A method and system for optical character verification of a color inkjet SD card

Abstract

The application discloses a color jet SD card optical character verification method and a verification system, and belongs to the technical field of digital image processing. The application firstly proposes a fast character verification method based on the HSV histogram characteristics, realizes fast verification of characters with large differences from a template image, quickly obtains the recognition result of characters with obvious differences, reduces the calculation amount of the next precise verification, and effectively improves the recognition efficiency. Secondly, a HOG feature extraction method resisting light and rotation changes is proposed, the verification accuracy of characters with small rotation and light changes is improved. Finally, a centralized generalized Jaccard coefficient is designed, the discrimination of the distance measure is enhanced, the verification precision of similar characters is improved, and the method can be widely applied to color jet image quality automatic inspection technology.

Description

TECHNICAL FIELD

[0001] The application relates to a color jet SD card optical character verification method and a verification system and belongs to the technical field of digital image processing. BACKGROUND

[0002] In the color jet process of the SD card, the printed surface characters of the SD card do not match the real model due to human errors or equipment failures, which increases the high cost of manufacturers and affects the use experience of consumers. At present, the manual verification method is generally used in the color jet process, which is low in efficiency and easy to be affected by subjective factors. Therefore, a color jet image quality automatic inspection method based on machine vision needs to be researched to replace manual inspection.

[0003] Optical character verification (OCV) is the core link of the color jet image quality automatic inspection, and the essence is to verify whether the surface characters of the to-be-detected image and the template image are the same. OCV has a wide range of applications in the fields of target detection and computer vision. The OCV algorithm can be divided into two categories: a verification algorithm based on deep learning and a verification algorithm based on matching.

[0004] In the research of deep learning verification algorithm, "Lin C H, Wang S H, Lin C J. Using convolutional neural networks for character verification on integrated circuit components of printed circuit boards [J]. Applied Intelligence, 2019, 49(11): 4022-4032." uses the structure and refinement mechanism in CNN to detect the characters of misplaced, missing and reverse polarity parts in printed circuit boards, with high accuracy. "Ribeiro F D S, Caliva F, Swainson M, et al. An adaptable deep learning system for optical character verification in retail food packaging [C] / / 2018 IEEE Conference on Evolving and Adaptive Intelligent Systems (EAIS). IEEE, 2018: 1-8." uses k-means clustering algorithm to cluster the features extracted by CNN, and uses k-nearest neighbor method combined with the centroid calculated by CNN to effectively identify the incorrect or missing information in retail packaging. Although the detection accuracy of the two algorithms is high, a large amount of data is needed for training and fitting, and the operation amount is large and the interpretability is weak.

[0005] In the research of matching verification algorithm, Liu Yifei, Zhang Xuming, and Ding Mingyue. Multi-core signal processor implementation of normalized cross-correlation grayscale image matching [J]. Computer Application, 2011, 31(12): 3334-3336. divides the detection area into six parts and searches in parallel, realizing a multi-core normalized cross-correlation image matching algorithm, but it is sensitive to the change of image gray value and has low robustness. Li Shenghui, Shi Ruizhi, and Ye Hui. Image matching algorithm combining color invariant and ORB feature [J]. Packaging Engineering, 2016, 37(15): 18-23. combines the color invariant of color image with ORB feature, which can realize matching under different light directions, but this algorithm requires the image to have more feature points, which has obvious limitations. Zuo Chuan, and Pang Chunjiang. Anti-perspective transformation fast image matching algorithm based on improved ORB [J]. Journal of Sensing Technology, 2018, 31(11): 1714-1720. uses the accelerated segmentation detection algorithm to extract ORB feature and uses Hamming distance as the similarity criterion, which has strong anti-perspective transformation capability, but the accuracy is not high. Liu Huan, Hao Jingrong, Ding Yongsheng, and Mao Min. Light-robust anti-fuzzy new combined invariant moment image matching method [J]. Journal of Sensing Technology, 2013, 26(09): 1258-1264. proposes a three-combined Hu invariant moment method for image matching, which can solve the problem of detection accuracy decline caused by fuzzy noise, but the algorithm depends on the correctness of the extracted corner information. The above matching-based algorithms will quickly decrease the matching verification accuracy in the OCV problem of light change, target rotation, or high similarity. Figure 1 (a) is an actually collected image, Figure 1 (b) is an SD card structure diagram, Figure 2 is a complex and diverse SD card schematic diagram in the production process. SUMMARY

