Print printing ink accumulation non-uniform feature-based authenticity discrimination method

Abstract

The invention relates to a authenticity discrimination method based on a seal printing ink accumulation non-uniform characteristic, and the method comprises the steps: collecting a forward imaging image and a surface fluctuation response enhanced image of a to-be-detected seal, and carrying out the registration of the two images, so as to determine an effective region of the seal; screening an ink accumulation dominant area in the effective area according to gray scale fluctuation or height response; positioning micro-defect candidate points with local ink deficiency or gray scale lightening in the dominant area; for each candidate point, carrying out accumulation backfill consistency check in a neighborhood determined according to the resolution or stroke width; the inspection comprises the following steps: whether lateral stacking enhancement formed by extrusion exists on two sides of the defect or not, whether trailing stacking enhancement consistent with the stacking direction exists on the downstream of the defect or not, and defect peripheral stacking enhancement is related to a space of a defect boundary and is distributed in a bypassing manner; and when the number of the candidate points passing the test reaches a preset threshold value, judging that the candidate points are true, otherwise, judging that the candidate points are suspected to be forged or unjudged. The misjudgment probability is reduced, and the authenticity confidence with statistical significance can be output.

Description

Technical Field

[0001] This invention relates to the fields of image processing and computer vision technology, specifically a method for determining the authenticity of printed text based on the characteristics of uneven ink accumulation. Background Technology

[0002] Based on current technologies CN113705330A and CN110728277A, it can be seen that existing methods for identifying the authenticity of seals or based on seal impression features still have significant shortcomings in terms of concept and implementation. The core problem is that most solutions still remain at the level of two-dimensional appearance structure or geometric consistency, making it difficult to truly depict the physical mechanism characteristics of the seal impression formation process. Taking CN113705330A as an example, this solution mainly relies on the comparison of grayscale length or area within a preset sampling area, and on extracting so-called special points by the intersection of the seal outline and the background, and then using the length and angle of the line connecting the special points to judge geometric consistency. Its essence is still a typical image comparison method based on shape, position, and geometric relationships. Under ideal conditions, this type of method has a certain ability to distinguish between crudely imitated or obviously deformed counterfeit seals, but its limitations are particularly prominent when facing current highly realistic counterfeiting technologies. Modern forgery techniques can use high-precision CNC engraving, laser carving, or high-resolution printing to make the overall outline, character shape, local grayscale distribution, and geometric proportions of the imprint highly consistent with the genuine seal. This makes statistical results based on the grayscale length, area, or random point connections of the sampling circle highly susceptible to falling within a preset threshold range, leading to misjudgments. This type of method is highly sensitive to seal alignment, size normalization, and center positioning. Even slight rotational errors, stretching distortions, paper wrinkles, or changes in scanning resolution during the imaging process can have a chain reaction effect on the position of the sampling area and the geometric calculation results, making the judgment results unstable. This scheme completely ignores the microscopic uneven accumulation of ink caused by force, friction, and material rheology during the actual imprinting process, treating the imprint only as a static two-dimensional graphic object. It cannot distinguish between genuine imprints that appear identical but have completely different formation mechanisms and mechanically copied or printed counterfeit imprints. This is the fundamental reason for its insufficient reliability in forensic identification and high-risk application scenarios.

[0003] In contrast, while CN110728277A incorporates advanced computer vision and deep learning concepts such as feature maps, candidate boxes, text region flattening, and OCR recognition, its technical objectives primarily focus on seal detection, localization, and text content recognition, rather than seal authenticity assessment. Its core concern is where the seal is and what is written on it, rather than whether it was formed by a genuine impression. This method relies on feature maps and candidate box classification results, and is significantly affected by the distribution of training data, network generalization ability, and the diversity of seal styles. Furthermore, its output mainly consists of position and text information, lacking direct ability to characterize the differences in the physical formation mechanisms of genuine and counterfeit seals. In authenticity assessment scenarios, even if text recognition is completely correct and position detection is accurate, the possibility of high-quality counterfeit seals or post-synthesized seal text cannot be ruled out. Moreover, this method is typically sensitive to imaging conditions, lighting changes, ink aging, and paper background; once the model is removed from its training environment, the probability of false positives and false negatives increases significantly. Summary of the Invention

[0004] The purpose of this invention is to provide a method for authenticity determination based on the characteristics of uneven ink accumulation in printed text, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0005] The present invention addresses the aforementioned technical problems by employing the following technical solution: a method for determining the authenticity of printed text based on the uneven accumulation characteristics of ink, comprising: acquiring a forward imaging image and a surface undulation response enhanced image of the printed text to be tested, and registering the two images to determine the effective area of ​​the printed text; within the effective area, filtering out the dominant ink accumulation area based on grayscale fluctuations or height response;

[0006] Locate candidate points for micro-defects such as local ink loss or grayscale lightening within the dominant area; for each candidate point, perform a backfill consistency check within the neighborhood determined by resolution or stroke width.

[0007] The inspection includes: whether there is lateral accumulation enhancement formed by extrusion on both sides of the defect, whether there is tailing accumulation enhancement downstream of the defect in the same direction as the accumulation, and whether the accumulation enhancement around the defect is spatially related to the defect boundary and distributed around it; when the number of candidate points that pass the inspection reaches a preset threshold, it is determined to be genuine, otherwise it is determined to be suspected forgery or undeterminable.

[0008] Furthermore, the determination of the effective area of ​​the imprint includes performing binary segmentation on the forward imaging image to obtain the imprint foreground, and calculating the local undulation amplitude in the surface undulation response enhancement image; pixels with undulation amplitudes lower than a preset threshold are removed from the foreground to form an effective area.

[0009] Furthermore, the neighborhood determined by resolution or stroke width is as follows: the skeleton of the stroke is extracted in the forward imaging image and the local stroke width is estimated, and 1 to 3 times the local stroke width is used as the neighborhood radius of the candidate point; and the stacking backfill consistency test is performed on at least two neighborhoods of different scales respectively, and the candidate point is determined to pass the test only when the multi-scale test results are consistent.

[0010] Furthermore, the consistency test employs the following method: lateral stacking enhancement, trailing stacking enhancement, and spatial correlation of the around-the-path distribution are calculated as three feature scores, and then weighted and summed according to preset weights to obtain the consistency score of the candidate points. Consistency score satisfy:

[0011]

[0012] in, This indicates the feature score for lateral stacking enhancement. This indicates the feature score for tail-stacking enhancement. This represents the spatial correlation characteristic score of the bypass distribution; They are respectively with The corresponding preset weight coefficients, and satisfy the following conditions: When consistency score Greater than the preset scoring threshold The candidate point is then determined to have passed the test; and based on the number of candidate points that have passed the test... and its consistency score Statistical output confidence level ,in To determine the number of candidate points that pass the test, the statistic must include at least the number of candidate points that pass the test. mean or variance The confidence level of the authenticity Used to distinguish results as real, suspected forgery, or undeterminable.

