Method for identifying distribution state of micro three-dimensional structure on surface of product based on mobile phone to verify authenticity of product

Abstract

The invention relates to the technical field of image identification, in particular to a method for identifying the distribution state of a micro three-dimensional structure on the surface of a product based on a mobile phone to verify the authenticity of the product. The method comprises the following steps: capturing an original image of a micro-stereo structure distribution state of a product surface by using a mobile phone camera, and preprocessing the original image; and performing feature extraction on the preprocessed original image through an image algorithm, calculating the similarity, introducing an anomaly detection algorithm to identify potential anomaly features, and further determining the anomaly features by using a local anomaly factor algorithm, thereby determining a re-checking result and verifying whether a product is a certified product or not. According to the method, the time for calculating the regional mean value is shortened by using the integrogram, the overall efficiency of the algorithm is improved, it is ensured that the algorithm can complete a large amount of calculation in a short time, the real-time requirement is met, abnormal points can be recognized more accurately, and the verification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, in particular to a method for identifying the distribution state of micro-stereoscopic structures on the surface of a product based on a mobile phone to verify the authenticity of the product. Background Art

[0002] During the manufacturing process, many products use printing finishing technology or mold processing technology to process the surface of the product with textures such as frosting, brushing, and star points, in order to improve the appearance of the surface. From the perspective of structural characteristics, the micro-stereoscopic structure presents a non-planar ups and downs in the vertical direction. After multiple three-dimensional structures are combined, the cross-section is formed with the same horizontal line as the reference, and the top and bottom surfaces show obvious unevenness. Under lighting conditions, due to the different reflection angles of the low and high points of each three-dimensional structure, diffuse reflection occurs, so that even a very small amount of uneven three-dimensional structure can present a state different from the mirror reflection of flat glass, similar to the diffuse reflection effect of frosted glass.

[0003] The unique optical properties of this micro-stereoscopic structure provide an important basis for product authenticity verification. However, existing image recognition technology faces many challenges when dealing with this type of micro-stereoscopic structure. For example: when the existing technology calculates the similarity of different positions in the image, if the mean is recalculated each time, it will lead to a large number of repeated calculations, and the calculation time will increase significantly. In addition, when the existing method identifies the distribution state of the micro-stereoscopic structure on the surface of the product, it may be affected by the global density change. At the junction of high-density and low-density areas, the existing method is difficult to accurately distinguish between normal points and abnormal points, resulting in a high false alarm rate that affects the accuracy of verification. Therefore, a method based on mobile phones to identify the distribution state of micro-stereoscopic structures on the surface of products to verify the authenticity of products is designed. Summary of the invention

[0004] The purpose of the present invention is to provide a method for identifying the distribution state of micro-stereoscopic structures on the surface of a product based on a mobile phone to verify the authenticity of the product, so as to solve the problem proposed in the above background technology that when calculating the similarity of different positions in an image, if the mean is recalculated each time, it will lead to a large number of repeated calculations and the existing method may be affected by global density changes when identifying the distribution state of micro-stereoscopic structures on the surface of a product.

[0005] To achieve the above object, the present invention provides a method for verifying the authenticity of a product based on identifying the distribution state of micro-stereoscopic structures on the surface of the product by a mobile phone, comprising the following steps:

[0006] S1. Use a mobile phone camera to capture the original image of the micro-stereoscopic structure distribution state on the surface of the product, and pre-process the original image;

[0007] S2. Extract features from the preprocessed original image through an image algorithm, compare the extracted features with the preset standard structure distribution template, calculate the similarity, introduce the integral graph for optimization in the process of calculating the similarity, and set a threshold in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic, and the final review result is notified to the user, otherwise it enters the S3 stage for review;

[0008] S3. Introduce anomaly detection algorithm to identify potential abnormal features, and then use local anomaly factor algorithm to further confirm the abnormal features, so as to determine the review results and verify whether the product is genuine.

[0009] As a further improvement of the technical solution, in S1, a built-in flash of the mobile phone is used to provide uniform lighting during the shooting process, and the mobile phone also has a built-in application that can adjust camera parameters according to the ambient light.

