A full-surface invisible anti-counterfeiting method and system based on a three-dimensional feature array

Abstract

This invention discloses a full-surface stealth anti-counterfeiting method and system based on a three-dimensional feature array, belonging to the field of stealth anti-counterfeiting technology. The invention acquires spectral images of the entire surface of an item using a spectral imaging device, identifies and extracts distributed feature points to form an original feature point set; clusters these feature points based on their spectral similarity to generate a feature map containing spatial data and spectral cluster information; encodes the spatial data and spectral clusters in the feature map into an original fingerprint, binds it to the item's production traceability information, and stores it in a database; during verification, images of a local area of ​​the item are acquired, a local feature map is extracted, and iterative spatial registration aligns the local feature map with the corresponding global feature map in the database. The number of matching points and geometric consistency score are analyzed, and authenticity is determined based on a dynamic threshold, outputting production traceability information. This reduces storage space and improves the efficiency of the verification process.

Description

Technical Field

[0001] This invention relates to the field of stealth anti-counterfeiting technology, specifically a full-surface stealth anti-counterfeiting method and system based on a three-dimensional feature array. Background Technology

[0002] In the field of existing anti-counterfeiting technology, widely used anti-counterfeiting methods mainly include QR codes, radio frequency identification (RFID) tags, holographic anti-counterfeiting labels, and some special ink printing labels. While these technologies can provide some anti-counterfeiting protection in the short term, they are essentially still classified as "reproducible labels." The information carried by QR codes or RFID tags is pre-set and deterministic; once obtained or cracked, a large number of counterfeit labels can be generated through copying, making it impossible to fundamentally eliminate counterfeiting. Although holographic anti-counterfeiting labels have a certain visual anti-counterfeiting effect, their manufacturing process has been gradually improved, and high-quality counterfeit holographic labels are frequently seen in the market.

[0003] Furthermore, the aforementioned anti-counterfeiting labels are typically attached to the surface of items in the form of "tags," lacking a physical bond between themselves and the item. In practical use, these tags are easily peeled off, replaced, or damaged, leading to a break in the connection between the item and its identity information. Even with the use of some special inks for printing, current technologies are mostly limited to single, visible or invisible labels, lacking guarantees for the randomness and uniqueness of the label itself. Once the label is completely copied or transferred, the traceability system cannot accurately determine the authenticity of the item. Therefore, existing anti-counterfeiting measures are significantly insufficient in terms of resistance to copying, resistance to damage, and the strength of their bond with the item itself, making it difficult to meet the security requirements for anti-counterfeiting and traceability of high-value items.

[0004] Therefore, this invention discloses a full-surface invisible anti-counterfeiting method and system based on a three-dimensional feature array to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a full-surface invisible anti-counterfeiting method and system based on a three-dimensional feature array to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a full-surface stealth anti-counterfeiting method based on a three-dimensional feature array, the method comprising the following steps: Step 1: Acquire spectral images of the entire surface of the object using a spectral imaging device, identify and extract the distributed feature points, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set; Step 2: Cluster and group feature points based on their spectral similarity to generate a feature map containing spatial data and spectral cluster information; Step 3: Encode the spatial data and spectral clusters in the feature map into original fingerprints, bind them with the production traceability information of the items, and store them in the database to establish a unique identity file for the items; Step 4: During verification, images of local areas of the item are collected, local feature maps are extracted, and the local feature maps are aligned with the global feature maps of the corresponding items in the database through iterative spatial registration. The number of matching points and geometric consistency scores are analyzed, and the authenticity is determined based on dynamic thresholds, and production traceability information is output.

[0007] Step one includes the following: The surface of an object is scanned using a spectral imaging device at at least two different excitation bands to obtain a sequence of monochromatic images of the object surface at different excitation bands, generating a spectral image data cube. The data cube includes spatial coordinate information and wavelength information, and records the excitation intensity corresponding to each excitation band and the response intensity within the spectral range. For each pixel in a monochrome image, calculate the spectral feature vector based on the response intensity and corresponding excitation intensity at each excitation band; Specifically, the ratio of the response intensity to the excitation intensity of each excitation band is recorded as the response value, and the response values ​​under all excitation bands are combined in a preset order to form the spectral feature vector of the pixel. Pixels whose response value in any dimension of the spectral feature vector exceeds the background noise threshold of the corresponding dimension are marked as candidate feature points; connected component analysis is performed on the binary image marked as candidate feature points to identify all interconnected pixel regions; for each connected region, the number of pixels contained is calculated as the region area, and pixels or connected regions whose region area is less than a preset minimum area threshold are removed. For each connected region that passes the area screening, calculate the centroid coordinates and the average spectral eigenvector. Combine the centroid coordinates and the average spectral vector to form a complete description of a feature point, generating the original feature point set.