[0013] Furthermore, the skeleton extraction involves refining the foreground of the strokes to obtain a single-pixel skeleton, and estimating the local stroke radius by transforming the distance from the skeleton point to the stroke boundary. The local stroke width is twice the local stroke radius. When the local stroke width fluctuates more than a preset proportion in the neighborhood of the candidate point, the median width in that neighborhood is taken as the scale benchmark for the neighborhood radius of the candidate point.

[0014] Furthermore, the at least two different scale neighborhoods include a first neighborhood and a second neighborhood determined by 1 times and 3 times the local stroke width, respectively; the consistency of the multi-scale test results includes: when a candidate point satisfies the correlation between lateral stacking enhancement and around-the-path distribution space in both the first and second neighborhoods, and satisfies tail-like stacking enhancement in at least one neighborhood, the candidate point is determined to pass the test.

[0015] Furthermore, before performing the stacking backfill consistency test, an adaptive directional neighborhood is constructed based on the tangential direction of the skeleton at the candidate point. The adaptive directional neighborhood is an elliptical neighborhood formed by the major axis along the tangential direction of the skeleton and the minor axis along the normal direction, wherein the length of the major axis is 2 to 6 times the width of the local stroke and the length of the minor axis is 0.8 to 2 times the width of the local stroke.

[0016] Furthermore, the three feature scores are calculated in at least two scale neighborhoods of the candidate point and then normalized and fused to obtain a consistency score; wherein, the lateral stacking enhancement score is obtained from the difference in gray-scale gradient or undulation response on both sides of the defect, the tailing stacking enhancement score is obtained from the unidirectional cumulative enhancement along the stroke direction, and the circumferential distribution spatial correlation score is obtained from the circumferential continuity of the enhancement response in the defect boundary neighborhood or the correlation coefficient along the boundary.

[0017] Furthermore, the weights and scoring thresholds are adaptively set based on imaging noise or ink contrast, including estimating the background noise variance and stroke contrast within the effective area, and accordingly performing noise compensation and weight adjustment on the feature scores; the authenticity confidence is calculated from the number of candidate points that pass the test and the mean and variance of the consistency scores.

[0018] Furthermore, the background noise variance is estimated by selecting a flat sub-region with fluctuation response below a preset threshold within the effective area, and the stroke contrast is calculated by the difference between the mean gray values ​​inside and outside the stroke or the local contrast index; and the noise variance and stroke contrast are jointly mapped to a noise compensation coefficient, which is used to uniformly normalize the three feature scores of lateral stacking enhancement, trailing stacking enhancement and spatial correlation of bypass distribution.

[0019] The beneficial effects of this invention are as follows: It simultaneously introduces positive imaging images and surface undulation response-enhanced images, and determines the effective area of ​​the printed text based on the registration of these two types of images. This effectively avoids the problem of being easily affected by paper texture, scanning noise, or print aging when relying solely on grayscale information. By screening the dominant ink accumulation area within the effective area and locating micro-defect candidate points with local ink loss or lighter grayscale, the discrimination focus is placed on the microscopic accumulation structure characteristics formed by force, ink flow, and extrusion during the actual printing process, rather than simple shape or contour similarity. This improves the ability to distinguish between genuine and counterfeit printed text from a mechanistic perspective.

[0020] This study introduces adaptive neighborhood construction based on stroke skeleton and local stroke width, multi-scale consistency testing, and directional adaptive elliptical neighborhood analysis. This allows for the stable characterization of lateral stacking enhancement, trailing stacking enhancement, and spatial correlation features of around-the-edge distribution around micro-defects at different scales and directions, and a consistency score is formed through weighted fusion. Simultaneously, an adaptive weight adjustment and noise compensation mechanism combining imaging noise and ink contrast ensures good robustness of the discrimination results under different imaging conditions and different print qualities. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the logic for determining the authenticity of printed ink based on the uneven accumulation characteristics of the ink.

[0022] Figure 2 This is a diagram showing the relationship between the stacking consistency verification function of the skeleton and the multi-scale directional adaptive neighborhood in this invention.

[0023] Figure 3 This is a functional relationship diagram for the multi-scale ink accumulation consistency analysis of this invention.

[0024] Figure 4 This is an example diagram of forward imaging grayscale statistics and threshold segmentation in Embodiment 1 of the present invention.

[0025] Figure 5 This is an example diagram of the statistical distribution and multi-scale neighborhood of the estimated stroke width using skeleton and distance transformation in Embodiment 1 of the present invention.

[0026] Figure 6 This is a comparison chart of the candidate point consistency score distribution and the pass determination in Embodiment 1 of the present invention.

[0027] Figure 7 This is an example diagram of consistency score calculation for the fusion of three feature scores and weights in Embodiment 2 of the present invention.

[0028] Figure 8 This is an example diagram of the two-scale neighborhood feature score calculation and normalization fusion in Embodiment 2 of the present invention.

[0029] Figure 9 This is an example diagram of the adaptive compensation coefficients and weights driven by imaging noise and ink contrast in Embodiment 2 of the present invention. Detailed Implementation

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] Combined with appendix Figure 1As shown, this invention provides a method for determining the authenticity of printed text based on the uneven ink accumulation characteristics. The method involves imaging the printed text under test, acquiring a forward-facing image and a surface undulation response enhanced image of the same text. The forward-facing image reflects the planar grayscale distribution characteristics of the text, while the surface undulation response enhanced image reflects the height variations or undulation response information caused by ink accumulation within the text area. The two images can be acquired using different imaging modes of the same imaging device, or using different imaging devices in the same spatial coordinate system, to ensure a corresponding spatial relationship between the two images. After obtaining the forward-facing image and the surface undulation response enhanced image, registration processing is performed on the two images. Through feature point matching, geometric transformation, or pixel-level alignment, the two images are made spatially corresponding, thereby eliminating the influence of differences in imaging angle, scale, or displacement. Through registration processing, any pixel in the forward-facing image can find a corresponding pixel position in the surface undulation response enhanced image, providing a basis for subsequent joint analysis.

[0032] After image registration, the effective area of ​​the imprint is determined in the registered image. The effective area is the region where ink imprints actually exist and have obvious imaging or fluctuation response characteristics. It is used to eliminate the interference of background areas and invalid noise areas on the discrimination results. Within the effective area of ​​the imprint, the imprint region is further analyzed based on the gray-level fluctuation changes in the forward imaging image or the height response distribution in the surface fluctuation response enhanced image. By comparing the gray-level change amplitude or fluctuation response intensity in different regions, the region where ink accumulation characteristics are dominant is screened out.

