An artificial marking error correction method and system based on an image grayscale calculation method

Abstract

The present invention relates to the field of image processing technology, and specifically provides a manual marking error correction method and system based on an image grayscale calculation method, including the following steps: Using laser scanning technology, scan the student's answer sheet into a digital image to generate a digital image of the student's answer sheet. In the present invention, through the improved laser scanning technology and image processing algorithm, the digital processing of a large number of test papers can be completed efficiently. The Sobel edge detection algorithm and morphological processing are used to accurately segment the text area, which can accurately locate the student's answering area, avoiding the situation of missing or misjudging. By using Otsu's adaptive threshold method to optimize the image clarity and the K-means clustering algorithm to analyze the handwriting features for more accurate scoring, the structural similarity index algorithm is used to analyze the answer sheet, and combined with a convolutional neural network to classify the questions, and the AR technology is used to feedback the marking situation, making the standardized scoring process more in line with the teaching requirements and improving the accuracy of manual marking.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a manual marking and correction method and system based on an image grayscale calculation method. Background Art

[0002] The field of image processing is a subfield of computer science that is dedicated to developing algorithms and techniques for acquiring, analyzing, modifying, and understanding images. This field explores how computers can be used to process and improve the quality of images in order to extract useful information from them. Image processing techniques have a wide range of applications in a variety of fields, including medical imaging, computer vision, security surveillance, image editing, and more.

[0003] The manual marking and correction method based on the image grayscale calculation method is a technology that combines image processing and grayscale calculation methods to assist the manual marking and correction process. It aims to improve the efficiency, accuracy and consistency of marking and correction, and is applicable to the scoring of student answer sheets and assignments. The purpose of the method is to improve the efficiency of marking and correction, reduce the time and labor required for manual marking, and improve the accuracy of scoring. This helps educational institutions to evaluate students' answers more effectively, reduce the burden on markers, and ensure consistency in scoring. The effects include improved efficiency, accuracy and consistency. Through image processing and grayscale calculation technology, the ability of automatic marking and correction is achieved, thereby reducing subjectivity, improving accuracy, and reducing the time and workload of marking. It usually includes steps such as image acquisition, preprocessing, grayscale calculation, marking and correction, and result generation to achieve these goals.

[0004] The main shortcoming of the existing manual marking method is the lack of assistance from scientific and technological means. The marking process relies entirely on manual work, which is inefficient and prone to errors. Without the use of technical means to process a large number of test papers, it is difficult to cope with large-scale answer review, making the marking work time-consuming and labor-intensive. Secondly, due to the lack of effective text area positioning and image optimization technology, it is impossible to avoid omissions or misjudgments. Thirdly, due to the lack of advanced technologies such as machine learning for handwriting feature analysis and question recognition, different scoring standards may occur, and even misjudgments may exist. Finally, there is a lack of an effective feedback system, a closed loop cannot be formed, and adjustments and improvements cannot be made at any time during the review process. These are all problems that the existing marking methods may cause in terms of scoring standards, accuracy, and efficiency. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a manual marking and correction method and system based on an image grayscale calculation method.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: a manual marking and correction method based on image grayscale calculation method, comprising the following steps:

[0007] S1: Use laser scanning technology to scan the student's answer sheet into a digital image to generate a digital image of the student's answer sheet;

[0008] S2: Based on the digital image of the student answer sheet, the Sobel edge detection algorithm and morphological processing are used to perform text area segmentation and extraction to generate a text image with grayscale features;

[0009] S3: Based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image to generate an optimized image;

[0010] S4: Based on the optimized image, using K-means clustering algorithm, analyzing the handwriting characteristics of the student to generate handwriting characteristic information;

[0011] S5: Based on the handwriting feature information, a structural similarity index algorithm is used to compare the grayscale distribution patterns of the answer sheet and the standard answer sheet, and a comparison result and analysis report are generated;

[0012] S6: Based on the comparison results and analysis reports, a convolutional neural network is used to identify and classify answer sheet question types, and AR technology is used to provide visual feedback to examiners to generate answer sheet question type recognition results and scoring feedback.

[0013] As a further solution of the present invention, laser scanning technology is used to scan the student's answer sheet into a digital image. The steps of generating the digital image of the student's answer sheet are specifically as follows:

[0014] S101: Based on the initialization of the laser scanner, the answer sheet is placed flat and fixed, and information about the answer sheet being fixed in place is obtained;

[0015] S102: Based on the information that the answer sheet is fixed in place, start a laser scanner to perform linear scanning to generate a preliminary digitized answer sheet image;

[0016] S103: Based on the preliminary digitized answer sheet image, a color balance and calibration method is used to adjust the image quality to generate a color-balanced digital answer sheet image;

[0017] S104: Based on the color-balanced digital answer sheet image, perform image size and resolution standardization to generate a digital image of the student answer sheet.

[0018] As a further solution of the present invention, based on the digital image of the student answer sheet, the steps of segmenting and extracting text regions using the Sobel edge detection algorithm and morphological processing to generate a text image with grayscale features are specifically as follows:

[0019] S201: Based on the digital image of the student answer sheet, convert the image using an image grayscale method to generate a grayscale answer sheet image;

[0020] S202: Based on the grayscale answer sheet image, applying the Sobel edge detection algorithm to generate an edge intensity image;

[0021] S203: Based on the edge intensity image, using morphological operations to generate a text area image with enhanced contrast;

[0022] S204: Based on the text region image with enhanced contrast, a region growing method is performed to generate a text image with grayscale features.

[0023] As a further solution of the present invention, based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image, and the steps of generating the optimized image are specifically as follows:

[0024] S301: Based on the text image with grayscale features, a grayscale histogram is calculated to generate a grayscale histogram distribution;

[0025] S302: Based on the grayscale histogram distribution, applying Otsu's algorithm to determine the optimal threshold value, and obtaining the optimal binarization threshold value;

[0026] S303: Based on the optimal binarization threshold, binarize the image to generate a binarized text image;

[0027] S304: Based on the binary text image, enhance the image using a histogram equalization technique to generate an optimized image.