[0006] In order to improve the efficiency and accuracy of the SD card optical character verification method, the present application provides a color jet SD card optical character verification method and verification system, and the technical solution is as follows:

[0007] The first object of the present application is to provide a color jet SD card optical character verification method, which comprises:

[0008] Step 1: convert the test image and the template image to the HSV color space, calculate the H, S, and V three-channel color histogram similarity of the template image and the test image using the Bhattacharyya coefficient, if the similarity value is less than the set threshold, directly output the final verification result "different", if the similarity value is not less than the set threshold, proceed to step 2 for further verification;

[0009] Step 2: extract the V channel image of the test image, and extract the HOG feature vector based on the V channel image;

[0010] Step 3: Calculate the similarity between the HOG feature vector of the image to be tested and the feature vector of the template image based on the centralized generalized Jaccard coefficient, and compare the similarity with a set threshold value. If the similarity is greater than the threshold value, output the final verification result as "same", otherwise output "different".

[0011] Optionally, the process of extracting the HOG feature vector in step 2 includes:

[0012] Step 21: Divide the image to be tested into four detection regions, and extract the V channel image V i Gamma correction is performed;

[0013] Step 22: Take 3x3 pixels as a gradient extraction unit, and take the center point p(x, y) as the rotation center. Rotate the surrounding 8 pixels clockwise by 1, 2, and 3 pixels in turn, and calculate the horizontal and vertical direction gradients of the gradient extraction unit without rotation and after rotating 3 times;

[0014] Step 23: Divide the image into cells and blocks of the same size, count the gradient direction histograms of all gradient extraction units without rotation and after rotating 3 times in each cell, and divide the gradient direction of 180 degrees of the cell into n direction blocks called bins, to obtain 4 n-dimensional feature vectors;

[0015] Step 24: Concatenate and combine the feature vectors of all cells in a block, normalize them, and perform sliding window operation along the row and column directions according to a certain step size to obtain multiple blocks. Then calculate the feature vector of each block, concatenate the feature vectors of each block together, and obtain the final 4 HOG feature vectors H i1 , H i2 , H i3 , H i4 .

[0016] Optionally, step 3 specifically includes:

[0017] Calculate the similarity between the 4 HOG feature vectors H i1 , H i2 , H i3 , H i4 of the image to be tested and the feature vector H t of the template image based on the centralized generalized Jaccard coefficient, select the maximum similarity, and compare the maximum similarity with a set threshold value. If the similarity of the four detection regions is greater than the threshold value, output the final verification result as "same", otherwise output "different".

[0018] Optionally, the central generalized Jaccard coefficient is:

[0019]

[0020] wherein, is a similarity value, is a feature vector H of the template image t , is a feature vector H of the image to be tested i , x i and y i are the i-th elements of H t and H i , respectively, and are the average values of all elements of H t and H i , respectively, and n represents the total dimension of the feature vector.

[0021] Optionally, the verification process of step 1 comprises:

[0022] Step 11: dividing ROI, and locating the image to be tested within the ROI using a template matching algorithm;

[0023] Step 12: converting the template image and the image to be tested from the RGB color space to the HSV color space;

[0024] Step 13: calculating the H, S, and V channel color histogram similarity of the template image and the image to be tested using the Bhattacharyya coefficient, comparing the three similarity values with a set threshold value, and if all are smaller than the threshold value, outputting the final verification result "different".

[0025] Optionally, the calculation formula of the Bhattacharyya coefficient is:

[0026]

[0027] wherein, p(i) and q(i) represent the probability distribution of the i-th gray level of the histogram of the template image and the image to be tested, respectively, and N represents the total number of gray levels.

[0028] A second object of the present application is to provide an inkjet SD card optical character verification system, comprising:

[0029] at least one storage medium, the at least one storage medium storing at least one instruction set for verifying the inkjet SD card optical character; and at least one processor in communication connection with the at least one storage medium,

[0030] When the color jet SD card optical character verification system is running, the at least one processor reads the at least one instruction set and implements the color jet SD card optical character verification method described above.

[0031] Optionally, the image acquisition device is further included for acquiring the SD card image to be detected and the template image.

[0032] Optionally, the display device is further included for outputting the final verification result.

[0033] Optionally, the image acquisition device is a camera.

