CN117593324A - Manual evaluation error investigation method and system based on image gray level calculation method - Google Patents

Abstract

The invention relates to the technical field of image processing, in particular to an artificial comment error method and system based on an image gray level calculation method, comprising the following steps: and scanning the student answer sheet into a digital image by using a laser scanning technology, and generating the digital image of the student answer sheet. According to the invention, the digital processing of a large number of test papers can be efficiently completed through an improved laser scanning technology and an image processing algorithm, a text region is accurately segmented through a Sobel edge detection algorithm and morphological processing, a student answer region can be accurately positioned, the condition of missed observation or misjudgment is avoided, the image definition is optimized through an Otsu's self-adaptive threshold method, handwriting characteristics are analyzed through a K-means clustering algorithm, more accurate scoring is performed, the answer rolls are analyzed through a structural similarity index algorithm, the questions are classified through a convolutional neural network, and the condition of evaluating the papers is fed back through an AR technology, so that the standardized scoring process is more in line with teaching requirements, and the accuracy of manual scoring is improved.

Description

Manual evaluation error investigation method and system based on image gray level calculation method

Technical Field

The invention relates to the technical field of image processing, in particular to an artificial evaluation error method and system based on an image gray level calculation method.

Background

The image processing field is a sub-field of computer science, devoting to developing algorithms and techniques for acquiring, analyzing, modifying and understanding images. This field explores how to use a computer to process and improve the quality of images in order to extract useful information therefrom. The image processing technology is widely applied to various fields such as medical imaging, computer vision, security monitoring, image editing and the like.

The manual paper marking error investigation method based on the image gray level calculation method is a technology combining the image processing and gray level calculation method and is used for assisting the manual paper marking and error investigation process. The method aims at improving the efficiency, accuracy and consistency of the examination paper and the investigation, and is suitable for scoring of the student examination papers and the homework. The method aims at improving the efficiency of the evaluation and the investigation, reducing the time and labor required by manual evaluation and improving the accuracy of the evaluation. This helps the educational institution evaluate the student's answers more effectively, relieves the cricket's burden, and ensures consistency of scoring. Its effects include improved efficiency, accuracy and consistency. Through image processing and gray level calculation technology, the automatic scoring and error investigation capabilities are realized, so that subjectivity is reduced, accuracy is improved, and time and workload of scoring are reduced. Typically including image acquisition, preprocessing, gray level calculation, scoring and error investigation, and result generation, to achieve these goals.

In the existing manual paper-making method, the main defects are that the auxiliary of technological means is lacking, the paper-making process is completely dependent on manual work, the efficiency is low and errors are easy to occur. The test paper is not processed in a large quantity by adopting a technical means, and large-scale answer sheet review is difficult to deal with, so that the time and effort are consumed in the paper review operation. Secondly, due to the lack of effective text region positioning and image optimization technology, missed view or misjudgment cannot be avoided. Again, since advanced techniques such as machine learning are not used for handwriting feature analysis and topic recognition, there may be different scoring criteria and even erroneous judgment. Finally, the lack of an effective feedback system does not allow for a closed loop to be formed and can not be readily adjusted and modified during review. These are all problems that existing review methods may lead to scoring criteria, accuracy, and efficiency.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides a manual evaluation error method and system based on an image gray level calculation method.

In order to achieve the above purpose, the present invention adopts the following technical scheme: an artificial evaluation error investigation method based on an image gray level calculation method comprises the following steps:

S1: scanning the student answer sheet into a digital image by using a laser scanning technology, and generating a digital image of the student answer sheet;

s2: based on the digital image of the student answer sheet, adopting a Sobel edge detection algorithm and morphological processing to segment and extract a text region, and generating a text image with gray features;

s3: based on the text image with gray features, performing contrast and definition optimization on the image by adopting an Otsu's self-adaptive threshold method, and generating an optimized image;

s4: based on the optimized image, performing handwriting feature analysis of student writing by using a K-means clustering algorithm to generate handwriting feature information;

s5: based on the handwriting characteristic information, comparing the gray distribution mode of the answer sheet with that of a standard answer sheet by adopting a structural similarity index algorithm to generate a comparison result and an analysis report;

s6: based on the comparison result and the analysis report, a convolutional neural network is adopted to identify and classify the answer questions based on the comparison result and the corresponding analysis report, and an AR technology is utilized to provide visual feedback for a commentator to generate an answer question identification result and a commentator feedback.