[0028] As a further solution of the present invention, based on the optimized image, the K-means clustering algorithm is used to analyze the handwriting characteristics of the students, and the steps of generating handwriting characteristic information are specifically as follows:

[0029] S401: Based on the optimized image, perform denoising and smoothing using image preprocessing technology to generate a clean and optimized image;

[0030] S402: Based on the cleaned and optimized image, a K-means clustering algorithm is used to perform pixel clustering to generate a handwriting feature clustering result;

[0031] S403: Based on the handwriting feature clustering result, using feature extraction technology to extract key handwriting information to generate handwriting feature information;

[0032] S404: Based on the handwriting feature information, a data sorting technology is used to unify the format to generate formatted handwriting feature information.

[0033] As a further solution of the present invention, based on the handwriting feature information, a structural similarity index algorithm is used to compare the grayscale distribution patterns of the answer sheet and the standard answer sheet, and the steps of generating the comparison results and analysis report are specifically as follows:

[0034] S501: Based on the formatted handwriting feature information, a standard answer sheet is retrieved using a database query technology to generate a standard answer sheet image;

[0035] S502: Based on the standard answer sheet image, a structural similarity index algorithm is used to perform grayscale mode comparison to generate comparison data;

[0036] S503: Based on the comparison data, using data analysis technology to identify key indicators and generate an analysis report;

[0037] S504: Based on the analysis report, visualization technology is used to display data and generate comparison results and analysis reports.

[0038] As a further solution of the present invention, based on the comparison results and the analysis report, a convolutional neural network is used to identify and classify the answer sheet question types, and AR technology is used to provide visual feedback to the examiner. The steps of generating the answer sheet question type recognition results and the scoring feedback are as follows:

[0039] S601: Based on the comparison result and analysis report, a convolutional neural network is used to identify question types and generate question type classification results;

[0040] S602: Based on the question type classification result, use deep learning technology to perform feature optimization to generate an optimized question type recognition result;

[0041] S603: Based on the optimized question type recognition result, using augmented reality technology to provide visual feedback and generate AR visual feedback;

[0042] S604: Based on the AR visual feedback, user interface design technology is used to integrate the interface and generate answer sheet question type recognition results and scoring feedback.

[0043] A manual paper marking and correction system based on an image grayscale calculation method, the manual paper marking and correction system based on an image grayscale calculation method is used to execute the above-mentioned manual paper marking and correction method based on an image grayscale calculation method, the system includes an answer sheet digitization module, a text area segmentation module, an image optimization module, a handwriting feature analysis module, an answer sheet comparison module, a question type recognition module, and a marking feedback module.

[0044] As a further solution of the present invention, the answer sheet digitization module digitizes the answer sheet based on laser scanning technology, and then uses image color balance and calibration methods to process the image quality, standardize the image size and resolution, and generate a digital image of the student answer sheet;

[0045] The text region segmentation module is based on the digital image of the student answer sheet, uses the image grayscale method to convert the color image into a grayscale image, uses the Sobel edge detection algorithm to depict the image edge information, uses morphological operations to enhance the contrast between the text and the background, and extracts the main text area through the region growing method to generate a text image with grayscale features;

[0046] The image optimization module calculates the grayscale distribution state of the image based on the text image with grayscale features using a grayscale histogram, determines the optimal segmentation threshold using Otsu's adaptive threshold method, performs binarization processing on the image, and obtains an optimized image;

[0047] The handwriting feature analysis module uses image preprocessing technology to clean and smooth the image based on the optimized image, uses K-means clustering algorithm to perform feature mining and clustering on the student's handwriting, and uses feature extraction technology to extract key handwriting information to obtain formatted handwriting feature information;

[0048] The answer sheet comparison module uses database query technology to retrieve the standard answer sheet based on the formatted handwriting feature information, compares the grayscale mode of the answer sheet with the standard answer sheet through the structural similarity index algorithm, uses data analysis technology to identify key indicators, and outputs comparison results and analysis reports;

[0049] The question type recognition module uses a convolutional neural network to recognize the question type of student answers based on the comparison results and analysis report, classifies the answer content, applies deep learning technology to optimize features, and generates optimized question type recognition results;

[0050] The grading feedback module uses augmented reality technology to provide visual feedback based on the optimized question type recognition results, present and feedback the results, and form answer sheet question type recognition results and grading feedback.

[0051] As a further solution of the present invention, the answer sheet digitization module includes an answer sheet scanning submodule, an image quality adjustment submodule, and an image standardization submodule;

[0052] The text region segmentation module includes an image grayscale submodule, an edge detection submodule, a morphological operation submodule, and a region growing submodule;

[0053] The image optimization module includes a grayscale histogram submodule, a threshold determination submodule, a binarization processing submodule, and an image enhancement submodule;

[0054] The handwriting feature analysis module includes an image preprocessing submodule, a pixel clustering submodule, a feature extraction submodule, and a data sorting submodule;

[0055] The answer sheet comparison module includes an answer sheet retrieval submodule, a grayscale comparison submodule, an indicator identification submodule, and a data display submodule;

[0056] The question type identification module includes a question type identification submodule and a feature optimization submodule;

[0057] The marking feedback module includes an AR feedback submodule and an interface integration submodule.

[0058] Compared with the prior art, the advantages and positive effects of the present invention are:

[0059] In the present invention, through the improved laser scanning technology and image processing algorithm, the digital processing of large quantities of test papers can be completed efficiently, thereby improving the efficiency of paper marking. The Sobel edge detection algorithm and morphological processing are used to accurately segment the text area, and the student's answer area can be accurately located to avoid omissions or misjudgments. By using Otsu's adaptive threshold method to optimize image clarity and K-means clustering algorithm to analyze handwriting features, more accurate scoring can be performed. The structural similarity index algorithm is used to analyze the answer sheet, and the convolutional neural network is combined to classify the questions. AR technology is used to feedback the marking situation, so that the standardized scoring process is more in line with teaching needs and the accuracy of manual scoring is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0061] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0062] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0063] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0064] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0065] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0066] Figure 7 This is a detailed flow chart of S6 of the present invention;

[0067] Figure 8 is a system flow chart of the present invention;

[0068] Fig. 9 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0070] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0071] Embodiment 1

[0072] See also Figure 1 The present invention provides a technical solution: a manual marking and correction method based on an image grayscale calculation method, comprising the following steps:

[0073] S1: Use laser scanning technology to scan the student's answer sheet into a digital image to generate a digital image of the student's answer sheet;

[0074] S2: Based on the digital image of the student answer sheet, the Sobel edge detection algorithm and morphological processing are used to segment and extract the text area to generate a text image with grayscale features;

[0075] S3: Based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image and generate an optimized image;

[0076] S4: Based on the optimized image, the K-means clustering algorithm is used to analyze the handwriting characteristics of the students and generate handwriting feature information;

[0077] S5: Based on the handwriting feature information, the structural similarity index algorithm is used to compare the grayscale distribution pattern of the answer sheet with the standard answer sheet, and the comparison results and analysis report are generated;

[0078] S6: Based on the comparison results and analysis reports, a convolutional neural network is used to identify and classify answer sheet question types, and AR technology is used to provide visual feedback to examiners to generate answer sheet question type recognition results and scoring feedback.