[0008] This invention effectively distinguishes real feature points from background noise and reduces ambient light interference by acquiring response data under at least two different excitation bands and calculating normalized spectral feature vectors. Connected component analysis is used to remove isolated noise points, and centroid coordinates are calculated, ensuring that the extracted feature points have stable spatial locations and spectral characteristics, providing high-quality foundational data for subsequent clustering and matching.

[0009] After the extraction of the original feature point set is completed, due to imaging noise, material response fluctuations and spatial sampling redundancy, the original feature point set may contain pseudo feature points or similar feature points. Directly using them for subsequent matching or recognition will introduce errors and increase the computational burden. Therefore, noise removal and redundancy compression are performed on the original feature point set. Step two includes the following: Perform spectral clustering on the original feature point set; The spectral clustering of the original feature point set includes: for each feature point in the original feature point set, calculating the similarity between the average spectral feature vector of the feature point and the average spectral feature vector of the remaining feature points, and grouping all feature points with a similarity greater than a preset similarity threshold into the same spectral cluster; Within each spectral cluster, a distance matrix is ​​constructed using the spatial coordinates of all points within the spectral cluster; the elements of the distance matrix are the Euclidean distances between two feature points within the spectral cluster. To reduce data redundancy and facilitate subsequent feature matching and storage, the distance matrix of each spectral cluster is normalized and quantized. The normalization process scales the elements in the distance matrix according to the maximum distance within the cluster, so that all distance values ​​fall within the range of zero to one, eliminating the influence of item size scaling on geometric features. The quantization process discretizes the normalized distance values ​​according to a preset quantization level value to obtain a distance code. The distance code of each spectral cluster is associated with and stored with the corresponding spectral cluster identifier to form a feature map.

[0010] This invention uses spectral clustering to group physically randomly distributed feature points according to their spectral attributes, transforming the disordered set of points into multiple clusters with independent geometric structures, facilitating subsequent encoding and matching. The intra-cluster distance matrix serves as the basis for geometric encoding, which is invariant to translation and rotation, effectively resisting the impact of changes in object placement on matching. By compressing complex feature point data into compact geometric codes, storage space is significantly reduced, and spatial registration and comparison in subsequent verification processes are accelerated.

[0011] Step three includes the following: This step employs a non-uniform quantization encoding method, converting each quantization level value in the geometric encoding into a fixed-length binary string. Unlike traditional uniform quantization, this non-uniform quantization encoding method divides the quantization interval based on the actual distribution characteristics of the quantization level values ​​in the training sample set. Specifically, by collecting quantized level value data of feature points of similar items and statistically analyzing the probability distribution of quantized level values, a finer quantization interval is used in areas with dense quantization level value distribution, and a coarser quantization interval is used in areas with sparse distribution; this can maximize the distinguishing ability of the encoding under a fixed binary length and improve the fingerprint recognition accuracy. Each quantization level value corresponds to a unique binary string of fixed length, and all quantization level values ​​have the same binary string length. All elements in the geometric encoding are converted into their corresponding binary strings in a preset order, and these binary strings are then concatenated in sequence to form the binary encoding of the spectral cluster. After obtaining the binary code of each spectral cluster, the binary codes of all clusters need to be combined into the original fingerprint of the item according to a unified rule. In order to avoid the problem of different fingerprints being generated for the same item due to different arrangement order of spectral clusters in the feature map, this step uses a hash sorting method based on the cluster center coordinates to sort the spectral clusters.

[0012] Specifically, for each spectral cluster, the average value of the spatial coordinates of all feature points within the spectral cluster is calculated to obtain the cluster center coordinates. A hash operation is performed on the cluster center coordinates to obtain a hash value. All spectral clusters are then arranged in ascending or descending order according to the size of the hash value. The hash sorting ensures that the arrangement order of the spectral clusters of the same item remains consistent under different collection conditions, as long as the content of the feature map remains unchanged, thereby ensuring that the generated fingerprint has uniqueness and repeatability. The binary codes of all sorted spectral clusters are concatenated in order to form a long binary sequence; on this basis, a check code generated by the number of spectral clusters and the total number of feature points is added; the concatenated binary sequence is recorded as the original fingerprint. The check code is generated by calculating the number of clusters and the total number of points, and is used to verify the integrity and correctness of the fingerprint data in the subsequent identification process.

[0013] To prevent the original fingerprints from being tampered with or leaked during storage and transmission, the original fingerprints need to be encrypted; the final digital fingerprint is then generated; the final digital fingerprint and the feature map are stored together in the database, and the final digital fingerprint is also bound to the item number and uploaded to the blockchain for evidence preservation.