[0034] The present application has the following beneficial effects:

[0035] The present application firstly proposes a fast character verification method based on the HSV histogram characteristics, realizes fast verification of characters with large differences from the template image, quickly obtains the recognition result of characters with obvious differences, reduces the calculation amount of the next precise verification, and effectively improves the recognition efficiency; secondly, proposes an HOG feature extraction method against light and rotation changes, improves the verification accuracy of characters with small rotation and light changes; and finally, designs a centralized generalized Jaccard coefficient, enhances the discrimination of the distance measure, and improves the verification accuracy of similar characters. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Figure 1 is a schematic diagram of an SD card, wherein (a) is an SD card image collected in an actual production environment, and (b) is a character distribution structure diagram of the SD card.

[0038] Figure 2 is a complex and diverse SD card image in the production process, wherein (a) is an SD card of different series, (b) is an SD card of the same series with similar characters, (c) is the same SD card under different lighting conditions, and (d) is an SD card with a small rotation.

[0039] Figure 3 is a general flowchart of the color jet SD card optical character verification method based on the centralized Jaccard matching of the present application.

[0040] Figure 4 (a) and (b) are respectively the HSV three-channel color histogram comparison effect diagrams of two sample SD card images in the second embodiment of the present application.

[0041] Figure 5 is the gradient extraction unit rotation schematic diagram of embodiment two of the present application.

[0042] Figure 6 is the flow chart of the improved HOG feature algorithm of the present application.

[0043] Figure 7 is the similarity curve diagram of three common distance measure calculations, wherein (a) is the cosine of angle similarity curve diagram, (b) is the correlation coefficient similarity curve diagram, and (c) is the generalized Jaccard coefficient similarity curve diagram.

[0044] Figure 8 is the similarity curve diagram of the improved generalized Jaccard coefficient calculation of the present application.

[0045] Figure 9 is the distance measure detection performance comparison diagram of the present application.

[0046] Figure 10 is the feature vector similarity comparison result diagram of the present application, wherein (a) is the logo area similarity value comparison diagram, (b) is the memory area similarity value comparison diagram, (c) is the serial number area similarity value comparison diagram, and (d) is the speed level area similarity value comparison diagram. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present application more clear, the following will make further detailed description on the embodiments of the present application in combination with the drawings.

[0048] Embodiment one:

[0049] The embodiment provides a color jet SD card optical character verification method, comprising:

[0050] Step 1: converting the to-be-detected image and the template image into the HSV color space, calculating the H, S and V three-channel color histogram similarity of the to-be-detected image and the template image by using the Bhattacharyya coefficient, if the similarity value is less than the set threshold value, directly outputting the final verification result “different”, if the similarity value is not less than the set threshold value, performing step 2 to continue verification;

[0051] Step 2: extracting the V channel image of the to-be-detected image, and extracting the HOG feature vector based on the V channel image;

[0052] Step 3: calculating the similarity of the HOG feature vector of the to-be-detected image and the feature vector of the template image based on the centralized generalized Jaccard coefficient, and comparing the similarity with the set threshold value, if the similarity is greater than the threshold value, outputting the final verification result as “same”, otherwise outputting “different”.

[0053] Embodiment Two:

[0054] The embodiment provides a color jet SD card optical character verification method based on centralized Jaccard matching, and the method is described in detail as follows. Figure 3 .

[0055] Firstly, the embodiment provides a fast verification method based on HSV color histogram analysis, which can accurately detect SD card images with large differences; secondly, aiming at the problem that the character verification accuracy of the SD card is affected by light changes and slight rotation, the resistance of the HOG feature to light and rotation changes is optimized by extracting the V channel image and the change angle feature, and the detection accuracy is improved; finally, aiming at the verification problem of similar SD card characters, a centralized generalized Jaccard coefficient is proposed, the discrimination of the distance measure is enhanced, and the precision detection accuracy of the color jet character is improved.

[0056] The fast verification method based on HSV color histogram analysis of the embodiment specifically includes the following steps.

[0057] Step 1. Dividing ROI (region of interest, such as Figure 1 (a) shown in the box), locating the to-be-detected image in the ROI by using a template matching algorithm;

[0058] Step 2. Converting the template image and the to-be-detected image from the RGB color space to the HSV color space;

[0059] Step 3. Calculating the H, S and V three-channel color histogram similarity of the template image and the to-be-detected image by using the Bhattacharyya coefficient, comparing the three similarity values with the set threshold value, and if all the three similarity values are less than the threshold value, outputting the final verification result "different". Figure 4 For Figure 2 (a1), Figure 2 (a2) The three-channel histogram comparison effect of the two kinds of SD cards. The Bhattacharyya coefficient calculation formula is as follows.