[0033] In forward-facing images, local locations with significantly lower grayscale values ​​than the surrounding area are detected; similarly, in images with enhanced surface undulation response, local locations with significantly reduced height response are detected. Regions that simultaneously satisfy the characteristics of reduced grayscale or reduced undulation response are marked as candidate micro-defects. These candidate micro-defects typically correspond to vacant areas left by insufficient ink coverage or displacement during the printing process. They are small in size and discretely distributed, and their identification in isolation is easily affected by noise or imaging errors. Therefore, further verification is needed in conjunction with surrounding accumulation characteristics.

[0034] For each identified micro-defect candidate point, a local neighborhood for analysis is constructed around it. The scale of the neighborhood is determined based on the imaging resolution or the stroke width at the corresponding location, ensuring that the neighborhood covers the ink accumulation influence area associated with the candidate point. The neighborhood can be a circular or elliptical region centered on the candidate point, and its direction can be appropriately adjusted according to the direction of the printed strokes to better reflect the actual shape of ink flow and accumulation. An accumulation backfill consistency test is performed within the neighborhood. By analyzing whether there is backfilling or accumulation enhancement caused by ink extrusion around the micro-defect candidate point, it is determined whether the defect is consistent with the physical accumulation characteristics caused by uneven ink stress during the actual printing process.

[0035] By comparing the changes in grayscale values ​​or the intensity of surface undulation response in the regions to the left and right or in the normal direction of the defect location, it is determined whether there is a significantly enhanced accumulation feature relative to the defect center. Lateral accumulation enhancement is usually manifested as a high grayscale or high undulation response band distributed laterally along the stroke, reflecting the physical characteristics of ink migrating to both sides under pressure. After completing the lateral accumulation enhancement analysis, a trailing accumulation enhancement analysis is further performed in the downstream region of the defect along the stroke direction. By detecting the continuous enhancement of grayscale or undulation response in the stroke extension direction after the defect, it is determined whether there is a trailing accumulation feature consistent with the ink accumulation direction. Trailing accumulation enhancement is usually manifested as an enhancement response that gradually decreases along the stroke direction but has directional consistency, used to characterize the trailing shape formed by ink flow and stretching during the printing contact and separation process.

[0036] Based on the above analysis, a comprehensive spatial correlation test is performed on the area surrounding the defect, focusing on the spatial relationship between the enhanced accumulation area and the defect boundary to determine whether the enhanced accumulation exhibits a continuous or semi-continuous distribution along the defect boundary. By comparing the spatial continuity and distribution direction of the enhanced response in the vicinity of the defect boundary, it is verified whether the enhanced accumulation conforms to the objective law that real ink is blocked and accumulates around the defect. After completing the tests for lateral enhanced accumulation, trailing enhanced accumulation, and spatial correlation of the distribution around each micro-defect candidate point, the candidate points that pass all or a preset combination of test conditions are statistically analyzed. When the number of candidate points that pass the test reaches a preset threshold, the printed image is determined to have the characteristics of uneven ink accumulation formed by real imprinting, and a true judgment result is output. When the number of candidate points that pass the test does not reach the preset threshold, a judgment result of suspected forgery or undeterminable is output based on the distribution of candidate points and the test results.

[0037] Combined with appendix Figure 2As shown, skeleton extraction involves thinning the foreground of the strokes, shrinking the original stroke region to a single-pixel-width skeleton located at the center of the stroke. This obtains a skeleton structure that represents the stroke's direction and connections without altering its topology. After obtaining the single-pixel skeleton, the local width of the strokes is estimated based on the skeleton structure. By performing a distance transformation on the foreground region of the strokes, the distance from each skeleton point to the nearest stroke boundary is calculated, and this distance is used as the local stroke radius at the corresponding location. Since strokes typically exhibit an approximately symmetrical cross-sectional shape in imaging, twice the local stroke radius is used as the local stroke width at the corresponding location to represent the actual thickness of the stroke at that position.

[0038] When constructing the analysis neighborhood for candidate micro-defect points, to avoid instability in the neighborhood scale due to large variations in stroke width within local areas, a statistical analysis is performed on the local stroke width distribution within the candidate point's neighborhood. When the fluctuation range of the local stroke width within the candidate point's neighborhood exceeds a preset proportion, it indicates that the stroke thickness variation within that area is significant. In this case, instead of directly using the single-point or average width as the neighborhood scale benchmark, the median of the local stroke widths within that neighborhood is taken as the scale benchmark for the candidate point's neighborhood radius.

[0039] To comprehensively validate the ink accumulation characteristics around micro-defect candidate points at different spatial scales, at least two analysis neighborhoods of different scales are constructed for each candidate point. The first neighborhood uses a scale equal to one times the width of the local stroke corresponding to the candidate point, and the second neighborhood uses a scale equal to three times the width of the local stroke. By setting neighborhoods of different scales, the local accumulation characteristics close to the stroke center and the ink flow and accumulation influence areas extending to a larger range can be covered respectively, thus avoiding the omission of true accumulation characteristics due to inappropriate selection of a single scale.

[0040] After constructing the first and second neighborhoods, a consistency test for ink backfilling is performed within each neighborhood to determine whether ink accumulation characteristics consistent with the physical imprinting process exist around the candidate point. During the consistency determination of the multi-scale test results, the candidate point is required to satisfy the conditions for lateral accumulation enhancement and spatial correlation of the surrounding distribution in both the first and second neighborhoods. This ensures that a stable accumulation pattern formed by ink extrusion and surrounding can be observed in different spatial ranges. Furthermore, considering that trailing accumulation enhancement is usually greatly affected by stroke direction and local stress conditions, and is only significant within certain scales, a candidate point is considered to exhibit accumulation characteristics consistent with the actual imprinting process under multi-scale analysis if at least one neighborhood in the first or second neighborhood satisfies the trailing accumulation enhancement condition.

[0041] When the above multi-scale consistency conditions are met simultaneously, the micro-defect candidate point is determined to pass the consistency test and is included in the subsequent statistical process of authenticity judgment; if the multi-scale consistency conditions are not met, the candidate point is considered to lack stable and consistent ink accumulation characteristics and fails the test.