[0010] As a further improvement of the present technical solution, in the S1, the mobile phone also has a built-in application that can adjust the camera parameters according to the ambient light by using the built-in light sensor of the mobile phone to read the light intensity of the current environment, and divide the light intensity into low light, medium light and high light; adjust the ISO sensitivity, exposure time, white balance and flash according to the light intensity.

[0011] As a further improvement of the technical solution, in S2, the extracted features are compared with the preset standard structure distribution template to calculate the similarity, as follows:

[0012] There is an image I(x,y) to be matched and a template T(u,v), where (x,y) and (u,v) are the coordinates of the image and template respectively;

[0013] Calculate the mean:

[0014]

[0015] Among them, μ T is the mean value of template T; W T is the width of the template image; H T is the height of the template image; T(u,v) is the pixel value of the template image at position (u,v); u=0,1,…,W T -1; v = 0, 1, ..., H T -1;

[0016]

[0017] Among them, μ I (x, y) is the mean value of image I in the area covered by the template; I(x+u, y+v) is the pixel value of the image to be matched at position (x+u, y+v);

[0018] Calculate the variance:

[0019]

[0020] Among them, σ T is the variance of template T;

[0021]

[0022] Among them, σ I (x, y) is the variance of the image to be matched in the area covered by the template;

[0023] Compute the normalized cross-correlation:

[0024]

[0025] Wherein, NCC(x,y) is the normalized cross-correlation value.

[0026] As a further improvement of the technical solution, in S2, an integral graph is introduced to optimize the process of calculating the similarity, and the optimization is as follows:

[0027] Construct the integral graph II of the image to be matched I(x,y) and the template T(u,v) I (x,y) and II T (u,v);

[0028]

[0029] Among them, II(x,y) is the integral image; the value of the integral image II(x,y) at point (x,y) is the cumulative sum of all pixel values ​​from the upper left corner (0,0) to point (x,y) of the original image;

[0030] Calculate the mean:

[0031]

[0032] Among them, μ' T is the mean value of the optimized template T; II T (W T -1,H T -1) is the integral image value at the lower right corner of the template; μ' I (x, y) is the mean value of image I in the area covered by the template; II I (x+W T -1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y+H T -1) the cumulative sum of all pixel values; II I(x+W T -1,y-1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y-1) the cumulative sum of all pixel values; II I (x-1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x-1,y+H T -1) the cumulative sum of all pixel values; II I (x-1, y-1) is the cumulative sum of all pixel values ​​from the upper left corner of the image (0, 0) to the position (x-1, y-1);

[0033] Calculate the variance:

[0034]

[0035] Among them, σ' T is the variance of the optimized template image; II T (W T -1,H T -1) is; σ' I (x, y) is the variance of the optimized image to be matched in the template coverage area;

[0036] Compute the normalized cross-correlation:

[0037]

[0038] Wherein, NCC'(x,y) is the optimized normalized cross-correlation value.

[0039] As a further improvement of the technical solution, in S2, a threshold is set in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic. The specific steps are as follows:

[0040] Slide the template T(u,v) on the image to be matched I(x,y) to calculate the NCC value of each position and find the position with the largest NCC value (x max ,y max ), then this position is the position where the template best matches in the image; a similarity threshold τ is set in advance, if NCC(x max ,y max )≥τ, the match is considered successful; otherwise, the match fails.

[0041] As a further improvement of the present technical solution, in S2, when it is preliminarily determined to be authentic, the preliminary judgment result is displayed to the user and a first confidence score is provided. If the preliminary judgment result is uncertain, the user is prompted to retake the image and adjust the shooting angle.

[0042] As a further improvement of the present technical solution, if the review result is determined in S3, a second confidence score is provided; if the review result is still uncertain, the user is advised to perform manual review, and finally the final review result is notified to the user.

[0043] As a further improvement of the technical solution, S3 is specifically as follows:

[0044] The extracted features are formed into a feature vector set X = {x 1 ,x 2 ,…,x n}, use the anomaly detection algorithm to perform the first anomaly detection on the feature vector set X, identify potential abnormal features, generate the anomaly score A(x), and then verify the judgment by calculating the local reachable density and local anomaly factor.

