Seal authenticity identification method based on ink permeation connected domain evolution

Abstract

The invention provides a seal authenticity identification method based on ink penetration connected domain evolution, and the method comprises the steps: constructing a discrete grid penetration depth distribution field of a seal image, executing the penetration depth-based connected domain tracking, and tracking the evolution process of a connected region formed by ink particles in the depth direction, extracting a life cycle feature set representing the penetration depth and the communication range; calculating a fourth-order standardized central moment of effective penetration depth distribution, and quantitatively capturing heavy tail distribution characteristics caused by printing ink physical penetration; and outputting an identification conclusion based on a comparison result of the kurtosis coefficient and the authenticity judgment threshold. According to the method, the microscopic penetration fingerprints of deep penetration are excavated, so that a physical seal with a sparse deep pit structure is effectively distinguished from a high-precision printed or electronic forged seal only with surface accumulation or mathematical uniformity characteristics, and accurate quantitative identification of physical authenticity of the seal is realized.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and pattern recognition technology, and in particular relates to a method for identifying the authenticity of seals based on the evolution of ink penetration connected domains. Background Technology

[0002] With the development of automated financial document processing and e-government, high-precision seal authentication has become an important research area for ensuring information security. Existing seal authentication methods mainly rely on appearance feature matching in computer vision, such as using SIFT / SURF operators for geometric shape comparison, using HSV / Lab color space for color difference analysis, or using LBP / GLCM for texture statistical classification. However, these methods rely excessively on the two-dimensional representational features of images, thus being severely inadequate in the face of high-precision forgeries.

[0003] Specifically, these methods ignore the fundamental difference in the physical formation mechanism of genuine and counterfeit seals—namely, the three-dimensional penetration of genuine ink into the paper fibers versus the two-dimensional accumulation of counterfeit ink on the paper surface. On the one hand, high-resolution (600 DPI and above) printing or copying technologies can highly realistically mimic the two-dimensional colors and textures of genuine seals, rendering appearance-based comparison algorithms completely ineffective. On the other hand, the binarization operation often used in preprocessing to simplify calculations in traditional methods forcibly erases crucial grayscale gradation information that reflects the depth of ink penetration, making physical penetration and surface accumulation indistinguishable at the data level.

[0004] Therefore, it is necessary to develop a method for identifying the authenticity of seals based on the evolution of ink penetration connected domains, which can overcome the limitations of two-dimensional visual features and identify the authenticity of seals from the perspective of physical causes. Summary of the Invention

[0005] The purpose of this invention is to provide a method for authenticating seals based on the evolution of ink penetration connectivity, so as to solve the problems of weak anti-high-precision forgery ability and loss of key physical information in traditional seal authentication schemes mentioned in the background art.

[0006] To achieve the above objectives, on the one hand, the present invention provides a method for authenticating seals based on the evolution of ink penetration connected domains, the method specifically comprising:

[0007] The original scanned image of the seal to be identified is obtained. The red channel is enhanced and mapped and numerically truncated to obtain a red enhanced grayscale image. The red enhanced grayscale image is then inverted and defined to obtain the penetration depth distribution field that characterizes the physical penetration depth of the ink.

[0008] Based on the penetration depth distribution field, a full-order filtering sequence of pixels is constructed, and connected component evolution tracking is performed. During the filtering process, the generation and destruction of connected components are tracked, and a life cycle feature set containing the life cycle information of connected components is extracted.

[0009] Calculate the effective penetration depth of each connected component evolution parameter pair in the life cycle feature set, and calculate the statistical decision index characterizing the heavy-tailed distribution characteristics;

[0010] The statistical judgment index is compared with the preset authenticity judgment threshold to obtain the seal authenticity identification result.

[0011] Based on the aforementioned scheme, the execution of connected component evolution tracking, which tracks the generation and destruction of connected components during the filtering process, specifically includes:

[0012] Initialize the root pixel index array and the threshold array for dynamically maintaining the topological properties of connected components; for the current pixel in the total order filtering sequence, retrieve its previous neighbor set in the discrete mesh graph;

[0013] If the previous neighbor set is empty, it is determined that a new connected component is generated. The potential energy value of the current pixel is written into the generation threshold array as the generation threshold. The current pixel is the root pixel of the new connected component.

[0014] If the preceding neighbor set points to two or more different root pixels, it is determined to be a connected component merging event, and the generation threshold retention principle is executed.

[0015] Based on the aforementioned scheme, the principle for retaining the generation threshold is as follows:

[0016] Find the root pixel with the minimum generation threshold and determine that the connected component it represents continues to survive; determine that the connected components represented by the other root pixels with generation thresholds higher than the minimum generation threshold are eliminated, and record the generation threshold of the eliminated connected component and the potential energy value of the current pixel as the merging threshold to construct the connected component evolution parameter pair.