As a further scheme of the invention, the laser scanning technology is used for scanning the student answer sheet into a digital image, and the step of generating the digital image of the student answer sheet comprises the following steps:

s101: based on the initialization of a laser scanner, placing the answer sheet flat and fixing, and obtaining the fixed-position information of the answer sheet;

s102: starting a laser scanner to perform linear scanning based on the answer sheet fixing in-place information, and generating a preliminary digital answer sheet image;

s103: based on the preliminary digital answer sheet image, adopting a color balance and calibration method to adjust the image quality, and generating a color balance digital answer sheet image;

s104: and based on the color balanced digital answer sheet image, performing image size and resolution standardization to generate a digital image of the student answer sheet.

As a further scheme of the invention, based on the digital image of the student answer sheet, a Sobel edge detection algorithm and morphological processing are adopted to segment and extract text areas, and the steps of generating the text image with gray features are specifically as follows:

s201: converting the image by using an image graying method based on the digital image of the student answer sheet to generate a gray answer sheet image;

s202: based on the gray-scale answer sheet image, applying a Sobel edge detection algorithm to generate an edge intensity image;

S203: generating a text region image of enhanced contrast using morphological operations based on the edge intensity image;

s204: and executing a region growing method based on the text region image with enhanced contrast, and generating a text image with gray scale characteristics.

As a further scheme of the invention, on the basis of the text image with gray features, an Otsu's self-adaptive threshold method is adopted to optimize the contrast and definition of the image, and the steps for generating the optimized image are specifically as follows:

s301: calculating a gray histogram based on the text image with gray features, and generating gray histogram distribution;

s302: based on the gray histogram distribution, determining an optimal threshold value by applying an Otsu's algorithm, and obtaining an optimal binarization threshold value;

s303: based on the optimal binarization threshold value, performing binarization processing on the image to generate a binarized text image;

s304: and based on the binarized text image, enhancing the image by using a histogram equalization technology, and generating an optimized image.

As a further scheme of the invention, based on the optimized image, the handwriting characteristic analysis of student writing is performed by utilizing a K-means clustering algorithm, and the step of generating handwriting characteristic information comprises the following steps:

S401: denoising and smoothing by adopting an image preprocessing technology based on the optimized image to generate a clean and optimized image;

s402: based on the clean and optimized image, performing pixel clustering by adopting a K-means clustering algorithm to generate a handwriting feature clustering result;

s403: based on the handwriting feature clustering result, extracting key handwriting information by adopting a feature extraction technology to generate handwriting feature information;

s404: based on the handwriting characteristic information, the format is unified by adopting a data sorting technology, and formatted handwriting characteristic information is generated.

As a further scheme of the invention, based on the handwriting characteristic information, a structural similarity index algorithm is adopted to compare the gray distribution mode of the answer sheet and the standard answer sheet, and the steps of generating a comparison result and an analysis report are specifically as follows:

s501: based on the formatted handwriting characteristic information, carrying out standard answer sheet retrieval by utilizing a database query technology, and generating a standard answer sheet image;

s502: based on the standard answer sheet image, adopting a structural similarity index algorithm to perform gray pattern comparison to generate comparison data;

s503: based on the comparison data, performing key index identification by adopting a data analysis technology, and generating an analysis report;

S504: based on the analysis report, a visualization technology is adopted for data display, and a comparison result and an analysis report are generated.

As a further scheme of the invention, based on the comparison result and the analysis report, a convolutional neural network is adopted to identify and classify the answer questions based on the comparison result and the corresponding analysis report, and an AR technology is utilized to provide visual feedback for a commentator, and the steps of generating the answer question identification result and the answer feedback are specifically as follows:

s601: based on the comparison result and the analysis report, performing topic identification by adopting a convolutional neural network to generate a topic classification result;

s602: based on the topic classification result, performing feature optimization by adopting a deep learning technology, and generating an optimized topic recognition result;

s603: based on the optimized topic identification result, performing visual feedback by adopting an augmented reality technology to generate AR visual feedback;

s604: based on the AR visual feedback, interface integration is performed by adopting a user interface design technology, and answer sheet recognition results and comment feedback are generated.

The manual examination paper investigation system based on the image gray level calculation method is used for executing the manual examination paper investigation method based on the image gray level calculation method, and comprises an examination paper digitizing module, a text region segmentation module, an image optimization module, a handwriting feature analysis module, an examination paper comparison module, a question recognition module and an examination paper feedback module.