[0042] To better align the analysis neighborhood with the actual shape of printed strokes and the direction of ink flow, an adaptive neighborhood is constructed based on the stroke skeleton direction at the candidate point's location. The adaptive neighborhood uses the skeleton tangential direction at the candidate point as the primary direction. By analyzing the local orientation of the skeleton near the candidate point, the tangential direction consistent with the stroke extension direction and the normal direction perpendicular to it are determined, thus providing a basis for setting the neighborhood's direction. After determining the direction, an elliptical analysis neighborhood is constructed centered on the candidate point. The major axis of the ellipse is arranged along the skeleton tangential direction to cover the main areas where ink flows and accumulates along the stroke direction, while the minor axis is arranged along the skeleton normal direction to cover areas where ink diffuses laterally due to compression. The scale of the elliptical neighborhood is adaptively set based on the local stroke width corresponding to the candidate point. The major axis length is set to two to six times the local stroke width to ensure coverage of effective areas for trailing accumulation and downstream accumulation enhancement, while the minor axis length is set to 0.8 to 2 times the local stroke width to accommodate the detection requirements of lateral accumulation features and avoid introducing excessive irrelevant background information. By employing an adaptive elliptical neighborhood based on the skeleton tangential direction, subsequent stacking and backfill consistency checks can be performed within a spatial range that highly matches the stroke shape and ink stacking direction.

[0043] Combined with appendix Figure 3 As shown, after constructing the directional adaptive neighborhood and obtaining the ink response data within the neighborhood of the candidate point, a consistency check of ink backfilling is performed on each micro-defect candidate point. By quantifying and calculating various ink accumulation characteristics, the physical accumulation law formed by uneven ink stress during the actual printing process is comprehensively evaluated. The consistency check calculates the scores of lateral accumulation enhancement features, tailing accumulation enhancement features, and spatial correlation features of around distribution to characterize the ink accumulation behavior around the candidate point in different directions and spatial structures.

[0044] In the feature fusion stage, the feature scores of lateral stacking enhancement, tail-like stacking enhancement, and spatial correlation of around-the-path distribution are weighted and summed according to preset weights to obtain a score that characterizes the overall stacking consistency of candidate points. Consistency score The following relationship must be satisfied:

[0045]

[0046] in, The score represents the lateral stacking enhancement feature, which reflects the degree of lateral stacking enhancement formed on both sides of the candidate point due to ink extrusion. This represents the score for the trailing accumulation enhancement feature, used to reflect the ink trailing and accumulation features downstream of the candidate point along the stroke direction; The score represents the spatial correlation characteristic of the bypass distribution, which is used to reflect the spatial continuity and bypass distribution characteristics between the accumulation enhancement area around the candidate point and the defect boundary. , , These are preset weighting coefficients corresponding to the scores of the three features mentioned above, used to adjust the contribution ratio of various stacked features in the consistency score, and satisfying the following constraints:

[0047]

[0048] By employing the weighted fusion method described above, a comprehensive evaluation of the consistency of candidate point clustering can be formed while ensuring the independence of various features. This consistency score is then obtained. Then, the consistency score will be... Compared with the pre-set scoring threshold When comparing, when the consistency score Greater than the scoring threshold When the corresponding micro-defect candidate point passes the backfill consistency test, the candidate point is included in the valid discrimination sample; when the consistency score is... Not greater than the scoring threshold If the candidate point fails the test, it will not be included in the subsequent statistical analysis of authenticity.

[0049] After completing the consistency test of all candidate points, statistical analysis is performed on the candidate points that pass the test to output the confidence level for overall authenticity determination. Confidence level of authenticity Based on the number of candidate points that passed the test and corresponding consistency score The statistics are calculated, among which This indicates the number of candidate points that passed the backfill consistency test. The statistic includes at least the consistency score of the candidate points that passed the test. mean or variance This is used to reflect the concentration and stability of the overall stacking consistency of candidate points. It is achieved by comprehensively considering the number of candidate points. Based on the statistical distribution characteristics of its consistency score, a confidence index is generated to characterize the authenticity of the seal under test. And based on the confidence level of authenticity The final judgment result will be categorized as genuine, suspected forgery, or undeterminable.

[0050] For each micro-defect candidate point, lateral buildup enhancement scores, trailing buildup enhancement scores, and spatial correlation scores of around-the-edge distribution are calculated in at least two different scales of the analysis neighborhood. By independently calculating the same type of buildup feature in neighborhoods of different scales, the buildup behavior at the local detail level and the ink flow and distribution characteristics in the extended range can be characterized simultaneously, thereby avoiding the omission or misjudgment of true buildup features due to inappropriate selection of a single scale.

[0051] In calculating the lateral build-up enhancement score, the grayscale gradient or surface undulation response intensity is analyzed in both regions on either side of the candidate point defect location. By comparing the differences in the enhancement responses in both regions on either side of the defect, the lateral build-up enhancement score, reflecting the ink's lateral migration characteristics under pressure, is obtained. The lateral build-up enhancement score effectively characterizes the physical behavior of ink build-up towards both sides of the stroke due to localized force concentration during actual printing. In calculating the tailing build-up enhancement score, a unidirectional cumulative analysis of the grayscale or undulation response is performed in the neighborhood along the stroke direction of the candidate point. By statistically analyzing the cumulative change in the enhancement response along the stroke extension direction, a tailing build-up enhancement score is obtained, characterizing the tailing and stretching of ink along the stroke direction during printing contact and separation. This score reflects the ink flow directionality and the continuous build-up characteristics formed in the downstream region of the stroke.

[0052] In calculating the spatial correlation score of the bypass distribution, a boundary neighborhood is constructed around the defect boundary of the candidate point. Spatial continuity analysis is performed on the enhanced response distribution within this neighborhood. By detecting the circumferential continuity of the enhanced response along the defect boundary or calculating the correlation between the enhanced response and the defect boundary, a spatial correlation score is obtained to characterize the bypass accumulation characteristics of ink after being blocked at the defect. This score can effectively distinguish between the continuous bypass structure formed by actual ink accumulation and the discrete enhanced response generated by random noise or non-physical factors.

[0053] After calculating the scores of the three features at different scales, the feature scores obtained at each scale are normalized, and the normalized results are then fused to form a consistency score for consistency verification. By combining multi-scale calculation with normalization fusion, the impact of dimensional differences between different scales and features on the scoring results can be reduced, making the final consistency score more stable and reliable.

[0054] The adaptive setting is based on a comprehensive estimation of the imaging noise level and ink contrast characteristics within the effective area of ​​the printed image, thus avoiding the risk of misjudgment introduced by using fixed parameters when the noise is high or the contrast is low. During the noise characteristic estimation process, a flat sub-region with a surface undulation response below a preset threshold is selected as the background region within the effective area of ​​the printed image. The background noise variance is estimated by statistically analyzing the fluctuations in pixel response values ​​within this flat sub-region, thus characterizing the noise intensity level under the current imaging conditions. Since the flat sub-region theoretically does not contain obvious ink accumulation structures, its response fluctuations mainly originate from imaging noise, thus accurately reflecting the system noise characteristics.

[0055] In the process of ink contrast estimation, the difference in grayscale mean between the printed stroke area and its adjacent background area is compared, or the difference in grayscale distribution inside and outside the stroke area is calculated using a local contrast index, to obtain the stroke contrast parameter used to characterize the salience of the ink. Stroke contrast reflects the clarity and distinguishability of the ink imprint under the current imaging conditions and is an important factor affecting the stability of the deposition features.