[0017] Based on the aforementioned scheme, the construction of the total order filtering sequence of pixels specifically includes:

[0018] The penetration depth distribution field is modeled as a discrete grid diagram, and all pixels are sorted in ascending order of their potential energy values.

[0019] For pixels with the same potential energy value, a lexicographical perturbation strategy is adopted to perform secondary sorting according to the row priority or column priority of the pixel coordinates, so as to obtain the strictly monotonically increasing total order filtering sequence.

[0020] Based on the aforementioned scheme, the step of obtaining a red-enhanced grayscale image through red channel enhancement mapping and numerical truncation processing, and then reversing its definition, specifically includes:

[0021] Extract the red, green, and blue channel component values ​​of each pixel in the original scanned image;

[0022] The red channel component is the main component. The weighted sum of the green and blue channel components is subtracted, and the calculation result is truncated to limit it to the range of effective pixel values ​​to obtain the red-enhanced grayscale value.

[0023] The red enhanced grayscale value is inverted to convert the grayscale peak region characterizing the physical penetration of the ink into a local minimum region in the penetration depth distribution field, thereby constructing the penetration depth distribution field.

[0024] Based on the aforementioned scheme, after calculating the effective penetration depth of each connected component evolution parameter pair in the lifecycle feature set, the method further includes:

[0025] A noise filtering threshold is preset, and the life cycle feature set is traversed. Pairs of connected component evolution parameters with an effective penetration depth less than the noise filtering threshold are removed, while effective connected component evolution features are retained to construct an effective life cycle feature set.

[0026] The statistical decision index is calculated based on the effective lifecycle feature set.

[0027] Based on the aforementioned scheme, the statistical decision index is the kurtosis coefficient, and its calculation steps include:

[0028] Calculate the mean of all feature lengths in the effective life cycle feature set, and calculate the second central moment, which characterizes the degree of dispersion of the distribution, and the fourth central moment, which characterizes the degree of extremeness of the tail of the distribution, based on the mean.

[0029] Calculate the ratio of the fourth central moment to the square of the second central moment;

[0030] Subtracting the preset normal distribution reference constant from the ratio yields the kurtosis coefficient, which is used to quantify the characteristics of the heavy-tailed distribution.

[0031] Based on the aforementioned scheme, after obtaining the result of the seal authenticity verification, the process further includes:

[0032] Based on the effective lifecycle feature set, the root pixel corresponding to the connected component evolution parameter pair with an effective penetration depth greater than the preset long-tail threshold is located by backtracking.

[0033] The root pixel is highlighted on the original scanned image, and the kurtosis coefficient and the seal authenticity identification results are combined to generate a seal authenticity verification report including the location of ink penetration points.

[0034] The present invention has the following advantages and effects compared with the prior art:

[0035] By constructing a penetration depth distribution field that characterizes the physical penetration depth of ink and performing connected component tracing based on penetration depth, the micro-texture of the stamp image is transformed into a set of life cycle features that quantize the life cycle of connected components, thus avoiding the loss of key grayscale gradient information caused by traditional binarization.

[0036] Furthermore, by using the fourth-order standardized central moment as the core statistical decision indicator, the significant heavy-tailed distribution characteristics caused by the physical penetration of ink are sensitively captured, and an identification mechanism that transcends shallow geometric morphology and directly addresses physical causes is established. This mechanism effectively distinguishes genuine physical stamps with sparse deep penetration characteristics from high-precision printed and electronically counterfeit stamps that only have surface accumulation characteristics of pigments or ink droplets, thus achieving accurate quantitative identification of the physical authenticity of stamps. Attached Figure Description

[0037] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0038] Figure 1 This is a flowchart of a method for identifying the authenticity of a seal based on the evolution of ink penetration connected domains, provided in an embodiment of the present invention. Detailed Implementation

[0039] To more clearly explain the purpose, technical solutions, and advantages of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein. On the contrary, these embodiments are provided so that the present invention will be more comprehensive and complete, and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0040] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.

[0041] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0042] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0043] The present invention will now be described in detail with reference to specific embodiments:

[0044] Example 1

[0045] As attached Figure 1 As shown, Embodiment 1 of this invention provides a method for authenticating seals based on the evolution of connected domains through ink penetration. This method aims to solve the technical problem in existing technologies where seal authentication based on image appearance features (such as color, shape, and texture contrast) is difficult to distinguish between forged seals generated by physical forgery (such as high-precision color laser printing) or digital forgery (such as image editing software synthesis) and physically printed seals. Specifically, from the interdisciplinary perspective of fluid mechanics and algebraic topology, the method utilizes the difference in connectivity (i.e., in the microscopic diffusion path) between the microscopic deep pit structure formed by the physical penetration of ink in a genuine seal and the flat structure accumulated on the surface of a fake seal based on toner / ink droplets. Specifically, a mathematical fingerprint is constructed that is insensitive to changes in illumination but extremely sensitive to connectivity evolution. This method includes: constructing a full-order filtering sequence of pixels based on the penetration depth distribution field and performing connected domain evolution tracking; tracking the generation and merging events of connected components during the filtering process, determining the time of death of connected components based on the generation threshold retention principle, thereby extracting a lifecycle feature set containing the lifecycle information of connected components.