As a further scheme of the invention, the answer sheet digitizing module digitizes the answer sheet based on a laser scanning technology, then performs image quality processing by using an image color balance and calibration method, performs image size and resolution standardization, and generates a digital image of the student answer sheet;

the text region segmentation module converts a color image into a gray image by using an image graying method based on a digital image of a student answer sheet, draws image edge information by using a Sobel edge detection algorithm, enhances the contrast between a text and a background by using morphological operation, extracts a main text region by using a region growing method, and generates a text image with gray characteristics;

the image optimization module calculates the gray distribution state of the image by adopting a gray histogram on the basis of the text image with gray characteristics, determines an optimal segmentation threshold by using an Otsu's self-adaptive threshold method, and carries out binarization processing on the image to obtain an optimized image;

the handwriting feature analysis module is used for carrying out image cleaning and smoothing by applying an image preprocessing technology based on the optimized image, carrying out feature mining and clustering on the student writing handwriting by adopting a K-means clustering algorithm, and extracting key handwriting information by utilizing a feature extraction technology to obtain formatted handwriting feature information;

The answer sheet comparison module is used for searching out standard answer sheets based on the formatted handwriting characteristic information by adopting a database query technology, comparing the gray patterns of the answer sheets and the standard answer sheets through a structural similarity index algorithm, and carrying out key index recognition by adopting a data analysis technology to produce comparison results and analysis reports;

the problem type recognition module performs student answer problem type recognition by using a convolutional neural network based on a comparison result and an analysis report, classifies answer content, performs feature optimization by using a deep learning technology, and generates an optimized problem type recognition result;

and the evaluation feedback module is used for providing visual feedback based on the optimized question type recognition result by using an augmented reality technology, and performing result presentation and feedback to form an answer sheet question type recognition result and an evaluation feedback.

As a further scheme of the invention, the answer sheet digitizing module comprises an answer sheet scanning sub-module, an image quality adjusting sub-module and an image normalizing sub-module;

the text region segmentation module comprises an image graying sub-module, an edge detection sub-module, a morphological operation sub-module and a region growing sub-module;

the image optimization module comprises a gray level histogram sub-module, a threshold value determination sub-module, a binarization processing sub-module and an image enhancement sub-module;

The handwriting feature analysis module comprises an image preprocessing sub-module, a pixel clustering sub-module, a feature extraction sub-module and a data arrangement sub-module;

the answer sheet comparison module comprises an answer sheet retrieval sub-module, a gray comparison sub-module, an index identification sub-module and a data display sub-module;

the question type recognition module comprises a question type recognition sub-module and a characteristic optimization sub-module;

the evaluation feedback module comprises an AR feedback sub-module and an interface integration sub-module.

Compared with the prior art, the invention has the advantages and positive effects that:

according to the invention, the digital processing of a large number of test papers can be efficiently completed by the improved laser scanning technology and image processing algorithm, so that the working efficiency of the paper evaluation is improved. The text region is accurately segmented by adopting a Sobel edge detection algorithm and morphological processing, so that the answer region of the student can be accurately positioned, and the condition of missed watching or misjudgment is avoided. The image definition is optimized by adopting an Otsu's self-adaptive threshold method, and handwriting characteristics are analyzed by a K-means clustering algorithm to carry out more accurate scoring. The answer rolls are analyzed by adopting a structural similarity index algorithm, the questions are classified by combining a convolutional neural network, and the condition of the evaluation roll is fed back by using an AR technology, so that the standardized scoring process meets teaching requirements, and the accuracy of manual scoring is improved.

Drawings

FIG. 1 is a schematic workflow diagram of the present invention;

FIG. 2 is a S1 refinement flowchart of the present invention;

FIG. 3 is a S2 refinement flowchart of the present invention;

FIG. 4 is a S3 refinement flowchart of the present invention;

FIG. 5 is a S4 refinement flowchart of the present invention;

FIG. 6 is a S5 refinement flowchart of the present invention;

FIG. 7 is a S6 refinement flowchart of the present invention;

FIG. 8 is a system flow diagram of the present invention;

FIG. 9 is a schematic diagram of a system framework of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Example 1