[0079] First, through the use of laser scanning technology, the answer sheet is quickly and accurately converted into a digital image, which greatly improves the efficiency of answer sheet processing. The generation of digital images provides a basis for subsequent automatic processing, realizing the transition from traditional manual marking to digital marking.

[0080] Secondly, through the Sobel edge detection algorithm and morphological processing, the text area is accurately segmented and extracted, which not only improves the readability of the text, but also lays the foundation for subsequent image processing. The Otsu's adaptive threshold method also performs well in optimizing image contrast and clarity, making the text information more eye-catching and helping to reduce scoring errors.

[0081] Furthermore, by analyzing the handwriting characteristics of students based on the K-means clustering algorithm, we can more accurately identify and classify the writing habits and characteristics of different students. This feature analysis not only helps to identify the differences between the answer sheets and the standard answer sheets, but also provides technical support for the personalized processing of the answer sheets.

[0082] The structural similarity index algorithm provides a new dimension for the comparison between the answer sheet and the standard answer sheet, namely the comparison of grayscale distribution patterns. This comparison method can more comprehensively reflect the correctness and completeness of the answer sheet content, bringing unprecedented depth and breadth to the marking work.

[0083] Finally, the application of convolutional neural networks makes the recognition and classification of answer sheet types more intelligent and efficient, greatly reducing the impact of human factors on the results during the marking process. The introduction of AR technology provides intuitive visual feedback for markers, greatly improving the accuracy and efficiency of marking.

[0084] See also Figure 2 , using laser scanning technology, scan the student's answer sheet into a digital image. The specific steps of generating the digital image of the student's answer sheet are:

[0085] S101: Based on the initialization of the laser scanner, the answer sheet is placed flat and fixed, and information about the answer sheet being fixed in place is obtained;

[0086] S102: Based on the information that the answer sheet is fixed in place, start the laser scanner to perform linear scanning to generate a preliminary digitized answer sheet image;

[0087] S103: Based on the preliminary digitized answer sheet image, the image quality is adjusted by using a color balance and calibration method to generate a color-balanced digital answer sheet image;

[0088] S104: Based on the color-balanced digital answer sheet image, perform image size and resolution standardization to generate a digital image of the student answer sheet.

[0089] In step S101, the answer sheet is fixed in place.

[0090] Obtaining information about the answer sheet being fixed in place requires the use of image processing algorithms, such as edge detection, to determine the bounding box of the answer sheet.

[0091] # Edge detection using Python and OpenCV

[0092] import cv2

[0093] image = cv2.imread('answer sheet image.jpg')

[0094] gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

[0095] edges = cv2.Canny(gray, threshold1, threshold2)

[0096] In step S102, a preliminary digitized answer sheet image is generated.

[0097] In step S103, image quality adjustment is performed

[0098] The specific algorithms and instructions for color balancing and calibrating an image will depend on the image processing library and software, but here are some of the operations involved:

[0099] Color Balance:

[0100] # Contrast and brightness adjustment using OpenCV

[0101] alpha = 1.5 # contrast enhancement factor

[0102] beta = 30 # Brightness enhancement factor

[0103] adjusted_image = cv2.convertScaleAbs(image, alpha=alpha, beta=beta)

[0104] Image Calibration:

[0105] Image calibration requires external reference points or a template image. Here is a code example:

[0106] # Find the reference point by template matching

[0107] template = cv2.imread('reference point template.jpg', 0)

[0108] result = cv2.matchTemplate(image, template, cv2.TM_CCOEFF_NORMED)

[0109] min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result)

[0110] #Perform perspective correction based on the reference point position

[0111] reference_points = [(x, y), (x + template_width, y), (x, y +template_height), (x + template_width, y + template_height)]

[0112] target_points = [(0, 0), (template_width, 0), (0, template_height),(template_width, template_height)]

[0113] matrix = cv2.getPerspectiveTransform(np.float32(reference_points),np.float32(target_points))

[0114] calibrated_image = cv2.warpPerspective(image, matrix, (template_width, template_height))

[0115] In step S104, image size and resolution normalization is performed

[0116] The specific algorithms and instructions for adjusting image size and resolution vary by application, but here are some examples:

[0117] Resize the image:

[0118] # Resize the image using OpenCV

[0119] resized_image = cv2.resize(image, (new_width, new_height))

[0120] Adjust resolution:

[0121] # Resolution adjustment using OpenCV

[0122] target_resolution = (new_width, new_height)

[0123] resized_image = cv2.resize(image, target_resolution, interpolation=cv2.INTER_LINEAR)

[0124] See also Figure 3 Based on the digital image of the student answer sheet, the Sobel edge detection algorithm and morphological processing are used to segment and extract the text area. The specific steps of generating a text image with grayscale features are as follows:

[0125] S201: Based on the digital image of the student answer sheet, convert the image using an image grayscale method to generate a grayscale answer sheet image;

[0126] S202: Based on the grayscale answer sheet image, applying the Sobel edge detection algorithm to generate an edge intensity image;

[0127] S203: Based on the edge intensity image, using morphological operations such as dilation and erosion to generate a text area image with enhanced contrast;

[0128] S204: Based on the text region image with enhanced contrast, a region growing method is performed to generate a text image with grayscale features.

[0129] In S201, grayscale conversion of the image is performed

[0130] In this step, the color image of the student answer sheet is converted into a grayscale image. This can be done using common image processing libraries such as OpenCV.

[0131] import cv2

[0132] # Read student answer sheet image

[0133] image = cv2.imread('student answer sheet image.jpg')

[0134] # Convert the image to grayscale

[0135] gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

[0136] In S202, Sobel edge detection is performed

[0137] In this step, the Sobel edge detection algorithm will be used to generate an edge intensity image.