[0046] Step S1: Obtain the original scanned image of the seal whose authenticity needs to be identified. Through red channel enhancement mapping and numerical truncation processing based on spectral characteristics, obtain a red enhanced grayscale image and invert its definition to obtain the penetration depth distribution field that characterizes the physical penetration depth of ink on the physical carrier (i.e., paper) corresponding to the original scanned image.

[0047] Step S2: Based on the penetration depth distribution field, construct the full-order filtering sequence of pixels and perform connected component evolution tracking; during the filtering process, track the generation and merging events of connected components, determine the time of death of connected components according to the generation threshold retention principle, and thus extract a life cycle feature set containing the life cycle information of connected components.

[0048] Step S3: Calculate the effective penetration depth of each connected component evolution parameter pair in the life cycle feature set, perform statistical analysis on the distribution pattern of the effective connected component evolution parameter pairs after removing connected component evolution parameter pairs that are less than the preset noise threshold, and calculate the statistical decision index (such as kurtosis coefficient) that characterizes the heavy-tailed distribution characteristics as a quantitative basis for identifying the authenticity of the seal.

[0049] Step S4: Compare the kurtosis coefficient with the preset authenticity determination threshold to obtain the seal authenticity identification result.

[0050] The steps described above will be explained in more detail below.

[0051] For step S1, the image of the stamp on the physical carrier that needs to be distinguished as genuine or fake is converted into a mathematically computable scalar field function. Since the original image contains complex background interference (such as black signature marks, printed table lines, and paper fiber textures), it must be converted into a pure mathematical scalar field that can characterize the physical penetration depth of the ink, ensuring that the input data for subsequent topology analysis has a high signal-to-noise ratio and physical consistency.

[0052] Specifically, the specific operation process for obtaining the penetration depth distribution field in step S1 is as follows:

[0053] S101. Image Acquisition and Definition: Acquire the original scanned image of the document to be authenticated using a high-resolution optical scanning device. To ensure that the microscopic penetration characteristics at the paper fiber level can be captured, the scanning resolution is preferably set to 600 DPI (dots per inch) or higher.

[0054] The original scanned image is a two-dimensional pixel matrix with dimensions of Any pixel in this matrix It has three color channel components: red channel component, green channel component, and blue channel component, with values ​​ranging from 0 to 10. An unsigned integer.

[0055] S102, Red Channel Enhancement Mapping: Addressing the reflectivity of stamp ink (typically cinnabar or synthetic red pigment) in the visible light spectrum (i.e., high reflectivity in the 600nm-700nm red band and high absorption in the 400nm-550nm blue-green band), and to suppress black text (all RGB channels are low) and background noise, this embodiment does not use the conventional grayscale weighted averaging method (such as the NTSC standard). Instead, it employs channel differential enhancement mapping and numerical truncation processing. By performing linear weighted operations on each color channel component of the original scanned image, a red-enhanced grayscale image is obtained. Any pixel in this image... coordinates The formula for calculating the pixel value is as follows:

[0056]

[0057] in, Represents pixels in a red-enhanced grayscale image. Pixel value at coordinates; These represent the values ​​of the red, green, and blue channel components at the corresponding coordinates in the original scanned image, respectively. The green channel suppression coefficient is preferably set to 0.5 in this embodiment; The blue channel suppression coefficient is preferably set to 0.5 in this embodiment; where... and The sum is preferably 1.0 to ensure complete suppression of white background and neutral gray noise; This is a numerical truncation function used to prevent calculation results from overflowing due to background color. The integer range of is defined as:

[0058]

[0059] S103. Inversion of Penetration Depth Distribution Field: To adapt to the mathematical definition of the expansion of sub-level sets with increasing parameters in subsequent connected component evolution analysis, and to simulate the physical process of water level rise inundating terrain, it is necessary to map high ink concentration to low terrain. Therefore, the red-enhanced grayscale image is inverted to construct a penetration depth distribution field to characterize the physical penetration depth of ink on the physical carrier (i.e., paper) corresponding to the original scanned image. The penetration depth distribution field The defining formula is:

[0060]

[0061] in, Represents the pixels in the penetration depth distribution field coordinates The potential energy at that point is a scalar. This represents the maximum quantization value for an 8-bit grayscale image. (In the penetration depth distribution field) In the middle, the deep penetration points of the physical seal correspond to The local minimum value is found in the blank area of ​​the paper, while the local maximum value is found in the blank area of ​​the paper.