[0014] This invention converts the geometric encoding of continuous space into a fixed-length binary string, enabling the physical, non-clonable characteristics of each item to be quantized into a unique, storable digital fingerprint. By encrypting and storing the fingerprint along with traceability information, and combining it with blockchain notarization, data tampering or forgery is prevented. Storing the feature map and digital fingerprint together preserves high-precision verification criteria while providing the identification information needed for rapid comparison. The quantization interval is adaptively divided based on the distribution characteristics of quantization level values ​​in the training samples, allowing for the allocation of more quantization levels in data-dense regions, improving encoding resolution; and compressing the encoding length in sparse regions, achieving a balance between storage efficiency and discriminative power. The non-uniform quantization method relies on the distribution of training data, increasing the difficulty for attackers to reverse engineer the geometric encoding. By hashing and sorting the cluster center coordinates, the fingerprints generated from the same feature map in different processing stages are kept in the same order, avoiding fingerprint differences caused by disordered splicing.

[0015] Step four includes the following: The spectral cluster with the most points in the local feature map is selected as the registration reference. The iterative nearest point algorithm is used to calculate the rotation matrix and translation vector of the feature points in the spectral cluster to the coordinate system of the global feature map in the database, so as to minimize the registration error. The transformation is then applied to all feature points in the local feature map to achieve spatial registration. In the registered local feature map, the number of feature point pairs whose spatial distance is less than a preset distance threshold and whose spectral cluster identifier is consistent with that in the global feature map is counted as the number of matching points. The geometric consistency score is calculated as one of all matching point pairs minus the average of the ratio of the spatial distance of the matching point pairs to the scale factor of the local feature map. The scale factor of the local feature map is the maximum span or average distance of the feature point distribution in the local feature map. The dynamic threshold is calculated by multiplying the ratio of the total number of feature points in the local feature map to the total number of feature points in the global feature map by a preset basic matching rate threshold. If the number of matching points is greater than or equal to the dynamic threshold and the geometric consistency score is greater than or equal to the preset minimum score threshold, the item to be tested is determined to be genuine and the corresponding production traceability information is output; otherwise, it is determined to be counterfeit.

[0016] This invention uses an iterative nearest-point algorithm to accurately register local point clouds with a global map, achieving accurate matching and positioning even when only a local area of ​​an object is scanned. It does not rely on a single label or fixed area; the distributed feature point matrix can still provide sufficient matching points for verification even in cases of local wear or contamination. In addition to the number of matching points, a geometric consistency score is calculated to effectively distinguish between accidental and true matches, reducing the false positive rate. The threshold for the number of matching points is dynamically adjusted based on the local sampling area, ensuring that the verification method maintains stable performance in distinguishing true from false in collection areas of different sizes.

[0017] Another aspect of this application provides a full-surface invisible anti-counterfeiting system based on a three-dimensional feature array. The system is applied to the above-mentioned full-surface invisible anti-counterfeiting method based on a three-dimensional feature array. The system includes a feature point screening module, a cluster analysis module, an identity information encryption module, and an alignment judgment module. The feature point screening module is used to acquire spectral images of the entire surface of an object through a spectral imaging device, identify and extract distributed feature points from them, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set. The clustering analysis module is used to cluster and group feature points based on their spectral similarity, generating a feature map containing spatial data and spectral cluster information. The identity information encryption module is used to encode the spatial data and spectral clusters in the feature map into the original fingerprint, and then bind it with the production traceability information of the item and store it in the database to establish a unique identity file for the item. The alignment judgment module is used for verification by acquiring images of local areas of the item, extracting local feature maps, aligning the local feature maps with the global feature maps of the corresponding items in the database through iterative spatial registration, analyzing the number of matching points and geometric consistency scores, determining authenticity based on dynamic thresholds, and outputting production traceability information.

[0018] The feature point filtering module includes a data feature analysis unit and a data filtering unit; The data feature analysis unit is used to scan the surface of an object using a spectral imaging device under at least two different excitation bands, acquire a spectral image data cube under each excitation band, record the excitation intensity and response intensity within the spectral range corresponding to each excitation band, and calculate a spectral feature vector for each pixel in the monochrome image based on the response intensity and corresponding excitation intensity under each excitation band. The data filtering unit is used to filter candidate feature points based on spectral feature vectors, analyze the connected region features of the candidate feature points, and generate an original feature point set.