[0138] import cv2

[0139] import numpy as np

[0140] # Calculate Sobel gradient

[0141] sobel_x = cv2.Sobel(gray_image, cv2.CV_64F, 1, 0, ksize=3)

[0142] sobel_y = cv2.Sobel(gray_image, cv2.CV_64F, 0, 1, ksize=3)

[0143] # Calculate edge strength image

[0144] edge_intensity = np.sqrt(sobel_x**2 + sobel_y**2)

[0145] In S203, morphological operations are performed

[0146] Use morphological operations such as dilation and erosion to enhance contrast and highlight text areas.

[0147] #Threshold edge intensity image

[0148] binary_image = cv2.threshold(edge_intensity, threshold_value, 255,cv2.THRESH_BINARY)

[0149] # Use dilation to enhance contrast

[0150] kernel = np.ones((5, 5), np.uint8)

[0151] dilated_image = cv2.dilate(binary_image, kernel, iterations=1)

[0152] # Further processing using corrosion operations

[0153] kernel = np.ones((3, 3), np.uint8)

[0154] text_region_image = cv2.erode(dilated_image, kernel, iterations=1)

[0155] In S204, the region growing method is performed

[0156] Perform region growing to generate a text image with grayscale features. This requires implementing the region growing algorithm according to specific requirements. The following is a simple example:

[0157] def region_growing(image, seed):

[0158] # Initialize the output image

[0159] output = np.zeros_like(image)

[0160] # Threshold

[0161] threshold = 50 # Adjust the threshold as needed

[0162] # Create a queue

[0163] queue = []

[0164] queue.append(seed)

[0165] # Region Growing

[0166] while len(queue)>0:

[0167] current_point = queue.pop(0)

[0168] # Check whether the current point meets the growth conditions

[0169] if image[current_point] <threshold:

[0170] output[current_point] = 255

[0171] # Add adjacent pixels to the queue

[0172] x, y = current_point

[0173] for i in range(-1, 2):

[0174] for j in range(-1, 2):

[0175] if 0<= x + i <image.shape[0] and 0<= y + j<image.shape[1]:

[0176] if output[x + i, y + j] == 0:

[0177] queue.append((x + i, y + j))

[0178] return output

[0179] # Select a seed point, for example (100, 100)

[0180] seed_point = (100, 100)

[0181] text_image = region_growing(text_region_image, seed_point)

[0182] See also Figure 4 Based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image. The steps to generate the optimized image are as follows:

[0183] S301: Calculating a grayscale histogram based on a text image with grayscale features to generate a grayscale histogram distribution;

[0184] S302: Based on the grayscale histogram distribution, the Otsu's algorithm is applied to determine the optimal threshold value to obtain the optimal binarization threshold value;

[0185] S303: Based on the optimal binarization threshold, binarize the image to generate a binarized text image;

[0186] S304: Based on the binary text image, the image is enhanced using a histogram equalization technique to generate an optimized image.

[0187] In S301, grayscale histogram calculation is performed

[0188] In this step, the grayscale histogram distribution of the text image will be calculated.

[0189] import cv2

[0190] import numpy as np

[0191] import matplotlib.pyplot as plt

[0192] # Read text image

[0193] text_image = cv2.imread('text image.jpg', cv2.IMREAD_GRAYSCALE)

[0194] # Calculate grayscale histogram

[0195] histogram = cv2.calcHist([text_image], [0], None,

[256] , [0, 256])

[0196] # Draw a histogram

[0197] plt.plot(histogram)

[0198] plt.title('Grayscale Histogram')

[0199] plt.xlabel('grayscale value')

[0200] plt.ylabel('Number of pixels')

[0201] plt.show()

[0202] In S302, Otsu's algorithm is executed to determine the optimal threshold

[0203] In this step, Otsu's algorithm is applied to determine the optimal binarization threshold.

[0204] # Use Otsu's algorithm to determine the optimal threshold

[0205] optimal_threshold = cv2.threshold(text_image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

[0206] print(f'Optimal binarization threshold: {optimal_threshold}')

[0207] In S303, binarization processing is performed

[0208] Based on the optimal binarization threshold, the text image will be binarized.

[0209] # Binarize text image

[0210] binary_text_image = cv2.threshold(text_image, optimal_threshold, 255,cv2.THRESH_BINARY)[1]

[0211] In S304, histogram equalization is performed

[0212] In this step, histogram equalization technique is used to enhance the contrast and sharpness of the image.

[0213] # Apply histogram equalization

[0214] equalized_image = cv2.equalizeHist(binary_text_image)

[0215] # Save the optimized image

[0216] cv2.imwrite('optimized image.jpg', equalized_image)

[0217] See also Figure 5 Based on the optimized image, the K-means clustering algorithm is used to analyze the handwriting characteristics of students. The steps to generate handwriting feature information are as follows:

[0218] S401: Based on the optimized image, image preprocessing technology is used to perform denoising and smoothing to generate a clean and optimized image;

[0219] S402: Based on the cleaned and optimized image, a K-means clustering algorithm is used to perform pixel clustering to generate a handwriting feature clustering result;

[0220] S403: Based on the handwriting feature clustering result, feature extraction technology is used to extract key handwriting information to generate handwriting feature information;

[0221] S404: Based on the handwriting feature information, data sorting technology is used to unify the format and generate formatted handwriting feature information.

[0222] In S401, image preprocessing is performed

[0223] In this step, the optimized image will be denoised and smoothed to produce a clean image. This helps reduce the impact of noise on subsequent analysis.

[0224] import cv2

[0225] # Read the optimized image

[0226] cleaned_image = cv2.imread('optimized image.jpg', cv2.IMREAD_GRAYSCALE)

[0227] # Perform image denoising (different denoising algorithms can be selected as needed)

[0228] cleaned_image = cv2.fastNlMeansDenoising(cleaned_image, None, h=10,templateWindowSize=7, searchWindowSize=21)

[0229] # Smooth the image to reduce noise

[0230] cleaned_image = cv2.GaussianBlur(cleaned_image, (5, 5), 0)

[0231] # Save the cleaned and optimized image

[0232] cv2.imwrite('cleaned optimized image.jpg', cleaned_image)

[0233] In S402, K-means clustering is performed

[0234] In this step, the K-means clustering algorithm will be used to cluster the pixels of the image to divide the image into different clusters, where each cluster represents a different handwriting feature.