[0056] After obtaining the background noise variance and stroke contrast parameters, the two are jointly mapped to generate noise compensation coefficients. These coefficients are then used to uniformly normalize the three feature scores: lateral stacking enhancement, trailing stacking enhancement, and spatial correlation of the surrounding distribution. By introducing noise compensation coefficients, the amplification effect of feature scores can be suppressed when noise is high or contrast is low, while the discriminative ability of feature scores can be enhanced when noise is low or contrast is high, thus making the feature scores obtained under different imaging conditions comparable.

[0057] After noise compensation and normalization of the feature scores, the corresponding weight coefficients and consistency score thresholds are adjusted based on the stability and reliability of the compensated features, making the weight allocation and threshold settings more consistent with the current imaging quality and ink performance characteristics of the printed text. In the final authenticity determination stage, the authenticity confidence level used for overall determination is calculated based on the number of candidate points that passed the backfill consistency test and the statistical characteristics of the corresponding consistency scores. The authenticity confidence level comprehensively considers the number of candidate points that passed the test, as well as the mean and variance of the consistency scores, and is used to reflect the overall performance of the candidate point backfilling features in terms of both quantity and stability.

[0058] Example 1:

[0059] A genuine official seal imprint was selected as the test object. Both a frontal imaging image and a surface undulation response enhanced image were acquired using the same imaging platform. The resolution of the frontal imaging image was set to 1200 dpi, corresponding to a pixel size of approximately 21 micrometers. The image grayscale range was 0-255, with the overall grayscale distribution of the ink area concentrated between 40-120 and the background area concentrated between 200-240. The surface undulation response enhanced image was output using the same spatial coordinate system, and its response value was a dimensionless enhancement value ranging from 0-1. The response value of areas with significant ink accumulation was typically greater than 0.35.

[0060] An adaptive thresholding method is used, and the segmentation threshold is determined to be 160 based on the overall grayscale histogram of the image. For example... Figure 4 As shown in the figure, the overall grayscale distribution curve of the frontal image and the distribution of foreground and background pixels are presented. Pixels with grayscale values ​​less than 160 are mainly concentrated in the ink area. After segmentation, pixels with grayscale values ​​less than 160 are identified as the foreground of the printed text, and the remaining pixels are identified as the background area. The segmentation results show that the foreground area of ​​the printed text contains approximately... Each pixel can completely cover the outline of the official seal and the internal stroke structure of the characters, while effectively eliminating the paper background area.

[0061] After obtaining the foreground region of the imprint, this foreground region is mapped onto the registered surface undulation response enhancement image, and the local undulation amplitude is calculated for each pixel within the foreground range. The calculation method involves constructing a local undulation amplitude around each pixel. A 5-pixel local window was used to calculate the difference between the maximum and minimum fluctuation response values ​​within the window, and this difference was taken as the local fluctuation amplitude corresponding to that pixel. Statistical analysis revealed that in stroke areas with significant ink accumulation, the local fluctuation amplitude was generally distributed between 0.12 and 0.38, while in areas with thinner ink coverage or blurred edges, the local fluctuation amplitude was mostly distributed between 0.04 and 0.10.

[0062] Based on the above statistical results, the local fluctuation amplitude threshold was set to 0.10. Subsequently, pixels with local fluctuation amplitudes below 0.10 within the foreground region of the printed image were removed from the foreground, retaining only pixels with local fluctuation amplitudes not lower than 0.10 as candidate valid regions. After this processing step, the number of pixels in the foreground region of the printed image was reduced from approximately [original value missing]. One reduced to approximately Of the 100 pixels removed, the majority were concentrated in areas with thin ink at the edges of the printed text, areas with localized false edges, and a small number of imaging noise points.

[0063] The regions retained after fluctuation amplitude screening exhibit a continuous spatial distribution, highly consistent with areas of sufficient ink accumulation in the printed strokes. These regions appear as areas with stable response values ​​and significant variations in the surface fluctuation response enhancement image, accurately reflecting the ink accumulation characteristics caused by uneven force during the printing process. Meanwhile, the eliminated low fluctuation amplitude regions were found in subsequent experiments to contribute little to micro-defect detection and consistency verification, and were prone to introducing noise interference. Comparative experimental results show that, with effective region screening, the number of false positives for micro-defect candidate points is reduced by approximately 32% compared to the unscreened approach, the variance of the consistency score is reduced by approximately 27%, and the overall authenticity judgment results are more stable and reliable.

[0064] After obtaining the stroke skeleton, a distance transformation operation is performed on each skeleton pixel to calculate the minimum distance from that skeleton point to the nearest stroke boundary pixel. This distance is then used as the local stroke radius at that location. Further, the local stroke radius is multiplied by 2 to obtain the local stroke width at the corresponding location. For example... Figure 5 As shown, the statistical results indicate that the width of the thinner strokes in the printed text is mainly concentrated in the range of 6-8 pixels, the width of the medium-thickness strokes is concentrated in the range of 9-13 pixels, while the width of the thicker strokes can reach 14-18 pixels, fully demonstrating the significant differences in stroke width in different areas.

[0065] After locating a candidate micro-defect point, the neighborhood scale used for analysis is adaptively determined based on the local stroke width corresponding to the candidate point's location. Figure 5 Taking a candidate point as an example, this candidate point is located on a stroke of medium thickness, and its local stroke width is calculated to be approximately 10 pixels. The neighborhood radius is defined as 1 to 3 times this local stroke width. Therefore, two analytical neighborhoods with radii of 10 pixels and 30 pixels are constructed respectively for subsequent stacking and backfill consistency verification.

[0066] Within a first neighborhood with a radius of 10 pixels, the local ink accumulation features close to the candidate point are analyzed. Calculations show that the average fluctuation response value on both sides of the candidate point in this neighborhood is approximately 0.42, significantly higher than the 0.18 at the center of the candidate point. Simultaneously, a continuously distributed enhanced response is detected near the defect boundary, indicating stable lateral accumulation enhancement and a circumferential distribution characteristic. Within a second neighborhood with a radius of 30 pixels, the overall flow and accumulation of ink along the stroke direction are further analyzed. The results show a continuous enhanced response band approximately 22 pixels long downstream of the candidate point, with an average fluctuation response value of approximately 0.36, exhibiting a clear trailing accumulation enhancement characteristic.