Referring to fig. 1, the present invention provides a technical solution: an artificial evaluation error investigation method based on an image gray level calculation method comprises the following steps:

s1: scanning the student answer sheet into a digital image by using a laser scanning technology, and generating a digital image of the student answer sheet;

s2: based on the digital image of the student answer sheet, adopting a Sobel edge detection algorithm and morphological processing to segment and extract a text region, and generating a text image with gray features;

s3: based on a text image with gray features, performing contrast and definition optimization on the image by adopting an Otsu's self-adaptive threshold method, and generating an optimized image;

s4: based on the optimized image, performing handwriting feature analysis of student writing by using a K-means clustering algorithm to generate handwriting feature information;

s5: based on handwriting characteristic information, comparing the gray distribution mode of the answer sheet with that of a standard answer sheet by adopting a structural similarity index algorithm to generate a comparison result and an analysis report;

s6: based on the comparison result and the analysis report, a convolutional neural network is adopted to identify and classify the answer questions based on the comparison result and the corresponding analysis report, and an AR technology is utilized to provide visual feedback for a commentator to generate an answer question identification result and a commentator feedback.

Firstly, through the use of a laser scanning technology, the answer sheet is rapidly and accurately converted into a digital image, and the answer sheet processing efficiency is greatly improved. The generation of the digital image provides a basis for subsequent automatic processing, and the transition from the traditional manual paper reading to the digital paper reading is realized.

Secondly, through a Sobel edge detection algorithm and morphological processing, the text region is accurately segmented and extracted, so that the readability of the text is improved, and a foundation is laid for subsequent image processing. The Otsu's self-adaptive threshold method is excellent in optimizing the contrast and definition of the image, so that the text information is more striking, and the reduction of the comment error is facilitated.

Furthermore, the handwriting characteristics written by the students are analyzed based on the K-means clustering algorithm, so that the writing habits and characteristics of different students can be more accurately identified and classified. The feature analysis is not only helpful for identifying the difference between the standard answer sheet, but also provides technical support for personalized treatment of the answer sheet.

The structural similarity index algorithm provides a new dimension for comparison between the answer sheet and the standard answer sheet, namely, comparison of gray distribution modes. The comparison method can more comprehensively reflect the correctness and the completeness of the answer sheet content, and brings unprecedented depth and breadth to the evaluation work.

Finally, the application of the convolutional neural network enables the identification and classification of the answer sheet questions to be more intelligent and efficient, and greatly reduces the influence of human factors on results in the process of evaluating the answer sheet. And visual feedback is provided for a commentator by introducing the AR technology, so that the accuracy and efficiency of commentator are greatly improved.

Referring to fig. 2, using a laser scanning technology, a digital image of a student answer sheet is scanned, and the steps for generating the digital image of the student answer sheet are specifically as follows:

s101: based on the initialization of a laser scanner, placing the answer sheet flat and fixing, and obtaining the fixed-position information of the answer sheet;

s102: starting a laser scanner to perform linear scanning based on the answer sheet fixing in-place information, and generating a preliminary digital answer sheet image;

s103: based on the preliminary digital answer sheet image, adopting a color balance and calibration method to adjust the image quality, and generating a color balance digital answer sheet image;

s104: based on the color balanced digital answer sheet image, image size and resolution standardization is executed, and a digital image of the student answer sheet is generated.

In step S101, the answer sheet is fixed in place

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

Edge detection using Python and OpenCV

import cv2

image=cv2.image ('answer sheet image. Jpg')

gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

edges = cv2.Canny(gray, threshold1, threshold2)

In step S102, preliminary digitized answer sheet image generation is performed

In step S103, image quality adjustment is performed

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

color balance:

contrast and brightness adjustment using OpenCV

alpha=1.5# contrast enhancement factor

beta=30# luminance enhancement factor

adjusted_image = cv2.convertScaleAbs(image, alpha=alpha, beta=beta)

Image calibration:

image calibration requires external reference points or template images. The following are code examples:

finding a reference point by template matching #

template = cv2.Imread ('reference point template. Jpg', 0)

result = cv2.matchTemplate(image, template, cv2.TM_CCOEFF_NORMED)

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

Perspective correction according to reference point

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

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

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

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

In step S104, image size and resolution normalization is performed

The specific algorithms and instructions for adjusting the image size and resolution will vary from application to application, with the following examples:

adjusting the image size:

image resizing using OpenCV

resized_image = cv2.resize(image, (new_width, new_height))

Adjusting the resolution:

resolution adjustment using OpenCV

target_resolution = (new_width, new_height)

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

Referring to fig. 3, based on a digital image of a student answer sheet, text region segmentation and extraction are performed by adopting a Sobel edge detection algorithm and morphological processing, and the steps of generating a text image with gray features are specifically as follows:

S201: converting the image by using an image graying method based on the digital image of the student answer sheet to generate a gray answer sheet image;

s202: based on the gray-scale answer sheet image, applying a Sobel edge detection algorithm to generate an edge intensity image;

s203: generating a contrast-enhanced text region image using morphological operations such as dilation and erosion based on the edge intensity image;

s204: based on the contrast-enhanced text region image, a region growing method is performed to generate a text image having gray features.