[0235] import numpy as np

[0236] # Convert the image to a one-dimensional array

[0237] pixels = cleaned_image.reshape((-1, 1))

[0238] # Define the number of clusters for K-means clustering (select according to actual situation)

[0239] k = 5

[0240] # Use K-means algorithm for clustering

[0241] kmeans = KMeans(n_clusters=k)

[0242] kmeans.fit(pixels)

[0243] # Get the cluster labels and convert them back to image shape

[0244] cluster_labels = kmeans.labels_.reshape(cleaned_image.shape)

[0245] # Save the clustering results

[0246] cv2.imwrite('clustering result.jpg', cluster_labels)

[0247] In S403, feature extraction is performed

[0248] In this step, key handwriting information, such as the area and center coordinates of each cluster, will be extracted from the handwriting feature clustering results.

[0249] # Calculate the area and center coordinates of each cluster

[0250] cluster_features = []

[0251] for i in range(k):

[0252] mask = (cluster_labels == i)

[0253] area = np.sum(mask)

[0254] moments = cv2.moments(mask)

[0255] cx = int(moments["m10"] / moments["m00"])

[0256] cy = int(moments["m01"] / moments["m00"])

[0257] cluster_features.append({'Cluster': i, 'Area': area, 'Centroid': (cx,cy)})

[0258] # Output clustering feature information

[0259] for feature in cluster_features:

[0260] print(f'Cluster {feature["Cluster"]} - Area: {feature["Area"]},Centroid: {feature["Centroid"]}')

[0261] In S404, data sorting is performed

[0262] Finally, the handwriting feature information is formatted to generate formatted handwriting feature information.

[0263] import pandas as pd

[0264] # Create a data frame to store handwriting feature information

[0265] feature_df = pd.DataFrame(cluster_features)

[0266] # Save formatted handwriting feature information to a CSV file

[0267] feature_df.to_csv('Handwriting feature information.csv', index=False)

[0268] See also Figure 6 Based on the handwriting feature information, the structural similarity index algorithm is used to compare the grayscale distribution pattern of the answer sheet with the standard answer sheet. The specific steps for generating the comparison results and analysis report are as follows:

[0269] S501: Based on the formatted handwriting feature information, a standard answer sheet is retrieved using a database query technology to generate a standard answer sheet image;

[0270] S502: Based on the standard answer sheet image, a structural similarity index algorithm is used to perform grayscale pattern comparison to generate comparison data;

[0271] S503: Based on the comparison data, use data analysis technology to identify key indicators and generate an analysis report;

[0272] S504: Based on the analysis report, visualization technology is used to display data and generate comparison results and analysis reports.

[0273] In S501, standard answer sheet retrieval is performed

[0274] First, retrieve the standard answer sheet from the database and convert it into an image. Here, it is assumed that the file path of the answer sheet image is saved in the database.

[0275] import cv2

[0276] import pandas as pd

[0277] # Query standard answer sheet information from the database

[0278] standard_answer_path = "path / to / standard_answer.jpg" # Assume this is the file path of the standard answer sheet

[0279] # Read the standard answer sheet image

[0280] standard_answer_image = cv2.imread(standard_answer_path, cv2.IMREAD_GRAYSCALE)

[0281] In S502, a structural similarity index comparison is performed.

[0282] Next, the structural similarity index (SSIM) algorithm is used to compare the answer sheet with the standard answer sheet.

[0283] from skimage.metrics import structural_similarity as ssim

[0284] # Read student answer sheet image

[0285] student_answer_path = "path / to / student_answer.jpg" # Assume this is the file path of the student answer sheet

[0286] student_answer_image = cv2.imread(student_answer_path, cv2.IMREAD_GRAYSCALE)

[0287] # Calculate the SSIM indicator

[0288] ssim_score, _ = ssim(standard_answer_image, student_answer_image,full=True)

[0289] # Output comparison results

[0290] print(f"SSIM Score: {ssim_score}")

[0291] In S503, data analysis is performed

[0292] Key indicators can be identified based on SSIM scores or other comparison data to generate analysis reports. This step usually needs to be customized according to specific needs and domain knowledge.

[0293] # Make a simple judgment based on the SSIM score

[0294] if ssim_score>0.8:

[0295] analysis_report = "The similarity between the student answers and the standard answers is high, and the answer quality is good."

[0296] else:

[0297] analysis_report = "The similarity between the student's answer sheet and the standard answer sheet is low. It is recommended to further review the student's answers."

[0298] # Output analysis report

[0299] print(analysis_report)

[0300] In S504, visualization is performed

[0301] Finally, visualization techniques can be used to present the comparison results and analysis reports.

[0302] import matplotlib.pyplot as plt

[0303] # Display student answer sheet images and standard answer sheet images

[0304] plt.subplot(1, 2, 1)

[0305] plt.imshow(student_answer_image, cmap='gray')

[0306] plt.title('Student Answer Sheet')

[0307] plt.axis('off')

[0308] plt.subplot(1, 2, 2)

[0309] plt.imshow(standard_answer_image, cmap='gray')

[0310] plt.title('Standard answer sheet')

[0311] plt.axis('off')

[0312] plt.show()

[0313] # Print analysis report

[0314] print(analysis_report)

[0315] See also Figure 7 Based on the comparison results and analysis reports, a convolutional neural network is used to identify and classify the answer sheet types, and AR technology is used to provide visual feedback to the examiner. The specific steps for generating answer sheet type recognition results and scoring feedback are as follows:

[0316] S601: Based on the comparison results and analysis report, a convolutional neural network is used to identify question types and generate question type classification results;

[0317] S602: Based on the question type classification result, deep learning technology is used to perform feature optimization to generate an optimized question type recognition result;

[0318] S603: Based on the optimized question type recognition result, augmented reality technology is used for visual feedback to generate AR visual feedback;

[0319] S604: Based on AR visual feedback, user interface design technology is used for interface integration to generate answer sheet question type recognition results and scoring feedback.

[0320] In S601, question type recognition is performed using a convolutional neural network

[0321] In this step, we will use Convolutional Neural Network (CNN) to identify the answer sheet types. First, we need to prepare a dataset, including answer sheet images of different types of questions and corresponding labels.