[0060]

[0061] In the formula, p(i) and q(i) respectively represent the probability distribution of the i-th gray level of the template image and the to-be-detected image histogram, and N represents the total number of gray levels.

[0062] The precision verification method based on HOG similarity analysis of the embodiment specifically includes the following steps.

[0063] When the template image and the image to be tested differ significantly, rapid verification can directly determine the final output result. However, when the apparent characters of the two images are similar, rapid verification will fail to provide an accurate judgment, requiring further precise verification. Feature extraction and similarity calculation are crucial steps in determining the accuracy of precise verification. Samples with slight rotations and lighting changes introduce interference factors during feature extraction. Furthermore, when the surface characters of the template image and the image to be tested are highly similar, the similarity calculation will also be biased. To address these issues, this embodiment optimizes both HOG features and similarity metrics, proposing an improved precise verification method based on feature similarity, as detailed below:

[0064] Step 1. Divide the image into four detection regions (e.g., ...). Figure 1 (b) As shown in the box), extract the V channel image for each region. i Gamma correction is performed. Compared to grayscale images, V-channel images exhibit higher illumination robustness.

[0065] Step 2. Using a 3×3 pixel gradient extraction unit and the center point p(x,y) as the rotation center, rotate the surrounding 8 pixels clockwise by 1, 2, and 3 pixels respectively. Calculate the horizontal and vertical gradients of the gradient extraction unit without rotation and after 3 rotations. The rotation diagram is shown below. Figure 5 As shown. For each gradient extraction unit, the surrounding pixels can be rotated 8 times, but the gradient magnitude and gradient direction of the last 4 rotations are the same as those of the first 4 rotations, so only the gradient information of the first 4 rotations is needed;

[0066] Step 3. Divide the image into cells and blocks of the same size, and calculate the gradient direction histograms of all gradient extraction units within each cell, both with and without rotation after 3 rotations. Divide the gradient directions of each cell into n directional blocks called bins by 180 degrees, resulting in 4 n-dimensional feature vectors;

[0067] Step 4. Concatenate and normalize the feature vectors of all cells within a block. Perform a sliding window operation along the row and column directions with a certain step size to obtain multiple blocks. Calculate the feature vector of each block and concatenate them to obtain the final four HOG feature vectors H. i1 H i2 H i3 H i4 The improved HOG feature algorithm flow is as follows: Figure 6 As shown;

[0068] Step 5. Calculate the four HOG feature vectors H of the image under test based on the centered generalized Jaccard coefficients. i1 H i2 H i3H i4 With the feature vector H of the template image t The similarity score is used to compare the highest similarity score with a set threshold. If the similarity score of all four detected regions is greater than the threshold, the final verification result is output as "same"; otherwise, "different" is output.

[0069] Distance measures can quantify the similarity of feature vectors and determine the verification accuracy. Common distance measures include cosine similarity, correlation coefficient, and generalized Jaccard coefficient, with the following expressions (2), (3), and (4).

[0070]

[0071]

[0072]

[0073] In the formula The similarity value is... The feature vector H of the template image t , H is the feature vector of the image to be tested. i x i and y i H respectively t and H i The i-th element, and H respectively t and H i The average of all elements, where n represents the total dimension of the eigenvector.

[0074] This embodiment designs a comparative experiment for these different distance measures. The template image is selected as... Figure 2 (b1) Characters within the boxed area; 100 images with the same character are selected as the validation sample set S, similar characters ( Figure 2 (b2) One hundred images of the boxed characters were used as the control sample set D. The comparative experimental curve is shown below. Figure 7 As shown. Among them, the cosine of the included angle ( Figure 7 (a) and correlation coefficient curve ( Figure 7 (b) shows crossover across multiple samples with small inter-class variance, which will introduce validation error; in contrast, the generalized Jaccard coefficient curve ( Figure 7 (c) has the best performance and increased inter-class variance, but there is still overlap on some samples (such as images 83 and 84), which poses a risk of misclassification.