In S201, image graying is performed

At this step, the color image of the student answer sheet is converted into a gray image. This may be done through a common library of image processing (such as OpenCV).

import cv2

# reading student answer sheet image

image=cv2.image ('student answer sheet image. Jpg')

# convert image to gray scale

gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

In S202, sobel edge detection is performed

At this step, a Sobel edge detection algorithm will be used to generate the edge intensity image.

import cv2

import numpy as np

Calculation of Sobel gradient

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

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

Calculation of edge intensity image #

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

In S203, morphological operations are performed

Morphological operations such as dilation and erosion are used to enhance contrast, highlighting text regions.

Thresholded edge intensity image

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

Enhancing contrast using dilation operation #

kernel = np.ones((5, 5), np.uint8)

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

# further treated with etching operation

kernel = np.ones((3, 3), np.uint8)

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

In S204, a region growing method is performed

A region growing method is performed to generate a text image having gray scale features. This requires implementation of a region growing algorithm according to specific requirements, the following being a simple example:

def region_growing(image, seed):

# initializing output image

output = np.zeros_like(image)

Threshold #

threshold=50# adjust threshold as needed

# Create queue

queue = []

queue.append(seed)

Region growth #

while len(queue)>0:

current_point = queue.pop(0)

Check if the current point satisfies the growth condition

if image[current_point]<threshold:

output[current_point] = 255

# add neighboring pixels to queue

x, y = current_point

for i in range(-1, 2):

for j in range(-1, 2):

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

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

queue.append((x + i, y + j))

return output

Select a seed point, e.g. (100 )

seed_point = (100, 100)

text_image = region_growing(text_region_image, seed_point)

Referring to fig. 4, based on a text image with gray features, the contrast and sharpness of the image are optimized by using Otsu's adaptive thresholding, and the steps for generating the optimized image are specifically as follows:

s301: calculating a gray histogram based on the text image with gray features, and generating gray histogram distribution;

s302: based on gray level histogram distribution, determining an optimal threshold value by applying Otsu's algorithm, and obtaining an optimal binarization threshold value;

s303: based on the optimal binarization threshold value, performing binarization processing on the image to generate a binarized text image;

s304: based on the binarized text image, the image is enhanced using a histogram equalization technique, generating an optimized image.

In S301, a calculation of a gray level histogram is performed

At this step, the gray level histogram distribution of the text image will be calculated.

import cv2

import numpy as np

import matplotlib.pyplot as plt

# reading text image

text_image=cv2.imread ('text image. Jpg', cv2.imread_ GRAYSCALE)

# calculation gray level histogram

histogram = cv2.calcHist([text_image], [0], None, [256], [0, 256])

Drawing histogram #

plt.plot(histogram)

plt.title ('gray histogram')

plt.xlabel ('grey scale value')

plt.ylabel ('pixel count')

plt.show()

In S302, otsu' S algorithm is executed to determine the optimal threshold

At this step, the Otsu's algorithm will be applied to determine the optimal binarization threshold.

Determining optimal threshold using Otsu's algorithm

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

print (f' optimal binarization threshold: { optimal_threshold })

In S303, binarization processing is performed

Based on the optimal binarization threshold, binarization processing is performed on the text image.

# binarized text image

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

In S304, histogram equalization is performed

At this step, histogram equalization techniques are used to enhance the contrast and sharpness of the image.

Application of histogram equalization

equalized_image = cv2.equalizeHist(binary_text_image)

# save optimized image

Iv 2.Imwrite ('optimized image. Jpg', equalized_image)

Referring to fig. 5, based on the optimized image, the handwriting feature analysis of the student writing is performed by using a K-means clustering algorithm, and the step of generating handwriting feature information specifically includes:

S401: denoising and smoothing by adopting an image preprocessing technology based on the optimized image to generate a clean and optimized image;

s402: based on the clean and optimized image, adopting a K-means clustering algorithm to perform pixel clustering to generate a handwriting feature clustering result;

s403: based on the handwriting feature clustering result, extracting key handwriting information by adopting a feature extraction technology to generate handwriting feature information;

s404: based on the handwriting characteristic information, the format is unified by adopting a data sorting technology, and the formatted handwriting characteristic information is generated.