[0045] As a further improvement of the technical solution, the review and judgment is performed by calculating the local reachable density and the local anomaly factor, as follows:

[0046] Compute the local reachability density:

[0047] For each feature vector x i , calculate the average distance d of its k nearest neighbors k (x i );

[0048]

[0049] Among them, lrd(x i ) is the data point x i The local reachability density; k is the number of neighbors; d(x i ,x j ) is x i and x j The distance between k (x j ) is the data point x j The distance to its kth nearest neighbor; N k (x i ) is the data point x i The set of k nearest neighbors of x; i is the i-th data point in the feature vector set; x j is the jth data point in the feature vector set;

[0050] Calculate the local outlier factor:

[0051]

[0052] Among them, LOF(x i ) is x iThe local anomaly factor of lrd(x j ) is the data point x j The local reachable density of

[0053] Review judgment:

[0054]

[0055] Among them, Confidence 2 Score the second confidence level; if Confidence 2 >Threshold 2 , where Threshold 2 is the second confidence threshold, the verification result is confirmed to be authentic and a second confidence score is provided; if Confidence 2 ≤Threshold 2 , if the review result is still uncertain, it is recommended that the user conduct a manual review.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. In the method for identifying the distribution state of micro-stereoscopic structure on the surface of a product based on a mobile phone to verify the authenticity of the product, an integral map is introduced for optimization in the process of calculating the similarity. The use of the integral map can reduce the time for calculating the regional mean, which means that no matter how large the template is, the time required to calculate the mean is at a constant level. When calculating the similarity of different positions in the image, if the mean is recalculated each time, a large number of repeated calculations will result. The use of the integral map can avoid this redundancy and improve the overall efficiency of the algorithm. The use of the integral map can ensure that the algorithm can complete a large number of calculations in a short time and meet the real-time requirements.

[0058] 2. In the method of using a mobile phone to identify the distribution state of the micro-stereoscopic structure on the surface of a product to verify the authenticity of the product, by calculating the local reachable density of each image data point, the density of the data point in its local environment can be more accurately reflected, which helps to distinguish normal points from abnormal points, especially at the junction of high-density and low-density areas. By comparing the local reachable density of a data point with its neighbors, abnormal points can be more accurately identified. The larger the value of the local abnormality factor, the more likely the point is to be an abnormal point, thereby improving the accuracy of verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 As shown, a method for identifying the distribution state of micro-stereoscopic structures on the surface of a product based on a mobile phone to verify the authenticity of the product is provided, including the following steps:

[0062] The anti-counterfeiting purpose is achieved by comparing and analyzing the image algorithm and the data of the printed three-dimensional structure. This mobile phone-based anti-counterfeiting solution not only improves the safety of the product, but also greatly enhances the user's sense of participation and trust, while also reducing the cost and complexity of traditional anti-counterfeiting measures.

[0063] S1. Use a mobile phone camera to capture the original image of the micro-stereoscopic structure distribution state of the product surface (such as paper or other printed surface or mold processing) (such as plastic, metal, etc.), and pre-process the original image;

[0064] In S1, the built-in flash of the phone is used to provide uniform lighting during shooting. The phone also has a built-in application that can adjust the camera parameters according to the ambient light to ensure the best image quality. The distribution state is manifested as dense point or line aggregation;

[0065] In S1, the phone also has a built-in application that can adjust the camera parameters according to the ambient light. By using the built-in light sensor of the phone to read the light intensity of the current environment, the light intensity is divided into low light, medium light and high light; ISO sensitivity, exposure time, white balance and flash are adjusted according to the light intensity. It ensures that high-quality images are captured under different lighting conditions, improves the accuracy of anti-counterfeiting verification and user experience, simplifies user operations, and enhances the reliability and adaptability of the system.

[0066] For example, the light intensity is divided into three levels, such as:

[0067] Low light: L<50L<50lux;

[0068] Medium light: 50≤L<500lux;

[0069] High light: L ≥ 500L ≥ 500lux;

[0070] Adjust ISO sensitivity:

[0071] When the light intensity is low: increase the ISO sensitivity to increase the image brightness, but avoid excessive noise;

[0072] ISO low =min(800,200+0.1×(50-L));

[0073] When the light intensity is medium: increase the ISO sensitivity moderately;

[0074] ISO medium =min(400,200+0.05×(L-50));

[0075] When the light intensity is high: lower the ISO sensitivity to reduce noise;

[0076] ISO high =max(100,200-0.05×(L-500));