[0067] By comparing and analyzing the consistency test results of backfilling in two different scale neighborhoods, it can be found that the candidate point satisfies the conditions of lateral stacking enhancement and spatial correlation of bypass distribution in both the first and second neighborhoods, and satisfies the condition of tail-like stacking enhancement in the second neighborhood, thus meeting the requirements for multi-scale consistency judgment. Further statistical analysis of micro-defect candidate points in the entire seal reveals that after adopting an adaptive neighborhood based on stroke width and a multi-scale consistency test scheme, 26 candidate points passed the test. In contrast, using only a single fixed neighborhood radius of 20 pixels, 39 candidate points passed the test, of which 14 were confirmed by manual review to have obvious misjudgments.

[0068] In further implementation, multiple rounds of refinement operations were performed on the foreground region of the printed text, gradually shrinking the strokes, which originally ranged in width from 6 to 18 pixels, into single-pixel skeleton structures located at the geometric center of the strokes. After processing, the resulting skeleton maintained the continuity and topological relationship of the original strokes in space, while effectively eliminating the influence of stroke width variations on the centerline position, providing a stable foundation for subsequent orientation analysis and scale estimation.

[0069] After obtaining the single-pixel skeleton, distance transformation calculations are performed on the foreground region of the printed text. For each skeleton point, the minimum distance to the nearest stroke boundary pixel is calculated, and this distance is used as the local stroke radius at that location. Statistical analysis reveals that the local stroke radius corresponding to a candidate point is 5 pixels, indicating a local stroke width of 10 pixels at that location. Further, an initial neighborhood is constructed centered on this candidate point, and the local stroke widths corresponding to all skeleton points within the neighborhood are statistically analyzed. The results show that the stroke width distribution range within this neighborhood is 8-15 pixels, with the ratio of the maximum to the minimum width being approximately 1.88, exceeding the preset width fluctuation ratio of 1.5, indicating significant variations in stroke thickness within this region.

[0070] To address the above situation, the widths of all local strokes within the neighborhood are sorted, and the median is taken as the scaling benchmark. The median width of the local strokes within this neighborhood is calculated to be 11 pixels. Therefore, 11 pixels is used as the scaling benchmark for the neighborhood radius of this candidate point to reduce the impact of extreme local width variations on neighborhood construction and make subsequent analysis more robust.

[0071] After determining the neighborhood scale benchmark, at least two analysis neighborhoods with different scales are constructed. The first neighborhood has a scale of 1 times the local stroke width (i.e., a radius of 11 pixels), and the second neighborhood has a scale of 3 times the local stroke width (i.e., a radius of 33 pixels). By setting neighborhoods with different scales, the first neighborhood mainly covers the local ink accumulation area close to the candidate point, while the second neighborhood is used to cover ink flow, tailing, and accumulation diffusion features over a larger area.

[0072] By analyzing the positional changes of continuous skeleton points near candidate points, the tangential direction of the stroke at that location is calculated, and an elliptical neighborhood is constructed using this tangential direction as the main direction. The major axis of the elliptical neighborhood is set along the tangential direction of the skeleton, and the minor axis is set along the normal direction of the skeleton. Based on a local stroke width of 11 pixels, the length of the major axis is set to 4 times the local stroke width, i.e., 44 pixels, and the length of the minor axis is set to 1.2 times the local stroke width, i.e., approximately 13 pixels. This forms a directionally adaptive analysis neighborhood that can cover the ink flow characteristics along the stroke direction while also taking into account the lateral accumulation characteristics.

[0073] Consistency checks on backfill accumulation were performed at both the first and second scales within the directional adaptive neighborhood. In the first neighborhood, by analyzing the undulation response distribution in the regions on both sides of the candidate point, the average undulation response value on both sides of the defect was found to be approximately 0.41, while the undulation response value at the center of the candidate point was approximately 0.19, indicating a lateral enhancement amplitude of 0.22. Furthermore, the enhanced region was continuously distributed along the defect boundary, satisfying the conditions for lateral stacking enhancement and spatial correlation of around-the-edge distribution. In the second neighborhood, a unidirectional cumulative analysis of the undulation response was further performed along the tangential direction of the skeleton. The results showed a continuous enhanced response region approximately 24 pixels long downstream of the candidate point, with an average response value of approximately 0.35, significantly higher than the 0.12 of the background region, satisfying the conditions for tail-like stacking enhancement. Simultaneously, the spatial correlation between lateral stacking enhancement and around-the-edge distribution was maintained at this scale.

[0074] Based on the combined test results of the first and second neighborhoods, it can be determined that the candidate point satisfies both lateral stacking enhancement and spatial correlation of bypass distribution in both neighborhoods of different scales, and satisfies the tailing stacking enhancement condition in the second neighborhood, thus meeting the multi-scale consistency judgment rule. Therefore, the candidate point is determined to have passed the stacking backfill consistency test. Figure 6 As shown, the consistency score distributions of all candidate points are compared and displayed, with the fixed neighborhood scheme using a threshold of 0.203 and the multi-scale scheme using a threshold of 0.202. The results show that the multi-scale scheme passes fewer candidate points but has higher stability.

[0075] The above process was applied to all micro-defect candidate points in the entire seal, resulting in 42 candidate points. Of these, 26 passed the test after introducing skeleton extraction, a local stroke width median scale benchmark, an adaptive elliptical neighborhood, and a multi-scale consistency check mechanism. In contrast, the control scheme, which only used a fixed circular neighborhood and did not employ an adaptive neighborhood, had 39 candidate points pass the test, of which 14 were manually verified as misjudged points that did not conform to the actual ink accumulation pattern.

[0076] Example 2:

[0077] Based on Example 1, the consistency test employs three complementary ink accumulation features: lateral accumulation enhancement features, trailing accumulation enhancement features, and spatial correlation features of around-the-edge distribution. These three features are then mapped to three numerical feature scores. For example... Figure 7 As shown, the three types of feature scores are visualized in radial form and, together with their corresponding weight coefficients, form the basis for calculating the consistency score. By weighting and fusing these feature scores, a consistency score is obtained to determine the authenticity of candidate points. .

[0078] Consistency score The calculation formula is defined as follows:

[0079]

[0080] in, This indicates the feature score for lateral stacking enhancement. This indicates the feature score for tail-stacking enhancement. This represents the spatial correlation characteristic score of the bypass distribution. , , These are the weight coefficients for the corresponding features, and they satisfy the weight normalization constraint:

[0081]

[0082] To balance the importance of lateral accumulation characteristics and ink flow characteristics along the stroke direction, and to appropriately consider the auxiliary role of around-the-stroke distribution characteristics in spatial structural stability, the weighting parameters are selected as follows:

[0083]

[0084] The above weights are configured in Figure 7 The scores are displayed in a dashed line format, synchronized with the feature scores, to visually represent the contribution ratio of each feature to the consistency score.