[0062] For step S2, unlike the existing technology that uses fixed threshold binarization to extract the seal shape, resulting in the loss of a large amount of microscopic grayscale gradient information and failing to capture the dynamic process of penetration, this step aims to quantify the connectivity evolution process of the local minimum region in the penetration depth distribution field f. By performing connected domain evolution tracking, the generation and extinction of connected components are tracked, and a set of life cycle features reflecting the microscopic physical penetration structure of the seal is extracted.

[0063] Specifically, the process of obtaining the lifecycle feature set in step S2 is as follows:

[0064] S201. Constructing a discrete mesh graph and a total order filtering sequence:

[0065] (1) Constructing a discrete grid map: the permeation depth distribution field Modeled as a discrete mesh graph , where the node set For the penetration depth distribution field The set of all pixels, where each pixel is a node and the edge set is an edge set. For the penetration depth distribution field The edge connecting any two adjacent pixels (i.e., those with a geometric distance of 1). In this embodiment, a 4-connected neighborhood (i.e., the four directions of up, down, left, and right) is preferably used to define the neighbor relationship.

[0066] (2) Total order arrangement: In order to establish the time axis of the filtering evolution, all pixels are arranged according to their potential values. Sort the pixels from smallest to largest. Specifically, let the total number of pixels be... Create an index sequence and rearrange the pixels using a sorting algorithm (such as quicksort).

[0067] (3) Uniqueness processing: This addresses the situation where multiple pixels in an image may have the same potential energy value. To ensure the determinism of the algorithm results and the uniqueness of the sequence, this embodiment employs a lexicographical perturbation strategy: when potential energy values ​​are the same, a secondary sort is performed according to the row-priority or column-priority order of pixel coordinates. This ultimately yields a strictly monotonically increasing pixel processing sequence. As the total order filtering sequence, it satisfies: .

[0068] Through the above steps, the process of the horizontal plane gradually rising is actually simulated at the physical level: the first pixel in the sequence corresponds to the position where the ink penetrates the deepest in the stamp (local minimum value), and as the sequence progresses, it corresponds to the shallow ink layer, paper fibers and finally the background area.

[0069] S202. Initialization of State Maintenance Structure and Neighborhood Search:

[0070] Before traversing the total order filtering sequence, i.e., before performing connected component evolution tracing, it is necessary to initialize the data structure used to dynamically maintain the evolution attributes of connected components. In this embodiment, a disjoint-set data structure related to arrays is preferably used for maintenance, and their length is the total number of pixels. It is used to store the state and attributes of each pixel during the dynamic evolution process, specifically including:

[0071] (1) Root pixel index array: Used to store the representative element (i.e., root pixel) index of the connected component to which each pixel belongs, so as to support fast lookup of the set to which a pixel belongs. Initially, the root pixel of all pixels is set to undefined or points to itself.

[0072] (2) Generate threshold array: used to record the birth potential value of the connected component with a certain pixel as the root pixel;

[0073] It should be noted that this array is a key storage structure for maintaining the initial information of the lifecycle of connected features. Specifically, when the total order filtering sequence traverses to a local minimum point in the penetration depth distribution field (corresponding to the physical deep penetration core of the ink), a new connected component is determined to be generated, and the potential energy value of this local minimum point is fixed and stored in this array as the birth potential energy value (i.e., the generation threshold). This generation threshold directly quantifies the initial penetration depth of the microtexture, providing necessary benchmark data for subsequent calculation of the effective penetration depth.

[0074] (3) Activation state array: used to mark whether a pixel has been visited.

[0075] Furthermore, following the total order filtering sequence constructed in S201, the current pixel is selected sequentially. As an object to be processed, and then:

[0076] Activation process: Set the current pixel... The activation state flag is set to True, and its root pixel is temporarily pointed to itself, treating it as an independent temporary set;

[0077] Neighborhood search: Check the current pixel. Four-neighbor neighbor pixels in a discrete mesh graph G;

[0078] Precedence filtering: Among neighboring pixels, identify all pixels whose activation state is marked as True, forming a precedence neighbor set. These neighboring pixels represent the areas already revealed at the current potential energy level.

[0079] S203. Perform evolution event determination based on root pixel query:

[0080] To accurately determine connectivity, a lookup operation (preferably with path compression optimization) is performed on each neighbor pixel in the preceding neighbor set through connected component evolution tracing to find the root pixel of their respective connected component.