[0077] Adjust exposure time:

[0078] When the light intensity is low: increase the exposure time to increase the brightness;

[0079] Exposure low =max(1 / 15,1 / 100+0.01×(50-L));

[0080] When the light intensity is medium: adjust the exposure time appropriately;

[0081] Exposure medium =max(1 / 100,1 / 100+0.005×(L-50));

[0082] When the light intensity is high: reduce the exposure time to avoid overexposure;

[0083] Exposure high =min(1 / 1000,1 / 100-0.005×(L-500));

[0084] Automatically adjust white balance:

[0085] When the light intensity is low light: Use the preset "Tungsten Light" white balance mode.

[0086] WB low = "Tungsten";

[0087] When the light level is medium: Use the preset "Auto" white balance mode.

[0088] WB medium = "Auto";

[0089] When the light intensity is bright: Use the preset "Daylight" white balance mode.

[0090] WB high = "Daylight";

[0091] Flash control:

[0092] When the light level is low: Turn on the flash and set it to a higher intensity.

[0093] Flash low = "High";

[0094] When the light level is medium: Turn on the flash and set it to medium intensity.

[0095] Flash medium = "Medium";

[0096] When the light intensity is high: turn off the flash.

[0097] Flash high = "Off";

[0098] In low-light conditions, the software algorithm reduces the noise caused by high ISO and maintains image clarity. If applicable, high dynamic range imaging technology is used to synthesize multiple photos with different exposures to preserve more highlights and shadows.

[0099] S2. Extract features from the preprocessed original image through an image algorithm, compare the extracted features with the preset standard structure distribution template, calculate the similarity, introduce the integral graph for optimization in the process of calculating the similarity, and set a threshold in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic, and the final review result is notified to the user, otherwise it enters the S3 stage for review;

[0100] In S2, the extracted features are compared with the preset standard structure distribution template to calculate the similarity, as follows:

[0101] There is an image I(x,y) to be matched and a template T(u,v), where (x,y) and (u,v) are the coordinates of the image and template respectively;

[0102] Calculate the mean:

[0103]

[0104] Among them, μ T is the mean value of template T; W T is the width of the template image; H T is the height of the template image; T(u,v) is the pixel value of the template image at position (u,v); u=0,1,…,W T -1; v = 0, 1, ..., HT -1; By calculating the mean, the influence of brightness changes on the matching results can be eliminated. The calculation of the mean helps to standardize the template so that it is not affected by global brightness changes, thereby improving the robustness of the matching.

[0105]

[0106] Among them, μ I (x, y) is the mean value of image I in the area covered by the template; I(x+u, y+v) is the pixel value of the image to be matched at position (x+u, y+v); this mean value will change as the template moves on the image. By calculating the local mean value, we can adapt to local brightness changes in the image and improve the accuracy of matching.

[0107] Calculate the variance:

[0108]

[0109] Among them, σ T is the variance of template T; it indicates the discreteness of the template pixel value; the calculation of variance helps to standardize the template so that it is not affected by contrast changes, thereby improving the robustness of matching.

[0110]

[0111] Among them, σ I (x, y) is the variance of the image to be matched in the area covered by the template, which indicates the degree of discreteness of the pixel values ​​in the area. By calculating the local variance, we can adapt to the local contrast changes in the image and improve the accuracy of matching.

[0112] Compute the normalized cross-correlation:

[0113]

[0114] Among them, NCC(x,y) is the normalized cross-correlation value; it is used to measure the similarity between template T and image I at position (x,y). The value range of NCC is [-1,1], where 1 indicates a complete match, -1 indicates a complete mismatch, and 0 indicates no correlation.

[0115] Normalized cross-correlation is a method widely used in image matching and template matching. It determines the best matching position by calculating the similarity between the image to be matched I(x, y) and the template T(u, v). By dividing by the standard deviation, the influence of brightness and contrast changes is eliminated, making the matching more robust. It can accurately reflect the similarity between the template and the image and is suitable for various complex image matching tasks.