[0085] For a candidate micro-defect point, three types of feature scores are calculated in its corresponding adaptive elliptical neighborhood and multi-scale neighborhood. Taking this candidate point as an example, the fluctuation response value at its center position is... The average fluctuation response values ​​of the regions on both sides of the defect are respectively and The lateral stacking enhancement characteristic score is calculated using the normalized difference between the enhancement amplitudes on both sides of the defect and the central response, and is defined as follows:

[0086]

[0087] After substituting the values, we get:

[0088]

[0089] The calculation result is in Figure 7 The values ​​are given in radial dimensioning to visually represent the degree of lateral packing reinforcement.

[0090] Within a larger-scale neighborhood of the same candidate point, a continuously enhanced response region was detected downstream of the stroke direction, with an average fluctuation response value of The average response value of the background area at the same scale is The tail-like stacking enhancement feature score is defined as the normalized result of the tail enhancement amplitude relative to the background response, i.e.:

[0091]

[0092] After substituting the values, we get:

[0093]

[0094] This trailing stacking enhancement feature is in Figure 7 The term "middle" corresponds to the significant radial component along the stroke direction, reflecting the characteristics of ink flow and accumulation along the force direction during the actual printing process.

[0095] The spatial correlation feature of the bypass distribution is used to describe whether the enhanced response around the defect boundary is continuously distributed along the defect contour. The enhanced response sequence is extracted within the neighborhood of the defect boundary, and its correlation coefficient along the boundary direction is calculated as a feature score. In this embodiment, the calculated boundary correlation coefficient is: Therefore, the definition is:

[0096]

[0097] This feature is in Figure 7 The third radial dimension is used to reflect the spatial consistency of the ink distribution around the defect boundary.

[0098] Substituting the above three feature scores into the consistency score calculation formula, we can obtain the consistency score of this candidate point as follows:

[0099]

[0100] Further calculations yield the following:

[0101]

[0102] Based on the statistical analysis results of the printed samples, the consistency score threshold is set as follows:

[0103]

[0104] like Figure 7As shown, the consistency score of this candidate point is 0.602, which is significantly higher than the scoring threshold of 0.202. Therefore, this candidate point is determined to pass the stacking backfill consistency test.

[0105] After completing the consistency check on individual candidate points, the above procedure is applied to all micro-defect candidate points detected within the valid area of ​​the imprint. A total of [number] micro-defect candidate points were detected. There are 100 micro-defect candidate points, of which the number of candidate points that pass the consistency test is 100.

[0106]

[0107] Further statistical analysis was performed on the consistency scores of the candidate points that passed the test, and their mean and variance were calculated as the basis for assessing the confidence level of authenticity. The set of consistency scores of the candidate points that passed the test was defined as follows: Its mean is defined as:

[0108]

[0109] The calculation result is:

[0110]

[0111] Meanwhile, its variance is defined as:

[0112]

[0113] The calculation result is:

[0114]

[0115] Based on the number of candidate points that passed the test Mean of consistency score and variance The overall output of the confidence level of the authenticity of the seal impression When the number of candidate points that pass the test is large, the mean consistency score is significantly higher than the threshold, and the score variance is small, it indicates that the candidate points have stable and consistent ink accumulation unevenness characteristics, and the print can be judged as genuine.

[0116] like Figure 8 As shown, for each candidate point, lateral stacking enhancement feature scores, tail-like stacking enhancement feature scores, and spatial correlation feature scores of around-the-path distribution are calculated in at least two different scale neighborhoods. The final consistency score used for consistency determination is obtained through normalization and fusion. The figure presents the three sets of feature scores—first-scale neighborhood, second-scale neighborhood, and fusion result—in a bar chart comparison to illustrate the role of multi-scale information in improving the stability of consistency determination.

[0117] A candidate point for a minor defect in an already identified genuine official seal imprint is selected. This candidate point is located on a stroke of medium thickness, with a local stroke width of approximately 10 pixels. Two scale neighborhoods are constructed according to the aforementioned rules, with the radius of the first scale neighborhood set to 10 pixels and the radius of the second scale neighborhood set to 30 pixels. Three types of feature scores are calculated in both scale neighborhoods to avoid interference from local anomalies or noise at a single scale on the judgment results.

[0118] Within the first-scale neighborhood, the mean fluctuation response of the symmetrical regions on both sides of the defect was extracted. The average response value of the left region was 0.41, the average response value of the right region was 0.43, and the response value at the center of the candidate point was 0.19. This confirms that there is obvious lateral stacking enhancement caused by compression on both sides of the defect. The lateral stacking enhancement feature score calculated at this scale is 0.55.

[0119] By extracting the enhanced response sequence along the defect boundary and analyzing its spatial continuity, it can be observed that the enhanced response exhibits a relatively complete circumferential distribution along the defect boundary. Further calculation of the correlation coefficient in the boundary direction yields a spatial correlation characteristic score of 0.60 for the circumferential distribution at this scale.

[0120] By performing unidirectional cumulative statistics on the fluctuation response along the tangential direction of the stroke skeleton, a continuous enhanced response region of about 24 pixels in length was detected downstream of the candidate point, with an average fluctuation response value of 0.35, while the average response value of the background region at the same scale is about 0.12. Thus, the tail-like stacking enhancement feature score at this scale was calculated to be 0.66.

[0121] Within the second-scale neighborhood, statistical results show that the average fluctuation response values ​​on both sides of the candidate point are still significantly higher than those at the center, with a lateral stacking enhancement feature score of 0.52. Meanwhile, the enhancement response near the defect boundary maintains a continuous distribution over a larger range, with a spatial correlation feature score of 0.58.

[0122] After obtaining the three feature scores corresponding to the two scale neighborhoods, the feature results at different scales are normalized. Taking the lateral stacking enhancement feature as an example, the feature scores of 0.55 and 0.52 at the first and second scales are weighted and fused to obtain a comprehensive lateral stacking enhancement score of approximately 0.54. Using the same method, the trailing stacking enhancement feature and the spatial correlation feature of the bypass distribution are normalized and fused, yielding comprehensive feature scores of approximately 0.66 and 0.59, respectively.

[0123] The feature scores obtained after multi-scale normalization and fusion are input into the consistency score calculation module to obtain the final consistency score of the candidate point. Since the candidate point stably exhibits lateral stacking enhancement, tail-like stacking enhancement, and spatial correlation characteristics of around-the-loop distribution in both neighborhoods of different scales, its consistency score is significantly higher than the scoring threshold, and it is determined to pass the stacking backfill consistency test.

[0124] like Figure 9 As shown, after filtering the effective area of ​​the printed text, a flat sub-region with low fluctuation response and uniform spatial distribution is selected within this effective area for background noise estimation. In the surface fluctuation response enhanced image, a set of pixels with fluctuation response values ​​below 0.05 is selected, and a connected sub-region with an area greater than 400 pixels is chosen as the noise estimation region. Statistical calculations of the fluctuation response values ​​within these regions yield a background noise variance of approximately 0.0025.