[0081] It should be noted that step 203 involves iterative processing of the total order permutation sequence pixel by pixel. This process is driven by the monotonically increasing potential energy value of each pixel, while using a disjoint-set data structure to dynamically maintain the pixel connectivity. The aim is to accurately capture the evolutionary events caused by local minima (corresponding to deep ink penetration) and saddle points (corresponding to connected junctions), thereby laying the foundation for subsequent extraction of raw data that can directly quantify the microscopic penetration structure of the ink.

[0082] Specifically, based on the similarities and differences of the root pixels, the following three evolution events are distinguished:

[0083] (1) Connected component generation event: If the previous neighbor set is empty (i.e., none of the four surrounding pixels are activated), it indicates that the current pixel is connected to the next node. The bottom of a pit at a local minimum is identified as a new connected component to be generated, and the current pixel is... potential energy value Write the generated threshold array. At this point, the pixel... This is the root pixel of the new connected component; this value directly quantifies the initial penetration depth of the ink in this microscopic region.

[0084] (2) Connected component expansion event: If the search results of all neighbor pixels in the previous neighbor set all point to the same root pixel, it indicates that the current pixel is connected to the root pixel. If it is merely an extension of an existing connected component, then a merge operation is performed on that pixel. The parent node index points to the unique root pixel, thus merging it into the connected component. This process does not generate new connected features (i.e., it does not generate new connected components);

[0085] (3) Connected component merging event and elimination determination: If the search result of the neighbor pixel points to two or more different root pixels, let it be... This indicates the current pixel. As a saddle point (i.e., a merging point), it connects multiple originally independent connected components. In this case, the generation threshold retention principle is as follows;

[0086] Query: Read the generation threshold corresponding to each root pixel from the generation threshold array. These thresholds physically represent the maximum penetration depth of each local ink region.

[0087] Comparison and Retention: Find the root pixel with the minimum generation threshold (i.e., the earliest generated and smallest value). The connected component it represents is determined to continue to survive; the ink region that is physically formed first (i.e., penetrates the deepest) will be retained after merging;

[0088] Extinction and Recording: Determine the root pixels whose generation threshold is higher than the minimum generation threshold. The represented connected component is destroyed. For each connected component that is determined to be destroyed... Obtain the connected component Generation threshold Set the merging threshold For the current pixel The potential energy value is used to construct and record each destroyed connected component. The connection evolution parameters ,in Characterizing the initial penetration depth of the ink particles, Characterizing the connectivity depth of a component submerged by surrounding ink, the connectivity evolution parameters affect... The life cycle length directly quantifies the independent microscopic contrast of the ink relative to the paper fiber background at that location;

[0089] Structure update: Perform a merge operation, removing all dead root pixels. The root pixel index is updated to point to the surviving root pixel. This is to maintain consistency in subsequent queries.

[0090] Step S204, Output of the lifecycle feature set:

[0091] After traversing all pixels in the total order filtering sequence, collect all connected component evolution parameter pairs obtained in step S203. Composition of life cycle feature set ,in, The total number of connected domain evolution parameter pairs monitored in the penetration depth distribution field f.

[0092] It should be noted that global connected components (usually representing the background) that have not disappeared by the end of the connected domain evolution tracking (when the potential energy value reaches its maximum, such as 255) are excluded in this embodiment because they do not carry the microscopic features of ink penetration.

[0093] For step S3, based on the life cycle feature set and combined with statistical principles, a decision factor that can quantitatively distinguish the difference between physical penetration and surface deposition is extracted.

[0094] It should be noted that genuine seals (i.e., physical stamping) involve fluid penetration, while the seal carrier, paper, is a porous medium made of interwoven plant fibers. Under the action of capillary force, the ink fluid will penetrate to very deep locations along certain specific wide fiber gaps (corresponding to the penetration depth distribution field). The local minimum point in the data, i.e., the very early generation threshold. These deep pits are often surrounded by towering, impermeable fibrous walls (corresponding to higher saddle points, i.e., later merging thresholds). This deep pit and high-walled structure results in its effective penetration depth ( The distribution of toner or ink droplets on the paper is extremely high and sparse, occurring only in a few fiber pores; while counterfeit stamps (printed / copied / digitally forged stamps) involve electrostatic adsorption or inkjet printing. The distribution of toner or ink droplets on the paper surface is limited by the print resolution and halftoning algorithm. They tend to form a uniform, shallow coverage on the paper surface, and their physical penetration structure is characterized by a large number of shallow pits (generation threshold). Larger) and low barriers (merging threshold) Slightly greater than the generation threshold This leads to its effective penetration depth. They are generally small and concentrated in distribution.