[0116] In S2, the integral graph is introduced to optimize the process of calculating similarity. The integral graph is a data structure used to quickly calculate the sum of pixels in a rectangular area of ​​an image, which can greatly accelerate the calculation of the mean and variance. After optimization, it is as follows:

[0117] Construct the integral graph II of the image to be matched I(x,y) and the template T(u,v) I (x,y) and II T (u,v);

[0118]

[0119] Among them, II(x,y) is the integral image; the value of the integral image II(x,y) at the point (x,y) is the cumulative sum of all pixel values ​​from the upper left corner (0,0) to the point (x,y) of the original image; the integral image can quickly calculate the sum of pixels in any sub-region, thereby greatly speeding up the calculation of the mean and variance.

[0120] Calculate the mean:

[0121]

[0122] Among them, μ' T is the mean value of the optimized template T; II T (W T -1,H T -1) is the integral image value in the lower right corner of the template, which represents the sum of all pixel values ​​in the entire template; μ' I (x, y) is the mean value of image I in the area covered by the template; II I (x+W T -1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y+H T -1), which is the integral image value in the lower right corner of the template, representing the sum of all pixels in the template coverage area and its left and above; II I (x+W T -1,y-1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y-1), which is the integral image value of the area below the template, representing the sum of all pixels above the right side of the template; II I (x-1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x-1,y+H T -1), which is the integral image value of the left area of ​​the template, representing the sum of all pixels below the left side of the template; II I(x-1, y-1) is the cumulative sum of all pixel values ​​from the upper left corner of the image (0, 0) to the position (x-1, y-1). This is the integral image value of the upper left corner of the template, representing the sum of all pixels in the upper left corner of the template;

[0123] Calculate the variance:

[0124]

[0125] Among them, σ' T is the variance of the optimized template image; II T (W T -1,H T -1) is; σ' I (x, y) is the variance of the optimized image to be matched in the template coverage area;

[0126] Compute the normalized cross-correlation:

[0127]

[0128] Among them, NCC'(x,y) is the optimized normalized cross-correlation value;

[0129] Using the integral graph, the time complexity of calculating the regional mean can be reduced from O(W T ×H T ) is reduced to O(1). This means that no matter how large the template is, the time required to calculate the mean is constant, which is especially important for large-sized templates;

[0130] When calculating the similarity of different positions in the image, if the mean is recalculated each time, a large number of repeated calculations will result. The use of integral graphs can avoid this redundancy and improve the overall efficiency of the algorithm. The use of integral graphs can ensure that the algorithm can complete a large number of calculations in a short time and meet the real-time requirements.

[0131] In S2, a threshold is set in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic. The specific steps are as follows:

[0132] Slide the template T(u,v) on the image to be matched I(x,y) to calculate the NCC value of each position. Usually, the sliding step is 1, that is, sliding pixel by pixel; find the position with the largest NCC value (x max ,y max ), then this position is the position where the template best matches in the image; a similarity threshold τ is set in advance, if NCC(x max ,y max )≥τ, the match is considered successful; otherwise, the match fails.

[0133] In S2, when the product is initially judged to be authentic, the user is shown the preliminary judgment result and a first confidence score is provided. If the preliminary judgment result is uncertain, the user is prompted to retake the image and adjust the shooting angle;

[0134] S3. Introduce anomaly detection algorithm to identify potential abnormal features, and then use local anomaly factor algorithm to further confirm the abnormal features, so as to determine the review results and verify whether the product is genuine;

[0135] If the review result in S3 is confirmed, a second confidence score is provided to indicate the system's confidence in the judgment result. If the review result is still uncertain, the user is advised to conduct a manual review, and finally the user is notified of the final review result.

[0136] The first confidence score is provided in the preliminary judgment stage, based on the results of normalized cross-correlation or other preliminary feature matching algorithms, and indicates the system's confidence in the preliminary judgment results. The second confidence score is provided in the review stage, based on the results of the anomaly detection algorithm, and indicates the system's confidence in the review judgment results.

[0137] S3 is as follows:

[0138] The extracted features are formed into a feature vector set X = {x 1 ,x 2 ,…,x n}, use the anomaly detection algorithm to perform the first anomaly detection on the feature vector set X, identify potential abnormal features, and generate anomaly scores A(x), which represent the degree of abnormality of each feature vector. The preliminary anomaly scores can help us focus on possible abnormal points more efficiently in the subsequent local anomaly factor calculation, thereby improving the calculation efficiency and accuracy. Specifically, by randomly splitting the data points, the path length of each data point is calculated. The shorter the path length, the more likely the data point is an abnormal point. Then, the local reachable density and local anomaly factor are calculated for review and judgment.