[0322] # Import required libraries

[0323] import tensorflow as tf

[0324] from tensorflow.keras import layers, models

[0325] from sklearn.model_selection import train_test_split

[0326] # Prepare the data set, X is the answer sheet image, y is the question type label

[0327] X, y = prepare_dataset()

[0328] # Divide into training set and test set

[0329] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

[0330] # Create a convolutional neural network model

[0331] model = models.Sequential()

[0332] model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(image_height, image_width, 1)))

[0333] model.add(layers.MaxPooling2D((2, 2)))

[0334] model.add(layers.Flatten())

[0335] model.add(layers.Dense(128, activation='relu'))

[0336] model.add(layers.Dense(num_classes, activation='softmax'))

[0337] # Compile the model

[0338] model.compile(optimizer='adam', loss='categorical_crossentropy',metrics=['accuracy'])

[0339] # Train the model

[0340] model.fit(X_train, y_train, epochs=10, validation_data=(X_test, y_test))

[0341] # Evaluate model performance

[0342] test_loss, test_acc = model.evaluate(X_test, y_test)

[0343] print(f"Test accuracy: {test_acc}")

[0344] In S602, feature optimization is performed using deep learning technology

[0345] At this step, deep learning techniques such as autoencoders can be used to optimize the question type recognition results.

[0346] # Create an autoencoder model

[0347] autoencoder = models.Sequential()

[0348] autoencoder.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(image_height, image_width, 1)))

[0349] autoencoder.add(layers.MaxPooling2D((2, 2)))

[0350] autoencoder.add(layers.Conv2D(64, (3, 3), activation='relu'))

[0351] autoencoder.add(layers.MaxPooling2D((2, 2)))

[0352] autoencoder.add(layers.Conv2DTranspose(64, (3, 3), activation='relu'))

[0353] autoencoder.add(layers.UpSampling2D((2, 2)))

[0354] autoencoder.add(layers.Conv2DTranspose(32, (3, 3), activation='relu'))

[0355] autoencoder.add(layers.UpSampling2D((2, 2)))

[0356] autoencoder.add(layers.Conv2D(1, (3, 3), activation='sigmoid',padding='same'))

[0357] # Compile the autoencoder model

[0358] autoencoder.compile(optimizer='adam', loss='mean_squared_error')

[0359] # Train the autoencoder

[0360] autoencoder.fit(X_train, X_train, epochs=10, validation_data=(X_test,X_test))

[0361] # Extract features using the encoder part

[0362] encoder = models.Model(inputs=autoencoder.input, outputs=autoencoder.get_layer('max_pooling2d_2').output)

[0363] encoded_X_train = encoder.predict(X_train)

[0364] encoded_X_test = encoder.predict(X_test)

[0365] In S603, augmented reality technology and AR visual feedback are performed

[0366] At this step, augmented reality technology is used to provide visual feedback to the examiner, typically involving the use of an AR library such as ARKit or ARCore to overlay information onto the examiner's field of view.

[0367] # Using AR technology to create visual feedback

[0368] # Here you need to develop according to the specific AR platform and library, such as using Unity and Vuforia.

[0369] # Overlay the question type recognition results onto the examiner's field of view

[0370] display_ar_feedback(question_type)

[0371] S604: User interface design and integration

[0372] Finally, a user interface needs to be created to integrate the question type recognition results and AR visual feedback so that examiners can easily view and record the grading results.

[0373] # Create the user interface

[0374] # This can be a web application, mobile application or desktop application, depending on the specific needs.

[0375] # Integrate question type recognition results and AR feedback

[0376] display_user_interface(question_type, ar_feedback)

[0377] # Allow examiners to enter information such as scores and save the results

[0378] save_grading_result(grading_result)

[0379] See also Figure 8 , a manual marking and correction system based on image grayscale calculation method, the manual marking and correction system based on image grayscale calculation method is used to execute the above-mentioned manual marking and correction method based on image grayscale calculation method, the system includes answer sheet digitization module, text area segmentation module, image optimization module, handwriting feature analysis module, answer sheet comparison module, question type recognition module, marking feedback module.

[0380] The answer sheet digitization module uses laser scanning technology to digitize the answer sheet, and then uses image color balance and calibration methods to process the image quality, standardize the image size and resolution, and generate a digital image of the student answer sheet;

[0381] The text region segmentation module is based on the digital image of the student answer sheet. It uses the image grayscale method to convert the color image into a grayscale image, uses the Sobel edge detection algorithm to depict the image edge information, uses morphological operations to enhance the contrast between the text and the background, and extracts the main text area through the region growing method to generate a text image with grayscale features.

[0382] The image optimization module uses grayscale histogram to calculate the grayscale distribution of the image based on the text image with grayscale features, uses Otsu's adaptive threshold method to determine the optimal segmentation threshold, performs binarization on the image, and obtains the optimized image;

[0383] The handwriting feature analysis module uses image preprocessing technology to clean and smooth the image based on the optimized image, uses the K-means clustering algorithm to mine and cluster the student's handwriting features, and uses feature extraction technology to extract key handwriting information to obtain formatted handwriting feature information;

[0384] The answer sheet comparison module uses database query technology to retrieve the standard answer sheet based on the formatted handwriting feature information, compares the grayscale mode of the answer sheet with the standard answer sheet through the structural similarity index algorithm, uses data analysis technology to identify key indicators, and outputs comparison results and analysis reports;

[0385] Based on the comparison results and analysis reports, the question type recognition module uses convolutional neural networks to identify the question types of students' answers, classify the answer content, and use deep learning technology to optimize features to generate optimized question type recognition results;

[0386] The grading feedback module is based on the optimized question type recognition results and uses augmented reality technology to provide visual feedback, present results and provide feedback, forming answer sheet question type recognition results and grading feedback.

[0387] First of all, the use of the answer sheet digitization module has significantly improved the efficiency of the traditional answer sheet processing process. The digital processing of the answer sheet not only improves the convenience and operability of the data, but also the application of laser scanning technology greatly improves the clarity and accuracy of the digitized image of the answer sheet. The image color balance and calibration method ensures the accuracy of the image color, making subsequent processing more accurate and reliable.

[0388] Secondly, the text region segmentation module is based on the digital image of the student answer sheet. Through image processing techniques such as grayscale, edge detection, and morphological operations, it can accurately extract the text region and convert the color image into a grayscale image, making the image information more prominent and more readable, which provides convenience for subsequent processing operations.

[0389] For the image optimization module, image processing based on grayscale features can improve image clarity, contrast, etc. The grayscale histogram and Otsu's adaptive threshold method can dynamically perform binarization processing based on the characteristics of the image itself, making the details of the text more eye-catching and reducing the error rate and complexity of subsequent processing.