[0019] The clustering analysis module includes a feature point clustering unit and a feature point distance analysis unit; The feature point clustering unit is used to perform spectral clustering on the original feature point set; The feature point distance analysis unit is used to construct a distance matrix within each spectral cluster using the spatial coordinates of all points within the spectral cluster; normalize and quantize the distance matrix of each spectral cluster to obtain the geometric code of the spectral cluster; and associate and store the geometric code of each spectral cluster with the corresponding spectral cluster identifier to form a feature map.

[0020] The identity information encryption module includes an original fingerprint analysis unit and a source tracing encryption unit; The original fingerprint analysis unit is used to obtain the geometric code of each spectral cluster in the feature map, and to convert each quantization level value in the geometric code into a fixed-length binary string using a non-uniform quantization encoding method. The binary strings corresponding to all quantization level values ​​are then concatenated sequentially to form the binary code of the spectral cluster. After obtaining the binary code of each spectral cluster, the binary codes of all clusters are combined into the original fingerprint of the item according to a unified rule. The traceability encryption unit is used to encrypt the original fingerprint and the production traceability information of the item to generate the final digital fingerprint; the final digital fingerprint and the feature map are stored together in the database, and the final digital fingerprint is also uploaded to the blockchain for evidence storage after being bound to the item number.

[0021] The alignment judgment module includes a spatial transformation unit and a scoring judgment unit; The spatial transformation unit is used to select the spectral cluster with the most points in the local feature map as the registration reference, and uses the iterative nearest point algorithm to calculate the rotation matrix and translation vector of the feature points in the spectral cluster to the global feature map coordinate system in the database to minimize the registration error. This transformation is then applied to all feature points in the local feature map. The scoring and judgment unit counts the number of feature point pairs in the registered local feature map that are spatially less than a preset distance threshold and have the same spectral cluster identifier as the number of matching points, and calculates the geometric consistency score. The dynamic threshold is calculated by multiplying the ratio of the total number of feature points in the local feature map to the total number of feature points in the global feature map by a preset basic matching rate threshold. If the number of matching points is greater than or equal to the dynamic threshold and the geometric consistency score is greater than or equal to the preset minimum score threshold, the item to be tested is determined to be genuine and the corresponding production traceability information is output; otherwise, it is determined to be counterfeit.

[0022] Compared with existing technologies, the beneficial effects of this invention are as follows: By acquiring response data under at least two different excitation bands and calculating normalized spectral feature vectors, this invention can effectively distinguish between real feature points and background noise, reducing ambient light interference. Connected component analysis is used to remove isolated noise points, and centroid coordinates are calculated, ensuring that the extracted feature points have stable spatial positions and spectral characteristics, providing high-quality foundational data for subsequent clustering and matching. This invention uses spectral clustering to group physically randomly distributed feature points according to their spectral attributes, transforming the disordered point set into multiple clusters with independent geometric structures, facilitating subsequent encoding and matching. Using the intra-cluster distance matrix as the basis for geometric encoding, this encoding is invariant to translation and rotation, effectively resisting the influence of changes in object placement posture on matching. Complex feature point data is compressed into compact geometric codes, significantly reducing storage space and accelerating spatial registration and comparison in subsequent verification processes. This invention converts the geometric code of continuous space into a fixed-length binary string, enabling the physical non-clonable features of each object to be quantified into a unique, storable digital fingerprint. By encrypting and storing fingerprints and traceability information, and combining this with blockchain notarization, data tampering or forgery is prevented. Storing feature maps and digital fingerprints together preserves high-precision verification criteria while providing the necessary identification information for rapid comparison. Adaptively dividing the quantization interval based on the distribution characteristics of quantization level values ​​in the training samples allows for the allocation of more quantization levels in data-dense regions, improving encoding resolution; compressing the encoding length in sparse regions achieves a balance between storage efficiency and discriminability. Non-uniform quantization methods rely on the distribution of training data, increasing the difficulty for attackers to reverse-engineer the geometric encoding. Hash operations on and sorting of cluster center coordinates ensure consistent fingerprint order generated from the same feature map in different processing stages, avoiding fingerprint differences caused by disordered splicing. This invention uses an iterative nearest-point algorithm to accurately register local point clouds with the global map, achieving accurate matching and positioning even when only scanning a local area of ​​an object. Without relying on a single label or fixed area, the distributed feature point array can still provide sufficient matching points for verification even under local wear or contamination. In addition to the number of matching points, a geometric consistency score is calculated to effectively distinguish between accidental and true matches, reducing the false positive rate. The threshold for the number of matching points is dynamically adjusted based on the local sampling area, so that the verification method can maintain stable performance in distinguishing between true and false data for sampling areas of different sizes. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to the present invention. Figure 2This is a schematic diagram of the structure of a full-surface invisible anti-counterfeiting system based on a three-dimensional feature array according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 The present invention provides a technical solution: a full-surface invisible anti-counterfeiting method based on a three-dimensional feature array, the method comprising the following steps: Step 1: Acquire spectral images of the entire surface of the object using a spectral imaging device, identify and extract the distributed feature points, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set; Step one includes the following: The surface of an object is scanned using a spectral imaging device at at least two different excitation bands to obtain a sequence of monochromatic images of the object surface at different excitation bands. A spectral image data cube is generated, which includes spatial coordinate information and wavelength information, and records the excitation intensity corresponding to each excitation band and the response intensity within the spectral range. For each pixel in a monochrome image, calculate the spectral feature vector based on the response intensity and corresponding excitation intensity at each excitation band; Specifically, the ratio of the response intensity to the excitation intensity of each excitation band is recorded as the response value, and the response values ​​under all excitation bands are combined in a preset order to form the spectral feature vector of the pixel. Pixels whose response value in any dimension of the spectral feature vector exceeds the background noise threshold of the corresponding dimension are marked as candidate feature points; connected component analysis is performed on the binary image marked as candidate feature points to identify all interconnected pixel regions; for each connected region, the number of pixels contained is calculated as the region area, and pixels or connected regions whose region area is less than a preset minimum area threshold are removed. For each connected region that passes the area screening, calculate the centroid coordinates and the average spectral eigenvector. Combine the centroid coordinates and the average spectral vector to form a complete description of a feature point, generating the original feature point set.