[0139] The local reachable density and local anomaly factor are calculated to verify the judgment, as follows:

[0140] Compute the local reachability density:

[0141] For each feature vector x i , calculate the average distance d of its k nearest neighbors k (x i );

[0142]

[0143] Among them, lrd(x i ) is the data point x iThe local reachability density; k is the number of neighbors; d(x i ,x j ) is x i and x j The distance between k (x j ) is the data point x j The distance to its kth nearest neighbor; N k (x i ) is the data point x i The set of k nearest neighbors of x; i is the i-th data point in the feature vector set; x j is the jth data point in the feature vector set;

[0144] Local reachability density lrd(x i ) represents the data point x i The density of x among its k nearest neighbors. Specifically, it is i The inverse of the average distance to its k nearest neighbors. i The k nearest neighbors of are very close to it, then lrd(x i ) will be very large, indicating that x i is in a high density region; on the contrary, if x i The k nearest neighbors of are far away from it, so lrd(x i ) will be very small, indicating that x i In a low-density area.

[0145] Calculate the local outlier factor:

[0146]

[0147] Among them, LOF(x i ) is x i The local anomaly factor of lrd(x j ) is the data point x j The local reachable density of

[0148] Local outlier factor LOF(x i ) represents the data point x i The degree of abnormality of x relative to its k nearest neighbors. Specifically, it is i The average of the local reachability densities of the k nearest neighbors is equal to x i The ratio of its local reachability density to its own. If LOF(x i ) is close to 1, indicating that x i The density of its neighbors is similar and it is unlikely to be an outlier. i ) is much greater than 1, indicating that x iIt is located in a low-density area and is an outlier relative to its neighbors.

[0149] By calculating the local reachable density of each data point, the density of the data point in its local environment can be more accurately reflected, which helps to distinguish normal points from abnormal points, especially at the junction of high-density and low-density areas. By comparing the local reachable density of a data point with its neighbors, abnormal points can be identified more accurately. The larger the value of the local anomaly factor, the more likely the point is to be an abnormal point.

[0150] Review judgment:

[0151]

[0152] Among them, Confidence 2 Score the second confidence level; if Confidence 2 >Threshold 2 , where Threshold 2 is the second confidence threshold, the verification result is confirmed to be authentic and a second confidence score is provided; if Confidence 2 ≤Threshold 2 , if the review result is still uncertain, the user is advised to conduct further manual review.

[0153] The above shows and describes 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 by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product using a mobile phone, characterized in that: The steps include: S1. Use a mobile phone camera to capture the original image of the micro-stereoscopic structure distribution state on the surface of the product, and pre-process the original image; S2. Extract features from the preprocessed original image through an image algorithm, compare the extracted features with the preset standard structure distribution template, calculate the similarity, introduce the integral graph for optimization in the process of calculating the similarity, and set a threshold in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic, and the final review result is notified to the user, otherwise it enters the S3 stage for review; S3. Introduce anomaly detection algorithm to identify potential abnormal features, and then use local anomaly factor algorithm to further confirm the abnormal features, so as to determine the review results and verify whether the product is genuine.

2. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 1, characterized in that: In the S1, the built-in flash of the mobile phone is used to provide uniform lighting during the shooting process. The mobile phone also has a built-in application that can adjust the camera parameters according to the ambient light.

3. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 2, characterized in that: In the S1, the mobile phone also has a built-in application that can adjust the camera parameters according to the ambient light by using the built-in light sensor of the mobile phone to read the light intensity of the current environment and divide the light intensity into low light, medium light and high light; Adjust ISO sensitivity, exposure time, white balance and flash according to light intensity.

4. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 3, characterized in that: In S2, the extracted features are compared with the preset standard structure distribution template to calculate the similarity, as follows: There is an image I(x,y) to be matched and a template T(u,v), where (x,y) and (u,v) are the coordinates of the image and template respectively; Calculate the mean: Among them, μ T is the mean value of template T; W T is the width of the template image; H T is the height of the template image; T(u,v) is the pixel value of the template image at position (u,v); u=0,1,…,W T -1; v = 0, 1, ..., H T -1; Among them, μ I (x, y) is the mean value of image I in the area covered by the template; I(x+u, y+v) is the pixel value of the image to be matched at position (x+u, y+v); Calculate the variance: Among them, σ T is the variance of template T; Among them, σ I (x, y) is the variance of the image to be matched in the area covered by the template; Compute the normalized cross-correlation: Wherein, NCC(x,y) is the normalized cross-correlation value.

5. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 4, characterized in that: In S2, the integral graph is introduced to optimize the process of calculating the similarity, and the optimization is as follows: Construct the integral graph II of the image to be matched I(x,y) and the template T(u,v) I (x,y) and II T (u,v); Among them, II(x,y) is the integral image; the value of the integral image II(x,y) at point (x,y) is the cumulative sum of all pixel values ​​from the upper left corner (0,0) to point (x,y) of the original image; Calculate the mean: Among them, μ' T is the mean value of the optimized template T; II T (W T -1,H T -1) is the integral image value at the lower right corner of the template; μ' I (x, y) is the mean value of image I in the area covered by the template; II I (x+W T -1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y+H T -1) the cumulative sum of all pixel values; II I (x+W T -1,y-1) is from the upper left corner of the image (0,0) to the position (x+W T -1,y-1) the cumulative sum of all pixel values; II I (x-1,y+H T -1) is from the upper left corner of the image (0,0) to the position (x-1,y+H T -1) the cumulative sum of all pixel values; II I (x-1, y-1) is the cumulative sum of all pixel values ​​from the upper left corner of the image (0, 0) to the position (x-1, y-1); Calculate the variance: Among them, σ' T is the variance of the optimized template image; II T (W T -1,H T -1) is; σ' I (x, y) is the variance of the optimized image to be matched in the template coverage area; Compute the normalized cross-correlation: Wherein, NCC'(x,y) is the optimized normalized cross-correlation value.

6. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 5, characterized in that: In S2, a threshold is set in advance. If the similarity exceeds the threshold, the product is preliminarily judged to be authentic. The specific steps are as follows: Slide the template T(u,v) on the image to be matched I(x,y) to calculate the NCC value of each position and find the position with the largest NCC value (x max ,y max ), then this position is the position where the template best matches in the image; a similarity threshold τ is set in advance, if NCC(x max ,y max )≥τ, the match is considered successful; otherwise, the match fails.

7. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 6, characterized in that: In S2, when the product is initially judged to be authentic, the preliminary judgment result is displayed to the user and a first confidence score is provided. If the preliminary judgment result is uncertain, the user is prompted to retake the image and adjust the shooting angle.

8. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 7, characterized in that: If the review result in S3 is confirmed, a second confidence score is provided. If the review result is still uncertain, the user is advised to perform manual review. Finally, the final review result is notified to the user.

9. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 8, characterized in that: The S3 is as follows: The extracted features are formed into a feature vector set X = {x1, x2, ..., x n }, use the anomaly detection algorithm to perform the first anomaly detection on the feature vector set X, identify potential abnormal features, generate the anomaly score A(x), and then verify the judgment by calculating the local reachable density and local anomaly factor.

10. The method for verifying the authenticity of a product by identifying the distribution state of micro-stereoscopic structures on the surface of the product based on a mobile phone according to claim 9, characterized in that: The review and judgment is performed by calculating the local reachable density and the local anomaly factor, as follows: Compute the local reachability density: For each feature vector x i , calculate the average distance d of its k nearest neighbors k (x i ); Among them, lrd(x i ) is the data point x i The local reachability density; k is the number of neighbors; d(x i ,x j ) is x i and x j The distance between k (x j ) is the data point x j The distance to its kth nearest neighbor; N k (x i ) is the data point x i The set of k nearest neighbors of x; i is the i-th data point in the feature vector set; x j is the jth data point in the feature vector set; Calculate the local outlier factor: Among them, LOF(x i ) is x i The local anomaly factor of lrd(x j ) is the data point x j The local reachable density of Review judgment: Among them, Confidence2 is the second confidence score; if Confidence2>Threshold2, where Threshold2 is the threshold of the second confidence, the review result is determined to be authentic and a second confidence score is provided; if Confidence2≤Threshold2, the review result is still uncertain, and the user is advised to perform manual review.

Images