[0390] The application of the handwriting feature analysis module enables the system to process answer sheets in a more refined and personalized manner. The K-means clustering algorithm is used to mine and cluster the features of students' handwriting, making the marking more in-depth and comprehensive, thereby increasing the accuracy of marking.

[0391] The answer sheet comparison module can compare the grayscale mode of the answer sheet with that of the standard answer sheet. The application of the structural similarity index algorithm makes the comparison more accurate, which can provide more objective scoring basis for examiners, reduce human errors and improve the fairness of scores.

[0392] The question type recognition module uses convolutional neural networks and deep learning technology to identify and optimize student answer types, which not only improves the speed of question type recognition, but also improves the accuracy of recognition, providing an effective reference for subsequent answer evaluation.

[0393] Finally, the grading feedback module uses augmented reality technology to provide visual feedback, allowing graders to understand the grading results more intuitively. This three-dimensional and intuitive feedback method can improve the grading quality and help graders make better judgments and decisions during the grading process.

[0394] See also Fig. 9 ,The answer sheet digitization module includes an answer sheet scanning submodule, an image quality adjustment submodule, and an image standardization submodule;

[0395] The text region segmentation module includes image grayscale submodule, edge detection submodule, morphological operation submodule, and region growing submodule;

[0396] The image optimization module includes a grayscale histogram submodule, a threshold determination submodule, a binarization processing submodule, and an image enhancement submodule;

[0397] The handwriting feature analysis module includes an image preprocessing submodule, a pixel clustering submodule, a feature extraction submodule, and a data sorting submodule;

[0398] The answer sheet comparison module includes answer sheet retrieval submodule, grayscale comparison submodule, indicator identification submodule, and data display submodule;

[0399] The question type recognition module includes a question type recognition submodule and a feature optimization submodule;

[0400] The marking feedback module includes an AR feedback submodule and an interface integration submodule.

[0401] Answer sheet digitization module:

[0402] The answer sheet scanning submodule uses a scanner or camera to convert paper answer sheets into digital images to ensure high resolution and clarity. The image quality adjustment submodule adjusts the quality of the scanned images, including removing noise, adjusting brightness and contrast to ensure image clarity. The image standardization submodule standardizes the images, including cropping edges and rotation correction, making the answer sheet images easier to analyze in subsequent processing.

[0403] Text region segmentation module:

[0404] The image grayscale submodule converts the answer sheet image into a grayscale image to reduce color information and facilitate text region segmentation. The edge detection submodule uses edge detection algorithms (such as Sobel or Canny) to detect the edges of the text region for subsequent segmentation. The morphological operation submodule performs morphological operations such as corrosion and dilation to further process the edge image, eliminate noise, and improve text segmentation. The region growing submodule uses the region growing algorithm to combine adjacent pixels into text regions, thereby segmenting the text in the answer sheet.

[0405] Image Optimization Module:

[0406] The grayscale histogram submodule analyzes the grayscale histogram of the answer sheet image and determines the parameters for subsequent processing, such as thresholds, based on the histogram information. The threshold determination submodule determines the appropriate threshold to separate the image into text and background based on the grayscale histogram. The binarization processing submodule binarizes the image to better separate the text and background. The image enhancement submodule uses filtering, enhancement and other technologies to improve the clarity of the text and ensure that the text is clear and recognizable.

[0407] Handwriting feature analysis module:

[0408] The image preprocessing submodule performs appropriate preprocessing, such as denoising, smoothing, and enhancement, to prepare the image for subsequent analysis. The pixel clustering submodule uses a clustering algorithm to divide the handwriting in the answer sheet into different blocks for subsequent feature extraction. The feature extraction submodule extracts the features of each handwriting block, such as shape, size, curves, etc., to help recognize handwritten text. The data organization submodule organizes and structures the feature information for subsequent answer sheet comparison.

[0409] Answer sheet comparison module:

[0410] The answer sheet retrieval submodule retrieves the standard answer or reference answer that matches the answer sheet based on the question information. The grayscale comparison submodule compares the grayscale of the student answer sheet image with the standard answer to determine whether it is correct or not. The indicator identification submodule identifies and records the various indicators in the scoring criteria for subsequent scoring. The data display submodule displays the comparison results and scoring criteria to the examiners in an easy-to-read manner to facilitate their scoring decisions.

[0411] Question type identification module:

[0412] The question type recognition submodule uses convolutional neural networks or other appropriate machine learning algorithms to classify and identify the question types in each answer sheet. The feature optimization submodule uses deep learning techniques, such as autoencoders, to optimize question type recognition results and improve classification accuracy.

[0413] Grading feedback module:

[0414] The AR feedback submodule uses augmented reality technology to provide visual feedback to examiners, superimposing recognition results on the answer sheet image, allowing examiners to more easily browse the answer sheet content. The interface integration submodule creates a user interface that integrates question type recognition results, AR feedback, and scoring functions so that examiners can easily view the answer sheet content, enter scores, and save scoring results.