[0075] It can be seen that compared with other distance measures, the generalized Jaccard coefficient has greater discrimination for similar characters. Since the dimension of the feature vector is large, and the dimension of the similar part of the character accounts for a small proportion of the total dimension, the generalized Jaccard coefficient cannot well amplify this part. The embodiment designs a centralized generalized Jaccard coefficient, and the expression is as follows.

[0076]

[0077] After improvement, the feature vector is centrally processed, and each element is subtracted by the average of all elements. Compared with before centralization, the element curve distribution of the squared or multiplied feature vector is more uniform, and the data is basically distributed above and below the origin, and the intra-class variance is smaller, so that the difference part of the two feature vectors can be better highlighted. The improved generalized Jaccard coefficient curve of the embodiment is as shown in Figure 8 , the data is not crossed and can be separated by a threshold curve 0.82, which increases the inter-class variance and improves the accuracy of character verification.

[0078] In order to further prove the beneficial effects of the present application, the character verification process designed according to the present application is tested for effectiveness based on the actually collected SD card images.

[0079] The test environment is: the CPU is AMD Ryzen 7 5800H, the memory is 16GB. The operating system is: Windows 10 Professional. The development environment is: Visual Studio 2015.

[0080] In order to verify the effectiveness and stability of the highly similar character verification algorithm proposed in the present application, the following comparative test is designed.

[0081] The test image set selects 5 template images in each of the 4 detection areas; for each template image, 100 images of the same character are selected as the verification sample set, and 100 images of similar characters are selected as the control sample set. The cosine of the included angle, the correlation coefficient, the normalized Euclidean distance, the combination of the cosine of the included angle and the correlation coefficient (addition and average), the generalized Jaccard coefficient, and the improved generalized Jaccard coefficient of the present application are used to calculate the similarity values of the template image and the verification sample set and the control sample set, respectively, a total of 100 groups. In order to intuitively observe the accuracy of the distance measure in the OCV problem, the Euclidean distances of the 100 similarity values are calculated, the greater the distance, the greater the inter-class variance, and the stronger the resolution ability of the distance measure. The Euclidean distance calculation formula is as follows.

[0082]

[0083] In the formula, x i and yi are the i-th similarity values of the verification sample set and the control sample set images, respectively. The experimental results are shown in Figure 9 The experimental results show that, for the 20 template images of the four functional areas, the average Euclidean distance of the logo area is 2.331, the average Euclidean distance of the memory area is 5.240, the average Euclidean distance of the serial number area is 3.471, the average Euclidean distance of the speed grade area is 4.649, and the Euclidean distance data of each detection area fluctuates slightly, which shows that it has high stability and strong resolution ability for similar images.

[0084] To verify the effectiveness of the micro-rotating character verification algorithm proposed in the present application, the following comparative experiment is designed.

[0085] The experimental image set uses one template image for each of the four detection areas; for each template image, 100 verification sample sets with micro-rotation of the same model are selected. The feature vectors of the template images and the images to be tested are extracted based on the traditional HOG feature and the improved HOG feature of the present application, respectively, and the similarity values of the feature vectors are compared. The experimental results are shown in Figure 10 The experimental results show that when the image appears micro-rotation, the similarity values of the four functional areas are significantly improved, with an average increase of 0.163, 0.132, 0.188, and 0.159, respectively, verifying the effectiveness of the improved HOG feature in resisting rotation changes.

[0086] To verify the accuracy and robustness of the verification method proposed in the present application, the following ablation experiment is designed.

[0087] The experimental image set uses 9 template images of different models; for each template image, 20 images under 3 acquisition conditions are selected as verification sample sets, a total of 9x3x20=540. Traditional HOG feature+generalized Jaccard (Traditional HOG+Generalized Jaccard), traditional HOG feature+improved generalized Jaccard (Traditional HOG+Revised Generalized Jaccard), improved HOG feature+generalized Jaccard (Revised HOG+Generalized Jaccard), and the present application (Revised HOG+Revised Generalized Jaccard) are used as verification methods, a total of 9x540=4860.

[0088] Table 1 Verification accuracy statistical table

[0089]

[0090] The test results show that the verification method proposed in the application has the highest verification accuracy of 99.15% compared with several other combined verification methods, and verifies the feasibility and robustness of the method, and can meet the production requirements.