In S401, image preprocessing is performed

At this step, the optimized image will be denoised and smoothed to generate a clean image. This helps to reduce the effect of noise on subsequent analysis.

import cv2

# reading an optimized image

clear_image=cv2. Imread ('optimized image. Jpg', cv2.Imread_ GRAYSCALE)

# image denoising (different denoising algorithms can be selected according to requirements)

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

# smooth image to reduce noise

cleaned_image = cv2.GaussianBlur(cleaned_image, (5, 5), 0)

Preservation of clean optimized images

Iv 2.Imwrite ('clean optimized image. Jpg', clean_image)

In S402, K-means clustering is performed

At this step, the images will be clustered in pixels using a K-means clustering algorithm to separate the images into different clusters, where each cluster represents a different handwriting feature.

import numpy as np

# convert image into one-dimensional array

pixels = cleaned_image.reshape((-1, 1))

# definition of the number of clusters of K-means clusters (selected according to actual circumstances)

k = 5

Clustering using K-means algorithm

kmeans = KMeans(n_clusters=k)

kmeans.fit(pixels)

# acquire cluster labels and convert them back to image shape

cluster_labels = kmeans.labels_.reshape(cleaned_image.shape)

# save clustering results

Iv 2.Imwrite ('clustering result. Jpg', cluster_labes)

In S403, feature extraction is performed

At this step, key handwriting information, such as the area, center coordinates, etc., of each cluster will be extracted from the handwriting feature clustering results.

# calculate the area and center coordinates of each cluster

cluster_features = []

for i in range(k):

mask = (cluster_labels == i)

area = np.sum(mask)

moments = cv2.moments(mask)

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

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

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

# output Cluster characteristic information

for feature in cluster_features:

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

In S404, data sort is performed

And finally, unifying the formats of the handwriting characteristic information to generate formatted handwriting characteristic information.

import pandas as pd

Creating a data frame to store handwriting feature information

feature_df = pd.DataFrame(cluster_features)

# save formatted handwriting feature information to CSV file

feature_df.to_csv ('handwriting feature information. Csv', index=false)

Referring to fig. 6, based on handwriting feature information, a structural similarity index algorithm is adopted to compare a gray distribution mode of an answer sheet with a standard answer sheet, and the steps of generating a comparison result and an analysis report are specifically as follows:

s501: based on the formatted handwriting characteristic information, carrying out standard answer sheet retrieval by utilizing a database query technology, and generating a standard answer sheet image;

S502: based on the standard answer sheet image, adopting a structural similarity index algorithm to carry out gray pattern comparison and generating comparison data;

s503: based on the comparison data, performing key index identification by adopting a data analysis technology to generate an analysis report;

s504: based on the analysis report, a visualization technology is adopted for data display, and a comparison result and an analysis report are generated.

In S501, standard answer sheet retrieval is performed

First, standard answer sheets are retrieved from a database and converted into images. Here it is assumed that the database stores the file path of the answer sheet image.

import cv2

import pandas as pd

Query standard answer sheet information from database

Standard_answer_path= "path/to/standard_answer. Jpg" # assume that this is the file path of a standard answer sheet

Standard answer sheet image read

standard_answer_image = cv2.imread(standard_answer_path, cv2.IMREAD_GRAYSCALE)

In S502, structural similarity index comparison is performed

Next, a Structural Similarity Index (SSIM) algorithm is used to compare the answer sheet with a standard answer sheet.

from skimage.metrics import structural_similarity as ssim

# reading student answer sheet image

student_answer_path= "path/to/student_answer. Jpg" # assume that this is the file path of student answer

student_answer_image = cv2.imread(student_answer_path, cv2.IMREAD_GRAYSCALE)

# calculate SSIM index

ssim_score, _ = ssim(standard_answer_image, student_answer_image, full=True)

# output comparison result

print(f"SSIM Score: {ssim_score}")

In S503, data analysis is performed

The analysis report may be generated based on identification of key indicators to the SSIM score or other comparison data. This step typically requires customization based on specific needs and domain knowledge.