[0095] Specifically, the obtained kurtosis coefficient The operation process is as follows:

[0096] S301, Effective penetration depth calculation and noise filtering:

[0097] Calculate the life cycle feature set Each connected component evolution parameter in the pair Effective penetration depth This characterizes the lifetime length of the corresponding connectivity feature, i.e., the corresponding connected component, during the filtration process. It is used to measure the independent depth contrast of the corresponding ink particle relative to its surrounding environment. The calculation formula is: The original permeation depth sequence was obtained. ;

[0098] Furthermore, considering the unavoidable sensor thermal noise and interference caused by the microscopic roughness of the paper surface during image acquisition, a large number of tiny features with extremely short lifecycles will be generated. To eliminate the influence of these non-critical background noises, a noise filtering threshold is preset. (This embodiment is preferred) =3), traverse the original permeation depth sequence, and remove all those that satisfy the condition. The connected component evolution parameter pairs, i.e. short-lived connectivity features, are retained only those that satisfy... The connected component evolution parameters are used as effective connectivity features to construct an effective lifecycle feature set. (in The effective penetration depth of the rearranged effective connectivity feature is k = 1...m.

[0099] S302. Kurtosis calculation based on fourth-order central moments:

[0100] Since the main difference between genuine and counterfeit seals is not in The mean, but in The distribution of the tail shape is analyzed, with genuine seals exhibiting a heavy-tailed distribution and counterfeit seals exhibiting a light-tailed distribution. To sensitively capture the heavy-tailed distribution characteristics in the connected feature distribution of genuine seals, this embodiment preferably uses the fourth-order central moment (i.e., kurtosis) as the quantification criterion. The specific calculation steps are as follows:

[0101] Calculate the effective lifecycle feature set mean With second-order center distance (i.e., variance): , ;

[0102] Calculate the fourth central moments of the effective lifecycle feature set This indicator utilizes the amplification effect of the fourth power, giving it a high weighting ability to capture abnormally large values ​​(i.e., deep penetration points) that are far from the mean. ;

[0103] Based on the second-order central moment and the fourth-order central moment Combined with numerical stability terms Calculate the kurtosis factor : The fourth central moment Weights used to capture extreme outliers (the long tail) in the distribution; denominator The square of the second central moment is used for normalization scaling; a constant of 3 is used for standardization, making the kurtosis of the normal distribution zero. To prevent division-by-zero errors caused by zero variance due to a solid color or perfectly uniform input image, the minimum stability constant... Preferred .

[0104] It should be noted that this kurtosis coefficient The tail thickness of the feature distribution is directly quantified. When When the value is significantly greater than 0, it indicates that the distribution has a very strong heavy-tailed characteristic, that is, there is a sparse but extremely large penetration depth feature, which is highly consistent with the physical penetration law of the deep pit and high wall of a genuine seal; conversely, when When the value is close to or less than 0, it indicates that the distribution tends to be flat or concentrated and lacks long tail extension, which is consistent with the imaging pattern of surface accumulation of fake seals.

[0105] For step S4, obtaining the seal authenticity verification result, the specific operation process is as follows:

[0106] Preset a threshold for determining authenticity The method for determining this threshold is as follows:

[0107] Collect no fewer than 1000 samples of genuine seals and 1000 samples of counterfeit seals confirmed as high-precision color photocopies / printed / electronic seals, and calculate their kurtosis coefficients respectively. The optimal split point that maximizes the distinction between the two classes of sample distributions is found using Support Vector Machine (SVM) or Otsu's method, and is used as the true / false determination threshold. In this embodiment, the preferred method is the one that has undergone extensive testing. .

[0108] The kurtosis coefficient Threshold for determining authenticity Perform a numerical comparison and execute the following branch logic:

[0109] (1) If If the image is found to be genuine, it indicates that the micro-texture in the image contains a large number of high-depth physical penetration structures, which are consistent with the hydrodynamic characteristics of ink physical penetration. The output identification result data is genuine seal, and a confidence score can be attached.

[0110] The confidence score Based on kurtosis coefficient Calculation of the degree to which the value deviates from the threshold:

[0111]

[0112] in For adjustment coefficients, Kurtosis coefficient The larger the value, the more extreme the heavy-tailed features, and the higher the credibility of the physical stamping.

[0113] (2) If If the micro-texture distribution in the image is too regular or flat, lacking evidence of deep penetration and interconnected evolution, it is determined to be a fake seal. This indicates that the micro-texture distribution in the image is too regular or flat, lacking evidence of deep penetration and interconnected evolution, and is consistent with the characteristics of pigment or ink droplet surface accumulation. The output identification result data is a fake seal.

[0114] Furthermore, to visually demonstrate the basis for the judgment, the effective lifecycle feature set obtained in step S301 can also be used. Backtrack to find all the satisfied And located in the long tail region (e.g.) The key connectivity evolution feature points (i.e., the root pixel corresponding to the deepest local penetration position) are identified, and these representative pixels are marked with a bright color (such as red anchor points) on the original scanned image. Combined with the kurtosis coefficient and the seal authenticity identification results, a seal authenticity verification report is generated for users to intuitively verify the location of the ink penetration points.