[0091] Part of the steps in the embodiments of the application can be implemented by software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0092] The above only describes the preferred embodiments of the application and is not intended to limit the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for verifying optical characters on a color inkjet SD card, characterized in that, The optical character verification method for inkjet SD cards includes: Step 1: Convert the image to be tested and the template image to the HSV color space. Calculate the similarity of the H, S, and V color histograms of the template image and the image to be tested using the Barthel coefficient. If the similarity value is less than the set threshold, the final verification result "dissimilar" is directly output. If the similarity value is not less than the set threshold, proceed to Step 2 for further verification. Step 2: Extract the V channel image of the image to be tested, and extract the HOG feature vector based on the V channel image; Step 3: Calculate the similarity between the HOG feature vector of the image to be tested and the feature vector of the template image based on the centered generalized Jaccard coefficient, and compare the similarity with a set threshold. If the similarity is greater than the threshold, the final verification result is output as "same", otherwise "different". The process of extracting HOG feature vectors in step 2 includes: Step 21: Divide the image to be tested into four detection regions, and extract the V channel image V from each region. i Perform Gamma correction; Step 22: Using 3×3 pixels as the gradient extraction unit and the center point p(x,y) as the rotation center, rotate the surrounding 8 pixels clockwise by 1, 2, and 3 pixels in sequence, and calculate the horizontal and vertical gradients of the gradient extraction unit after no rotation and after 3 rotations respectively. Step 23: Divide the image into cells and blocks of the same size, and calculate the gradient direction histograms of all gradient extraction units in each cell with no rotation and after 3 rotations. Divide the gradient direction of the cell into n direction blocks called bins by 180 degrees to obtain 4 n-dimensional feature vectors. Step 24: Concatenate and normalize the feature vectors of all cells within a block. Perform a sliding window operation along the row and column directions with a certain step size to obtain multiple blocks. Calculate the feature vector of each block and concatenate the feature vectors of each block to obtain the final four HOG feature vectors H. i1 H i2 H i3 H i4 .

2. The optical character verification method for inkjet SD cards according to claim 1, characterized in that, Step 3 specifically includes: The four HOG feature vectors H of the image under test are calculated based on the centered generalized Jaccard coefficient. i1 H i2 H i3 H i4 With the template image feature vector H t The similarity is selected, and the largest similarity is compared with the set threshold. If the similarity of all four detection regions of the two is greater than the threshold, the final verification result is output as "same"; otherwise, "different" is output.

3. The method for verifying optical characters on a color inkjet SD card according to any one of claims 1-2, characterized in that, The centralized generalized Jaccard coefficient is: in, The similarity value. The feature vector H of the template image t , H is the feature vector of the image to be tested. i x i and y i H respectively t and H i The i-th element, and H respectively t and H i The average of all elements, where n represents the total dimension of the eigenvector.

4. The optical character verification method for inkjet SD cards according to claim 1, characterized in that, The verification process in step 1 includes: Step 11: Divide the ROI and locate the image to be tested within the ROI using a template matching algorithm; Step 12: Convert the template image and the image to be tested from the RGB color space to the HSV color space; Step 13: Calculate the similarity of the H, S, and V channel color histograms of the template image and the image to be tested using the Bach coefficient. Compare the three similarity values ​​with a set threshold. If all three values ​​are less than the threshold, output the final verification result "dissimilar".

5. The method for verifying optical characters on a color inkjet SD card according to claim 4, characterized in that, The formula for calculating the Bartholomew's coefficient is as follows: Where p(i) and q(i) represent the probability distribution of the i-th gray level of the histogram of the template image and the image to be tested, respectively, and N represents the total number of gray levels.

6. A color inkjet SD card optical character verification system, characterized in that, The inkjet SD card optical character verification system includes: At least one storage medium storing at least one instruction set for verifying optical characters on a color inkjet SD card; and at least one processor communicatively connected to the at least one storage medium. When the inkjet SD card optical character verification system is running, the at least one processor reads the at least one instruction set and implements the inkjet SD card optical character verification method as described in any one of claims 1-5.

7. The optical character verification system for inkjet SD cards according to claim 6, characterized in that, It also includes an image acquisition device for acquiring images of the SD card to be inspected and template images.

8. The optical character verification system for inkjet SD cards according to claim 6, characterized in that, It also includes a display device for outputting the final verification results.

9. The optical character verification system for inkjet SD cards according to claim 7, characterized in that, The image acquisition device is a camera.

Images