the distribution ratio data of the color characteristics in the differentiated area, identify the change gradient, analyze the contrast relationship between colors according to the difference data, and generate a color feature distribution analysis result; S502: Based on the color feature distribution analysis result, the geometric morphological features of the target area are analyzed, the boundary length and area ratio of the morphological parameters are calculated, the difference between the morphological features is determined, the geometric features are classified into types, and a geometric morphological classification analysis result is generated; S503: Based on the geometric morphology classification analysis result, the pattern features of the target area are analyzed, the boundary lines and texture distribution of the pattern features are analyzed, the contrast ratio of the pattern features between regions is calculated, the multi-level feature data of the region is integrated, and the hierarchical feature difference distribution determination result is generated.

10. An efficient review system for the security of different types of content with a single system, characterized by: The efficient method for reviewing the security of different types of content by a single system according to any one of claims 1 to 9, wherein the system comprises: The normalization module performs size normalization based on the image input content, analyzes the global color distribution, extracts histogram data, calls edge texture parameters to analyze local edge changes and texture distribution, performs global and local feature vectorization encoding, and generates a set of color and texture basic feature vectors; The segmentation modeling module segments the target area according to preset rules based on the color and texture basic feature vector set, performs spatial geometric relationship analysis and modeling, analyzes the boundary connectivity of the segmented area, aggregates and calculates the geometric morphological structure, and generates a spatial geometric morphological feature set; The vector matching module performs feature hierarchical analysis and semantic vector matching based on the spatial geometric morphological feature set, calculates the category distribution relationship and dynamic score through the vector matching rule, and generates a dynamic matching score matrix according to the distribution result; The risk screening module sets the risk threshold based on the dynamic matching score matrix, screens the target area with the risk value parameter meeting the conditions, calls the characteristic distribution center point parameter, performs density aggregation operation, calculates the characteristic distribution coverage and overlap rate of the target area, and generates a high-density characteristic distribution parameter set; The hierarchical analysis module hierarchically analyzes the color distribution characteristics, geometric morphology characteristics and detail pattern characteristics of the target area based on the high-density feature distribution parameter set, calculates the difference value distribution, and generates a hierarchical feature difference distribution determination result.

Images