[0115] To verify the effectiveness of the seal authentication method based on ink penetration connectivity evolution proposed in this invention, a test dataset containing genuine seals (with physical penetration characteristics) and counterfeit seals (with surface deposition characteristics) was constructed for verification. The test environment consisted of an Intel Core i7 processor and 16GB of memory.

[0116] In the test, the aforementioned threshold for determining authenticity was addressed. Parameter sensitivity analysis was performed. Experimental results show that, considering the balance between accuracy, precision, and recall, Under this threshold setting, the comprehensive performance indicators of the method of the present invention are shown in the table below:

[0117] Performance indicators numerical values Accuracy 92.30% Precision 97.00% Recall 87.30% F1 score 0.9189 AUC value 0.9652

[0118] Experimental data show that the mean of the statistical decision index (kurtosis coefficient) of genuine seal samples is significantly greater than 0 (indicating a heavy-tailed distribution), while the mean of the index of fake seal samples is close to 0 (indicating a normal or light-tailed distribution). The difference between the two is significant, which verifies the effectiveness of the present invention in identifying based on the evolution characteristics of connected components.

[0119] To further illustrate the technical advantages of this invention, the method of this invention was compared with existing SIFT feature matching, LBP texture classification, and CNN deep learning methods, including its performance in anti-high-precision forgery scenarios such as color laser printing. The results are shown in the table below, where n represents the total number of pixels in the image:

[0120] method accuracy F1 score High-precision counterfeiting resistance Complexity Method of this embodiment 92.30% 0.9189 96.00% O(nlogn) SIFT feature matching 78.00% 0.7800 45.00% O(nlogn) HSV color space 72.00% 0.7200 35.00% O(n) LBP Texture Classification 81.00% 0.8100 52.00% O(n) Binarized morphology 68.00% 0.6800 28.00% O(n) CNN deep learning 89.00% 0.8900 75.00% High, GPU dependent

[0121] As can be seen from the table above, compared with traditional methods (SIFT) that rely on the appearance features of two-dimensional images and deep learning methods (CNN) that rely on a large amount of training data, this invention has significant advantages in identifying high-precision counterfeit seals by extracting microscopic physical penetration depth features (accuracy improved to 96.00%), and has higher computational efficiency, with a single conventional-sized image processing time of only about 12ms, meeting the requirements of real-time processing.

[0122] In summary, in this embodiment, by constructing a penetration depth distribution field of a discrete grid and performing connected component tracing based on penetration depth, the microscopic texture of the seal image is transformed into a set of life cycle features that quantify the life cycle of connected components. Furthermore, by using the fourth-order normalized central moment (kurtosis) as the core statistical decision factor, the significant heavy-tailed distribution characteristics caused by the physical penetration of ink (i.e., deep pit and high wall structure) are sensitively captured. Essentially, an identification mechanism that transcends shallow geometric morphology and directly addresses physical causes is established. This effectively distinguishes genuine seals with sparse deep penetration characteristics from high-precision printed and electronically counterfeited seals that only have surface accumulation characteristics of pigments / ink droplets / or mathematical uniformity of digital images. This overcomes the technical deficiency of traditional morphological analysis in identifying highly realistic counterfeits and achieves accurate quantitative identification of the physical authenticity of seals.

[0123] Example 2

[0124] Embodiment 2 of the present invention provides a computing device, which can be a personal computer, server, smart terminal, or other computing device. The computing device includes a processing component, a storage component, and a communication component.

[0125] The storage component is used to store one or more computer instructions, and the storage component may include volatile memory (such as random access memory RAM) and / or non-volatile memory (such as read-only memory ROM, flash memory, hard disk, etc.).

[0126] The processing component may be a central processing unit (CPU), a digital signal processor (DSP), or other form of processing unit. The processing component is communicatively connected to the storage component via a bus or other connection method.

[0127] When the one or more computer instructions are invoked and executed by the processing component, the processing component is able to perform all or part of the steps of a seal authenticity identification method based on ink penetration connectivity evolution as described in Embodiment 1.

[0128] Example 3

[0129] Embodiment 3 of the present invention provides a computer-readable storage medium, which may be a non-volatile storage medium, such as an optical disc, a hard disk, a USB flash drive, a flash memory, etc.

[0130] The computer-readable storage medium stores a computer program or instructions that, when executed by the processing component of a computer or processor, cause the computer or processor to perform a method for authenticating seals based on ink penetration connectivity evolution as described in Embodiment 1.