[0415] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A manual marking and correction method based on image grayscale calculation method, It is characterized in that The following steps are involved: Using laser scanning technology, scan the student's answer sheet into a digital image to generate a digital image of the student's answer sheet; Based on the digital image of the student answer sheet, the Sobel edge detection algorithm and morphological processing are used to perform text area segmentation and extraction to generate a text image with grayscale features; Based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image to generate an optimized image; Based on the optimized image, using K-means clustering algorithm, the handwriting characteristics of the student are analyzed to generate handwriting characteristic information; Based on the handwriting feature information, a structural similarity index algorithm is used to compare the grayscale distribution patterns of the answer sheet and the standard answer sheet, and a comparison result and analysis report are generated; Based on the comparison results and analysis reports, a convolutional neural network is used to identify and classify the answer sheet types, and AR technology is used to provide visual feedback to the examiners to generate answer sheet type recognition results and scoring feedback; Based on the optimized image, the K-means clustering algorithm is used to analyze the handwriting characteristics of the students. The steps of generating handwriting characteristic information are as follows: Based on the optimized image, using image preprocessing technology to perform denoising and smoothing to generate a clean and optimized image; Based on the cleaned and optimized image, a K-means clustering algorithm is used to perform pixel clustering to generate a handwriting feature clustering result; Based on the handwriting feature clustering result, using feature extraction technology to extract key handwriting information and generate handwriting feature information; Based on the handwriting feature information, a data sorting technology is used to unify the format to generate formatted handwriting feature information; Based on the handwriting feature information, the structural similarity index algorithm is used to compare the grayscale distribution patterns of the answer sheet and the standard answer sheet, and the steps of generating the comparison results and analysis report are as follows: Based on the formatted handwriting feature information, a standard answer sheet is retrieved using a database query technology to generate a standard answer sheet image; Based on the standard answer sheet image, a structural similarity index algorithm is used to perform grayscale pattern comparison to generate comparison data; Based on the comparison data, using data analysis technology to identify key indicators and generate an analysis report; Based on the analysis report, visualization technology is used to display data and generate comparison results and analysis reports; Based on the comparison results and analysis reports, a convolutional neural network is used to identify and classify the answer sheet types, and AR technology is used to provide visual feedback to the examiner. The specific steps for generating answer sheet type recognition results and scoring feedback are as follows: Based on the comparison results and analysis report, a convolutional neural network is used to identify question types and generate question type classification results; Based on the question type classification results, deep learning technology is used to perform feature optimization to generate optimized question type recognition results; Based on the optimized question type recognition result, augmented reality technology is used for visual feedback to generate AR visual feedback; Based on the AR visual feedback, user interface design technology is used to integrate the interface and generate answer sheet question type recognition results and grading feedback.

2. The manual marking and correction method based on the image grayscale calculation method according to claim 1, It is characterized in that Use laser scanning technology to scan the student's answer sheet into a digital image. The steps for generating a digital image of the student's answer sheet are as follows: Based on the initialization of the laser scanner, the answer sheet is placed flat and fixed to obtain the information that the answer sheet is fixed in place; Based on the information that the answer sheet is fixed in place, a laser scanner is started to perform linear scanning to generate a preliminary digitized answer sheet image; Based on the preliminary digitized answer sheet image, a color balance and calibration method is used to adjust the image quality to generate a color-balanced digital answer sheet image; Based on the color-balanced digital answer sheet image, image size and resolution standardization is performed to generate a digital image of the student answer sheet.

3. The manual marking and correction method based on the image grayscale calculation method according to claim 1, It is characterized in that Based on the digital image of the student answer sheet, the steps of segmenting and extracting text regions using the Sobel edge detection algorithm and morphological processing to generate a text image with grayscale features are as follows: Based on the digital image of the student answer sheet, convert the image using an image grayscale method to generate a grayscale answer sheet image; Based on the grayscale answer sheet image, applying the Sobel edge detection algorithm to generate an edge intensity image; Based on the edge intensity image, using morphological operations to generate a text region image with enhanced contrast; Based on the text region image with enhanced contrast, a region growing method is performed to generate a text image with grayscale features.

4. The manual marking and correction method based on the image grayscale calculation method according to claim 1, It is characterized in that Based on the text image with grayscale features, Otsu's adaptive threshold method is used to optimize the contrast and clarity of the image. The steps of generating the optimized image are as follows: Based on the text image with grayscale features, a grayscale histogram is calculated to generate a grayscale histogram distribution; Based on the grayscale histogram distribution, an optimal threshold is determined by applying Otsu's algorithm to obtain an optimal binarization threshold; Based on the optimal binarization threshold, binarize the image to generate a binarized text image; Based on the binarized text image, a histogram equalization technique is used to enhance the image to generate an optimized image.

5. A manual marking and correction system based on image grayscale calculation method, It is characterized in that The manual marking and correction system based on the image grayscale calculation method is used to execute the manual marking and correction method based on the image grayscale calculation method described in any one of claims 1 to 4, and the system includes an answer sheet digitization module, a text area segmentation module, an image optimization module, a handwriting feature analysis module, an answer sheet comparison module, a question type recognition module, and a marking feedback module.

6. The manual marking and correction system based on the image grayscale calculation method according to claim 5, It is characterized in that The answer sheet digitization module digitizes the answer sheet based on laser scanning technology, and then uses image color balance and calibration methods to process the image quality, standardize the image size and resolution, and generate a digital image of the student answer sheet; The text region segmentation module is based on the digital image of the student answer sheet, uses the image grayscale method to convert the color image into a grayscale image, uses the Sobel edge detection algorithm to depict the image edge information, uses morphological operations to enhance the contrast between the text and the background, and extracts the main text area through the region growing method to generate a text image with grayscale features; The image optimization module calculates the grayscale distribution state of the image based on the text image with grayscale features using a grayscale histogram, determines the optimal segmentation threshold using Otsu's adaptive threshold method, performs binarization processing on the image, and obtains an optimized image; The handwriting feature analysis module uses image preprocessing technology to clean and smooth the image based on the optimized image, uses K-means clustering algorithm to perform feature mining and clustering on the student's handwriting, and uses feature extraction technology to extract key handwriting information to obtain formatted handwriting feature information; The answer sheet comparison module uses database query technology to retrieve the standard answer sheet based on the formatted handwriting feature information, compares the grayscale mode of the answer sheet with the standard answer sheet through the structural similarity index algorithm, uses data analysis technology to identify key indicators, and outputs comparison results and analysis reports; The question type recognition module uses a convolutional neural network to recognize the question type of student answers based on the comparison results and analysis report, classifies the answer content, applies deep learning technology to optimize features, and generates optimized question type recognition results; The grading feedback module uses augmented reality technology to provide visual feedback based on the optimized question type recognition results, present and feedback the results, and form answer sheet question type recognition results and grading feedback.

7. The manual marking and correction system based on the image grayscale calculation method according to claim 5, It is characterized in that The answer sheet digitization module includes an answer sheet scanning submodule, an image quality adjustment submodule, and an image standardization submodule; The text region segmentation module includes an image grayscale submodule, an edge detection submodule, a morphological operation submodule, and a region growing submodule; The image optimization module includes a grayscale histogram submodule, a threshold determination submodule, a binarization processing submodule, and an image enhancement submodule; The handwriting feature analysis module includes an image preprocessing submodule, a pixel clustering submodule, a feature extraction submodule, and a data sorting submodule; The answer sheet comparison module includes an answer sheet retrieval submodule, a grayscale comparison submodule, an indicator identification submodule, and a data display submodule; The question type identification module includes a question type identification submodule and a feature optimization submodule; The marking feedback module includes an AR feedback submodule and an interface integration submodule.

Images