Simple determination according to SSIM score

if ssim_score>0.8:

analysis_report= "the similarity of student answer sheet and standard answer sheet is higher, and the answer quality is good. "

else:

analysis_report= "the similarity between student answer sheet and standard answer sheet is low, and further examination of student answer sheet condition is recommended. "

# output analysis report

print(analysis_report)

In S504, visualization is performed

Finally, the comparison results and analysis reports may be presented using visualization techniques.

import matplotlib.pyplot as plt

# display student answer sheet image and standard answer sheet image

plt.subplot(1, 2, 1)

plt.imshow(student_answer_image, cmap='gray')

plt.title ('student's answer sheet)

plt.axis('off')

plt.subplot(1, 2, 2)

plt.imshow(standard_answer_image, cmap='gray')

plt.title ('Standard answer sheet')

plt.axis('off')

plt.show()

# print analysis report

print(analysis_report)

Referring to fig. 7, based on the comparison result and the analysis report, the method adopts a convolutional neural network to identify and classify the answer sheet type based on the comparison result and the corresponding analysis report, and provides visual feedback for the commentator by using AR technology, and the steps of generating the answer sheet type identification result and the commentary feedback are specifically as follows:

s601: based on the comparison result and the analysis report, adopting a convolutional neural network to identify the problem type, and generating a problem type classification result;

s602: based on the topic classification result, performing feature optimization by adopting a deep learning technology, and generating an optimized topic recognition result;

s603: based on the optimized topic recognition result, performing visual feedback by adopting an augmented reality technology to generate AR visual feedback;

S604: based on AR visual feedback, interface integration is performed by adopting a user interface design technology, and answer sheet recognition results and comment feedback are generated.

In S601, performing topic recognition using convolutional neural network

At this step, a convolutional neural network (Convolutional Neural Network, CNN) will be used to identify the answer sheet. First, a dataset needs to be prepared, including answer images of different types of questions and corresponding labels.

# import required library

import tensorflow as tf

from tensorflow.keras import layers, models

from sklearn.model_selection import train_test_split

# prepare dataset, X is answer sheet image, y is question label

X, y = prepare_dataset()

# division training set and test set

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

# creation convolutional neural network model

model = models.Sequential()

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

model.add(layers.MaxPooling2D((2, 2)))

model.add(layers.Flatten())

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

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

# compiling model

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

Training model #

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

Model performance evaluation #

test_loss, test_acc = model.evaluate(X_test, y_test)

print(f"Test accuracy: {test_acc}")

In S602, feature optimization is performed using deep learning techniques

At this step, a deep learning technique, such as an automatic encoder (Autoencoder), may be used to optimize the topic recognition result.

# creation of automatic encoder model

autoencoder = models.Sequential()

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

autoencoder.add(layers.MaxPooling2D((2, 2)))

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

autoencoder.add(layers.MaxPooling2D((2, 2)))

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

autoencoder.add(layers.UpSampling2D((2, 2)))

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

autoencoder.add(layers.UpSampling2D((2, 2)))

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

# compiling automatic encoder model

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

# training automatic encoder

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

# use encoder section to extract features

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

encoded_X_train = encoder.predict(X_train)

encoded_X_test = encoder.predict(X_test)

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

At this step, the visual feedback will be provided to the reviewer using augmented reality techniques, which typically involve the use of an AR library, such as ARKit or ARCore, to superimpose the information into the reviewer's field of view.

Creating visual feedback using AR technology

# here needs to be developed according to specific AR platforms and libraries, e.g. using Unity and Vuforia et al.

# superimpose the question recognition result in the field of view of the panelist

display_ar_feedback(question_type)

S604: user interface design and integration

Finally, a user interface needs to be created to integrate the question recognition results with the AR visual feedback so that the reviewer can conveniently view and record the review results.

# Create user interface

This may be a Web application, mobile application or desktop application, depending on the specific requirements.

# integration question type recognition result and AR feedback

display_user_interface(question_type, ar_feedback)

# allow the panelist to enter information such as scoring and save the result of the panel

save_grading_result(grading_result)

Referring to fig. 8, an artificial evaluation error system based on an image gray level calculation method is used for executing the above artificial evaluation error method based on the image gray level calculation method, and the system includes an answer sheet digitizing module, a text region segmentation module, an image optimizing module, a handwriting feature analysis module, an answer sheet comparison module, a question recognition module, and an answer sheet feedback module.