[0026] After the extraction of the original feature point set is completed, due to imaging noise, material response fluctuations and spatial sampling redundancy, the original feature point set may contain pseudo feature points or similar feature points. Directly using them for subsequent matching or recognition will introduce errors and increase the computational burden. Therefore, noise removal and redundancy compression are performed on the original feature point set. Step 2: Cluster and group feature points based on their spectral similarity to generate a feature map containing spatial data and spectral cluster information; Step two includes the following: Perform spectral clustering on the original feature point set; Spectral clustering of the original feature point set includes: for each feature point in the original feature point set, calculating the similarity between the average spectral feature vector of the feature point and the average spectral feature vector of the remaining feature points, and grouping all feature points with a similarity greater than a preset similarity threshold into the same spectral cluster; Example 1: In this example, the similarity between average spectral feature vectors is calculated using cosine similarity; Within each spectral cluster, a distance matrix is ​​constructed using the spatial coordinates of all points within the cluster; the elements of the distance matrix represent the Euclidean distance between two feature points within the spectral cluster. To reduce data redundancy and facilitate subsequent feature matching and storage, the distance matrix of each spectral cluster is normalized and quantized. Normalization scales the elements in the distance matrix according to the maximum intra-cluster distance, so that all distance values ​​fall within the range of zero to one, eliminating the influence of item size scaling on geometric features. Quantization discretizes the normalized distance values ​​according to a preset quantization level to obtain distance codes. The distance code of each spectral cluster is associated with and stored with the corresponding spectral cluster identifier to form a feature map.

[0027] Example 2: In this example, there are 3 feature points A, B and C in the spectral cluster, and their spatial coordinates are as follows: point A (0, 0), point B (3, 4), point C (6, 0); Calculate the Euclidean distance between any two points: the distance between A and B is 5; the distance between A and C is 6; the distance between C and B is 5. Therefore, the distance matrix is: ; The normalization process assigns a maximum intra-cluster distance of 6. Dividing all elements of the matrix by 6 yields the normalized distance matrix: ; The interval [0, 1] is evenly divided into 4 levels, and the codes corresponding to each level are as follows: [0, 0.25) is code 0; [0.25, 0.50) is code 1; [0.50, 0.75) is code 2; [0.75, 1.00] is code 3; Therefore, the distance encoding is ; Step 3: Encode the spatial data and spectral clusters in the feature map into original fingerprints, bind them with the production traceability information of the items, and store them in the database to establish a unique identity file for the items; Step three includes the following: This step employs a non-uniform quantization encoding method, converting each quantization level value in the geometric encoding into a fixed-length binary string. Unlike traditional uniform quantization, the non-uniform quantization encoding method divides the quantization interval based on the actual distribution characteristics of the quantization level values ​​in the training sample set. Specifically, by collecting quantized level value data of feature points of similar items and statistically analyzing the probability distribution of quantized level values, a finer quantization interval is used in areas with dense quantization level value distribution, and a coarser quantization interval is used in areas with sparse distribution; this can maximize the distinguishing ability of the encoding under a fixed binary length and improve the fingerprint recognition accuracy. Example 3: In this example, the quantization level values ​​of all feature point pairs in the training sample set are statistically analyzed, and the probability distribution is as follows: 10 ~ 15: accounting for 35% of the total sample; 15 ~ 20: accounting for 45% of the total sample; 20 ~ 30: accounting for 10% of the total sample; 30 ~ 80: accounting for 10% of the total sample. The preset total number of quantization levels is 8, meaning each distance value is converted into a 3-bit binary code, with an encoding range of 0 to 7. Based on the distribution density of the distance values, a non-uniform quantization interval is used: 10.0 ~ 11.25 is coded as 000; 11.25 ~ 12.5 as 001; 12.5 ~ 13.75 as 010; 13.75 ~ 15.0 as 011; 15.0 ~ 17.5 as 100; 17.5 ~ 20.0 as 101; 20.0 ~ 30.0 as 110; and 30.0 ~ 80.0 as 111. Each quantization level value corresponds to a unique binary string of fixed length, and all quantization level values ​​have the same binary string length. All elements in the geometric encoding are converted into their corresponding binary strings in a preset order, and these binary strings are then concatenated in sequence to form the binary encoding of the spectral cluster. After obtaining the binary code of each spectral cluster, the binary codes of all clusters need to be combined into the original fingerprint of the item according to a unified rule. In order to avoid the problem of different fingerprints being generated for the same item due to different arrangement order of spectral clusters in the feature map, this step uses a hash sorting method based on the cluster center coordinates to sort the spectral clusters.