[0125] Statistical analysis of the grayscale distribution of the stroke region and the background region in the forward-facing image was performed to estimate the contrast of the ink strokes. By randomly selecting pixels inside the strokes and adjacent background pixels, the mean grayscale value was calculated separately. The statistical results showed that the average grayscale value inside the strokes was approximately 85, while the average grayscale value of the background region was approximately 220, resulting in a mean grayscale difference of approximately 135.

[0126] Based on the statistical results of background noise variance and stroke contrast, the noise compensation coefficient was determined to be approximately 0.92. This compensation coefficient was then uniformly applied to the feature scores of three categories: lateral stacking enhancement, trailing stacking enhancement, and spatial correlation of around-the-path distribution, and normalized accordingly. Taking a candidate micro-defect as an example, its feature scores before compensation were 0.54, 0.66, and 0.59, which were adjusted to 0.50, 0.61, and 0.54 after compensation.

[0127] Under conditions of low noise level and high stroke contrast, lateral stacking enhancement features and trailing stacking enhancement features are still maintained as the main discrimination criteria. At the same time, the stability role of the spatial correlation feature of the bypass distribution in the overall score is appropriately increased. Finally, the weight parameters are determined to be 0.38, 0.34 and 0.28.

[0128] Statistical results show that after adopting an adaptive weighting and noise compensation mechanism, the consistency scores of candidate points that passed the test in genuine official seal imprints are more concentrated, with a mean score of approximately 0.57 and a variance of approximately 0.018. Further applying the above process to all micro-defect candidate points in the entire seal imprint, a total of 42 candidate points were detected, of which 26 passed the consistency test. Based on the number of candidate points, the mean consistency score, and the variance, the authenticity confidence level of the seal imprint is comprehensively output, thereby achieving a stable distinction between genuine, suspected forgery, and undetermined results.

[0129] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for determining the authenticity of printed text based on the characteristics of uneven ink accumulation, characterized in that... include: Acquire a forward imaging image and a surface undulation response enhanced image of the print to be tested, and register the two images to determine the effective area of ​​the print; Within the effective area, the dominant ink accumulation region is selected based on grayscale fluctuations or high response. Locate candidate micro-defects with localized ink loss or lighter grayscale within the dominant area; For each candidate point, a stacking backfill consistency check is performed within the neighborhood determined by resolution or stroke width. The inspection includes: whether there is lateral accumulation enhancement formed by extrusion on both sides of the defect, whether there is tailing accumulation enhancement downstream of the defect in the same direction as the accumulation, and whether the accumulation enhancement around the defect is spatially related to the defect boundary and distributed around it; when the number of candidate points that pass the inspection reaches a preset threshold, it is determined to be genuine, otherwise it is determined to be suspected forgery or undeterminable.

2. The method for determining authenticity based on the uneven accumulation characteristics of printed ink as described in claim 1, characterized in that... The determination of the effective area of ​​the imprint includes performing binary segmentation on the forward imaging image to obtain the imprint foreground, and calculating the local undulation amplitude in the surface undulation response enhancement image; pixels with undulation amplitudes lower than a preset threshold are removed from the foreground to form the effective area.

3. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 1, characterized in that... The neighborhood determined by resolution or stroke width is as follows: skeleton extraction of strokes in the forward imaging image and estimation of local stroke width, with 1 to 3 times the local stroke width as the neighborhood radius of the candidate point; and the stacking backfill consistency test is performed on at least two neighborhoods of different scales respectively, and the candidate point is determined to pass the test only when the multi-scale test results are consistent.

4. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 1, characterized in that... The consistency test is conducted by calculating three feature scores for lateral stacking enhancement, trailing stacking enhancement, and spatial correlation of around distribution, and then weighting and summing them according to preset weights to obtain a consistency score for candidate points. When the consistency score is greater than a preset score threshold, the candidate point is determined to have passed the test. Based on the number of candidate points that have passed the test and the statistics of their consistency scores, the authenticity confidence level is output to distinguish the results as real, suspected forgery, or undeterminable.

5. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 3, characterized in that... The skeleton extraction process involves refining the foreground of the strokes to obtain a single-pixel skeleton, and estimating the local stroke radius by transforming the distance from the skeleton point to the stroke boundary. The local stroke width is twice the local stroke radius. When the local stroke width fluctuates more than a preset proportion in the neighborhood of the candidate point, the median width in that neighborhood is taken as the scale benchmark for the neighborhood radius of the candidate point.

6. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 3, characterized in that... The at least two different scale neighborhoods include a first neighborhood and a second neighborhood determined by 1 times and 3 times the local stroke width, respectively; the consistency of the multi-scale test results includes: when a candidate point satisfies the correlation between lateral stacking enhancement and around-the-path distribution space in both the first and second neighborhoods, and satisfies tail-like stacking enhancement in at least one neighborhood, the candidate point is determined to pass the test.

7. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 3, characterized in that... Before performing the backfill consistency test, an adaptive directional neighborhood is constructed based on the tangential direction of the skeleton at the candidate point. The adaptive directional neighborhood is an elliptical neighborhood formed by the major axis along the tangential direction of the skeleton and the minor axis along the normal direction, wherein the length of the major axis is 2 to 6 times the width of the local stroke and the length of the minor axis is 0.8 to 2 times the width of the local stroke.

8. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 4, characterized in that... The three feature scores are calculated in at least two scale neighborhoods of the candidate point and then normalized and fused to obtain a consistency score. The lateral stacking enhancement score is obtained from the difference in gray-scale gradient or undulation response on both sides of the defect, the tailing stacking enhancement score is obtained from the unidirectional cumulative enhancement along the stroke direction, and the spatial correlation score of the around distribution is obtained from the circumferential continuity of the enhancement response in the defect boundary neighborhood or the correlation coefficient along the boundary.

9. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 4, characterized in that... The weights and scoring thresholds are adaptively set based on imaging noise or ink contrast, including estimating the background noise variance and stroke contrast within the effective area, and accordingly performing noise compensation and weight adjustment on the feature scores; the authenticity confidence is calculated from the number of candidate points that pass the test and the mean and variance of the consistency scores.

10. The method for authenticity determination based on the uneven accumulation characteristics of printed ink as described in claim 9, characterized in that... The background noise variance is estimated by selecting a flat sub-region with fluctuation response below a preset threshold within the effective area. The stroke contrast is calculated by the difference between the mean gray values ​​inside and outside the stroke or the local contrast index. The noise variance and stroke contrast are mapped together into a noise compensation coefficient, which is used to uniformly normalize the three feature scores of lateral stacking enhancement, trailing stacking enhancement, and spatial correlation of around distribution.

Images