[0131] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims. It should be understood that the invention is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for authenticating seals based on the evolution of ink penetration connected domains, characterized in that, include: The original scanned image of the seal to be identified is obtained. The red channel is enhanced and mapped and numerically truncated to obtain a red enhanced grayscale image. The image is then inverted and defined to obtain the penetration depth distribution field that characterizes the physical penetration depth of the ink. Based on the penetration depth distribution field, a full-order filtering sequence of pixels is constructed, and connected component evolution tracking is performed. During the filtering process, the generation and destruction of connected components are tracked, and a life cycle feature set containing the life cycle information of connected components is extracted. Calculate the effective penetration depth of each connected component evolution parameter pair in the life cycle feature set, and calculate the statistical decision index characterizing the heavy-tailed distribution characteristics; The statistical judgment index is compared with the preset authenticity judgment threshold to obtain the seal authenticity identification result.

2. The method for authenticating seals based on ink penetration connectivity evolution according to claim 1, characterized in that, The execution of connected component evolution tracing, which tracks the generation and destruction of connected components during the filtering process, specifically includes: Initialize the root pixel index array and the threshold array for dynamically maintaining the topological properties of connected components; for the current pixel in the total order filtering sequence, retrieve its previous neighbor set in the discrete mesh graph; If the previous neighbor set is empty, it is determined that a new connected component is generated. The potential energy value of the current pixel is written into the generation threshold array as the generation threshold. The current pixel is the root pixel of the new connected component. If the preceding neighbor set points to two or more different root pixels, it is determined to be a connected component merging event, and the generation threshold retention principle is executed.

3. The method for authenticating seals based on ink penetration connectivity evolution according to claim 2, characterized in that, The principle for retaining the generation threshold is as follows: Find the root pixel with the minimum generation threshold and determine whether the connected component it represents continues to survive. The connected components represented by the root pixels whose generation thresholds are higher than the minimum generation threshold are determined to be eliminated, and the generation thresholds of the eliminated connected components and the potential energy value of the current pixel, which serves as the merging threshold, are recorded to construct the connected component evolution parameter pair.

4. The method for authenticating seals based on ink penetration connectivity evolution according to claim 1, characterized in that, The constructed pixel-level total order filtering sequence specifically includes: The penetration depth distribution field is modeled as a discrete grid diagram, and all pixels are sorted in ascending order of their potential energy values. For pixels with the same potential energy value, a lexicographical perturbation strategy is adopted to perform secondary sorting according to the row priority or column priority of the pixel coordinates, so as to obtain the strictly monotonically increasing total order filtering sequence.

5. The method for authenticating seals based on ink penetration connectivity evolution according to claim 1, characterized in that, The process of enhancing the red channel mapping and numerical truncation to obtain a red-enhanced grayscale image and then reversing its definition specifically includes: Extract the red, green, and blue channel component values ​​of each pixel in the original scanned image; The red channel component is the main component. The weighted sum of the green and blue channel components is subtracted, and the calculation result is truncated to limit it to the effective pixel value range to obtain the red-enhanced grayscale value. The red enhanced grayscale value is inverted to convert the grayscale peak region characterizing the physical penetration of the ink into a local minimum region in the penetration depth distribution field, thereby constructing the penetration depth distribution field.

6. The method for authenticating seals based on ink penetration connectivity evolution according to claim 2, characterized in that, After calculating the effective penetration depth of each connected component evolution parameter pair in the lifecycle feature set, the method further includes: A noise filtering threshold is preset, and the life cycle feature set is traversed. Pairs of connected component evolution parameters with an effective penetration depth less than the noise filtering threshold are removed, while effective connected component evolution features are retained to construct an effective life cycle feature set. The statistical decision index is calculated based on the effective lifecycle feature set.

7. The method for authenticating seals based on ink penetration connectivity evolution according to claim 6, characterized in that, The statistical decision index is the kurtosis coefficient, and its calculation steps include: Calculate the mean of all feature lengths in the effective life cycle feature set, and calculate the second central moment, which characterizes the degree of dispersion of the distribution, and the fourth central moment, which characterizes the degree of extremeness of the tail of the distribution, based on the mean. Calculate the ratio of the fourth central moment to the square of the second central moment; Subtracting the preset normal distribution reference constant from the ratio yields the kurtosis coefficient, which is used to quantify the characteristics of the heavy-tailed distribution.

8. The method for authenticating seals based on ink penetration connectivity evolution according to claim 7, characterized in that, After obtaining the result of the seal authentication, the process also includes: Based on the effective lifecycle feature set, the root pixel corresponding to the connected component evolution parameter pair with an effective penetration depth greater than the preset long-tail threshold is located by backtracking. The root pixel is highlighted on the original scanned image, and the kurtosis coefficient and the seal authenticity identification results are combined to generate a seal authenticity verification report including the location of ink penetration points.

Images