[0028] Specifically, for each spectral cluster, the average value of the spatial coordinates of all feature points within the spectral cluster is calculated to obtain the cluster center coordinates. A hash operation is performed on the cluster center coordinates to obtain a hash value. All spectral clusters are then arranged in ascending or descending order according to the size of the hash value. The hash sorting ensures that the arrangement order of the spectral clusters of the same item remains consistent under different collection conditions, as long as the content of the feature map remains unchanged, thereby ensuring that the generated fingerprint has uniqueness and repeatability. The binary codes of all sorted spectral clusters are concatenated in order to form a long binary sequence; on this basis, a check code generated by the number of spectral clusters and the total number of feature points is added; the concatenated binary sequence is recorded as the original fingerprint. The check code is generated by calculating the number of clusters and the total number of points, and is used to verify the integrity and correctness of the fingerprint data in the subsequent identification process.

[0029] Example 4: In this example, the check code is generated by performing a cyclic redundancy check or an XOR operation on the number of clusters and the total number of points.

[0030] To prevent the original fingerprints from being tampered with or leaked during storage and transmission, the original fingerprints need to be encrypted; the final digital fingerprint is then generated; the final digital fingerprint and the feature map are stored together in the database, and the final digital fingerprint is also bound to the item number and uploaded to the blockchain for evidence preservation.

[0031] Example 5: In this example, the XOR encryption method is used to perform an XOR operation on the original fingerprint and the production traceability information of the item to generate the final digital fingerprint; the production traceability information includes, but is not limited to, the production batch, production date, manufacturer, logistics records, quality inspection report identity association data of the item. This information is first converted into a binary sequence of the same length as the original fingerprint, and then XORed bit by bit with the original fingerprint. Step 4: During verification, images of local areas of the item are collected, local feature maps are extracted, and the local feature maps are aligned with the global feature maps of the corresponding items in the database through iterative spatial registration. The number of matching points and geometric consistency scores are analyzed, and the authenticity is determined based on dynamic thresholds, and production traceability information is output.

[0032] Step four includes the following: The spectral cluster with the most points in the local feature map is selected as the registration reference. The iterative nearest point algorithm is used to calculate the rotation matrix and translation vector that transforms the feature points in the spectral cluster to the coordinate system of the global feature map in the database, so as to minimize the registration error. The transformation is then applied to all feature points in the local feature map to achieve spatial registration. In the registered local feature map, the number of feature point pairs whose spatial distance to the global feature map is less than a preset distance threshold and whose spectral cluster identifier is consistent with the global feature map is counted as the number of matching points. The geometric consistency score is calculated as one of all matching point pairs minus the average ratio of the spatial distance of the matching point pairs to the scale factor of the local feature map. The scale factor of the local feature map is the maximum span or average distance of the feature point distribution in the local feature map. The dynamic threshold is calculated by multiplying the ratio of the total number of feature points in the local feature map to the total number of feature points in the global feature map by a preset basic matching rate threshold. If the number of matching points is greater than or equal to the dynamic threshold and the geometric consistency score is greater than or equal to the preset minimum score threshold, the item to be tested is determined to be genuine and the corresponding production traceability information is output; otherwise, it is determined to be counterfeit.

[0033] Please see Figure 2 The present invention provides a technical solution: a full-surface invisible anti-counterfeiting system based on a three-dimensional feature array, the system including a feature point screening module, a cluster analysis module, an identity information encryption module and an alignment judgment module; The feature point filtering module is used to acquire spectral images of the entire surface of an object through a spectral imaging device, identify and extract distributed feature points from them, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set. The clustering analysis module is used to cluster and group feature points based on their spectral similarity, generating a feature map that includes spatial data and spectral cluster information. The identity information encryption module is used to encode the spatial data and spectral clusters in the feature map into the original fingerprint, and then bind it with the production traceability information of the item and store it in the database to establish a unique identity file for the item. When the alignment judgment module is used for verification, it acquires images of local areas of the item, extracts local feature maps, aligns the local feature maps with the global feature maps of the corresponding items in the database through iterative spatial registration, analyzes the number of matching points and geometric consistency scores, determines authenticity based on dynamic thresholds, and outputs production traceability information.

[0034] The feature point filtering module includes a data feature analysis unit and a data filtering unit; The data feature analysis unit is used to scan the surface of an object using a spectral imaging device under at least two different excitation bands, acquire a spectral image data cube under each excitation band, record the excitation intensity and response intensity within the spectral range corresponding to each excitation band, and calculate the spectral feature vector for each pixel in the monochrome image based on the response intensity and the corresponding excitation intensity under each excitation band. The data filtering unit is used to filter candidate feature points based on spectral feature vectors, analyze the connected region features of the candidate feature points, and generate the original feature point set.

[0035] The clustering analysis module includes a feature point clustering unit and a feature point distance analysis unit; Feature point clustering units are used to perform spectral clustering on the original feature point set; The feature point distance analysis unit is used to construct a distance matrix within each spectral cluster using the spatial coordinates of all points within the spectral cluster. The distance matrix of each spectral cluster is then normalized and quantized to obtain the geometric code of the spectral cluster. The geometric code of each spectral cluster is associated with and stored with the corresponding spectral cluster identifier to form a feature map.

[0036] The identity information encryption module includes an original fingerprint analysis unit and a source tracing encryption unit; The original fingerprint analysis unit is used to obtain the geometric code of each spectral cluster in the feature map, and to convert each quantization level value in the geometric code into a fixed-length binary string using a non-uniform quantization encoding method. The binary strings corresponding to all quantization level values ​​are then concatenated sequentially to form the binary code of the spectral cluster. After obtaining the binary code of each spectral cluster, the binary codes of all clusters are combined into the original fingerprint of the item according to a unified rule. The traceability encryption unit is used to encrypt the original fingerprint and the production traceability information of the item to generate the final digital fingerprint; the final digital fingerprint and the feature map are stored together in the database, and the final digital fingerprint is also uploaded to the blockchain for evidence storage after being bound to the item number.

[0037] The alignment judgment module includes a spatial transformation unit and a scoring judgment unit; The spatial transformation unit is used to select the spectral cluster with the most points in the local feature map as the registration reference. The iterative nearest point algorithm is used to calculate the rotation matrix and translation vector that transforms the feature points in the spectral cluster to the coordinate system of the global feature map in the database, so as to minimize the registration error. This transformation is then applied to all feature points in the local feature map. In the registered local feature map, the scoring and judgment unit counts the number of feature point pairs that are spatially less than a preset distance threshold and have the same spectral cluster identifier as the number of matching points in the global feature map, and calculates the geometric consistency score. The dynamic threshold is calculated by multiplying the ratio of the total number of feature points in the local feature map to the total number of feature points in the global feature map by a preset basic matching rate threshold. If the number of matching points is greater than or equal to the dynamic threshold and the geometric consistency score is greater than or equal to the preset minimum score threshold, the item to be tested is determined to be genuine and the corresponding production traceability information is output; otherwise, it is determined to be counterfeit.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A full-surface invisible anti-counterfeiting method based on a three-dimensional feature array, characterized in that, The method includes the following steps: Step 1: Acquire spectral images of the entire surface of the object using a spectral imaging device, identify and extract the distributed feature points, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set; Step 2: Cluster and group feature points based on their spectral similarity to generate a feature map containing spatial data and spectral cluster information; Step 3: Encode the spatial data and spectral clusters in the feature map into original fingerprints, bind them with the production traceability information of the items, and store them in the database to establish a unique identity file for the items; Step 4: During verification, images of local areas of the item are collected, local feature maps are extracted, and the local feature maps are aligned with the global feature maps of the corresponding items in the database through iterative spatial registration. The number of matching points and geometric consistency scores are analyzed, and the authenticity is determined based on dynamic thresholds, and production traceability information is output.

2. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 1, characterized in that: Step one includes the following: Using a spectral imaging device, the surface of the object is scanned at at least two different excitation bands to obtain a spectral image data cube for each excitation band, and the excitation intensity and response intensity within the spectral range corresponding to each excitation band are recorded. For each pixel in a monochrome image, a spectral feature vector is calculated based on the response intensity and corresponding excitation intensity in each excitation band. Candidate feature points are selected based on the spectral feature vectors, and the connected region features of the candidate feature points are analyzed to generate an original feature point set.

3. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 2, characterized in that: Step two includes the following: Perform spectral clustering on the original feature point set; Within each spectral cluster, a distance matrix is ​​constructed using the spatial coordinates of all points within the spectral cluster; the elements of the distance matrix are the Euclidean distances between two feature points within the spectral cluster. The distance matrix of each spectral cluster is normalized and quantized to obtain the geometric code of the spectral cluster. The geometric code of each spectral cluster is then associated with and stored with the corresponding spectral cluster identifier to form a feature map.

4. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 3, characterized in that: The spectral clustering of the original feature point set includes: for each feature point in the original feature point set, calculating the similarity between the average spectral feature vector of the feature point and the average spectral feature vector of the remaining feature points, and grouping all feature points with a similarity greater than a preset similarity threshold into the same spectral cluster.

5. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 3, characterized in that: Step three includes the following: For each spectral cluster in the feature map, the geometric code of the spectral cluster is obtained. The non-uniform quantization coding method is used to convert each quantization level value in the geometric code into a fixed-length binary string. The binary strings corresponding to all quantization level values ​​are then concatenated in sequence to form the binary code of the spectral cluster. After obtaining the binary code of each spectral cluster, the binary codes of all clusters are combined into the original fingerprint of the item according to a unified rule; The original fingerprint and the product's production traceability information are encrypted to generate the final digital fingerprint. The final digital fingerprint and the feature map are stored together in the database. At the same time, the final digital fingerprint is bound to the product number and uploaded to the blockchain for evidence preservation.

6. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 5, characterized in that: The quantization interval of non-uniform quantization coding is determined based on the distribution characteristics of quantization level values ​​in the training samples.

7. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 6, characterized in that: The original fingerprint of the finished product includes calculating the cluster center coordinates of each spectral cluster, performing a hash operation on the cluster center coordinates to obtain a hash value, sorting all spectral clusters, concatenating the binary codes of the sorted spectral clusters in sequence, and adding a check code generated by the number of spectral clusters and the total number of feature points. The concatenated binary sequence is recorded as the original fingerprint.

8. The full-surface stealth anti-counterfeiting method based on a three-dimensional feature array according to claim 5, characterized in that: Step four includes the following: The spectral cluster with the most points in the local feature map is selected as the registration reference. The iterative nearest point algorithm is used to calculate the rotation matrix and translation vector that transforms the feature points in the spectral cluster to the coordinate system of the global feature map in the database, so as to minimize the registration error. This transformation is then applied to all feature points in the local feature map. In the registered local feature map, the number of feature point pairs that are spatially less than a preset distance threshold and have the same spectral cluster identifier as the global feature map is counted as the number of matching points, and the geometric consistency score is calculated. The dynamic threshold is calculated by multiplying the ratio of the total number of feature points in the local feature map to the total number of feature points in the global feature map by a preset basic matching rate threshold. If the number of matching points is greater than or equal to the dynamic threshold and the geometric consistency score is greater than or equal to the preset minimum score threshold, the item to be tested is determined to be genuine and the corresponding production traceability information is output; otherwise, it is determined to be counterfeit.

9. A full-surface stealth anti-counterfeiting system based on a three-dimensional feature array, wherein the system is applied to the full-surface stealth anti-counterfeiting method based on a three-dimensional feature array as described in any one of claims 1-8, characterized in that, The system includes a feature point filtering module, a cluster analysis module, an identity information encryption module, and an alignment judgment module; The feature point screening module is used to acquire spectral images of the entire surface of an object through a spectral imaging device, identify and extract distributed feature points from them, record the spatial coordinates and spectral response data of each feature point, and form an original feature point set. The clustering analysis module is used to cluster and group feature points based on their spectral similarity, generating a feature map containing spatial data and spectral cluster information. The identity information encryption module is used to encode the spatial data and spectral clusters in the feature map into the original fingerprint, and then bind it with the production traceability information of the item and store it in the database to establish a unique identity file for the item. The alignment judgment module is used for verification by acquiring images of local areas of the item, extracting local feature maps, aligning the local feature maps with the global feature maps of the corresponding items in the database through iterative spatial registration, analyzing the number of matching points and geometric consistency scores, determining authenticity based on dynamic thresholds, and outputting production traceability information.

Images