The answer sheet digitizing module is used for digitizing the answer sheet based on the laser scanning technology, and then performing image quality processing by using an image color balance and calibration method, and performing image size and resolution standardization to generate a digital image of the student answer sheet;

The text region segmentation module converts a color image into a gray image by using an image graying method based on a digital image of a student answer sheet, draws image edge information by using a Sobel edge detection algorithm, enhances the contrast between a text and a background by using morphological operation, extracts a main text region by using a region growing method, and generates a text image with gray characteristics;

the image optimization module calculates the gray distribution state of the image by adopting a gray histogram on the basis of the text image with gray characteristics, determines an optimal segmentation threshold by adopting an Otsu's self-adaptive threshold method, and carries out binarization processing on the image to obtain an optimized image;

the handwriting feature analysis module performs image cleaning and smoothing by applying an image preprocessing technology based on the optimized image, performs feature mining and clustering on the student writing handwriting by adopting a K-means clustering algorithm, and performs key handwriting information extraction by utilizing a feature extraction technology to obtain formatted handwriting feature information;

the answer sheet comparison module is used for searching out standard answer sheets based on the formatted handwriting characteristic information by adopting a database query technology, comparing the gray patterns of the answer sheets and the standard answer sheets through a structural similarity index algorithm, and carrying out key index recognition by adopting a data analysis technology to produce comparison results and analysis reports;

The problem type recognition module performs student answer problem type recognition by using a convolutional neural network based on the comparison result and the analysis report, classifies answer content, performs feature optimization by using a deep learning technology, and generates an optimized problem type recognition result;

and the comment feedback module provides visual feedback based on the optimized question type recognition result by using an augmented reality technology, and performs result presentation and feedback to form an answer sheet question type recognition result and comment feedback.

Firstly, the use of the answer sheet digitization module obviously improves the efficiency of the traditional answer sheet processing flow, the digitization processing of the answer sheet not only improves the convenience and operability of data, but also greatly improves the definition and precision of the digitized image of the answer sheet by the application of the laser scanning technology, and the image color balance and calibration method ensures the accuracy of the image color, so that the subsequent processing is more accurate and reliable.

And secondly, the text region segmentation module can accurately extract the text region based on the digital image of the student answer sheet through image processing technologies such as graying, edge detection, morphological operation and the like, and convert the color image into a gray image, so that the image information is more prominent, the readability is stronger, and convenience is provided for subsequent processing operation.

Aiming at an image optimization module, image processing based on gray features can improve image definition, contrast and the like, and binarization processing can be dynamically carried out according to the characteristics of an image by adopting a gray histogram and Otsu's self-adaptive threshold method, so that details of a text are more striking, and the error rate and complexity of subsequent processing can be reduced.

By the application of the handwriting characteristic analysis module, the system can conduct answer sheet processing more finely and more personally, and the characteristics of the handwriting of the student are mined and clustered through the K-means clustering algorithm, so that the evaluation can be deeper and more comprehensive, and the evaluation precision is increased.

The answer sheet comparison module can compare the gray patterns of the answer sheet and the standard answer sheet, and the structural similarity index algorithm is applied to enable the comparison to be more accurate, so that a more objective scoring basis can be provided for the commentators, personal errors are reduced, and the score fairness is improved.

The problem type recognition module performs recognition and optimization of the student answer sheet problem type through a convolutional neural network and a deep learning technology, so that the problem type recognition speed is improved, the recognition accuracy is improved, and an effective reference is provided for subsequent answer sheet evaluation.

Finally, the evaluation feedback module provides visual feedback by using the augmented reality technology, so that an evaluation person can more intuitively know the evaluation result, the three-dimensional and intuitive feedback mode can improve the evaluation quality, and the evaluation person can be helped to better make judgment and decision in the evaluation process.

Referring to fig. 9, the answer sheet digitizing module includes an answer sheet scanning sub-module, an image quality adjusting sub-module, and an image normalizing sub-module;

the text region segmentation module comprises an image graying sub-module, an edge detection sub-module, a morphological operation sub-module and a region growing sub-module;

the image optimization module comprises a gray level histogram sub-module, a threshold value determination sub-module, a binarization processing sub-module and an image enhancement sub-module;

the handwriting feature analysis module comprises an image preprocessing sub-module, a pixel clustering sub-module, a feature extraction sub-module and a data arrangement sub-module;

the answer sheet comparison module comprises an answer sheet retrieval sub-module, a gray comparison sub-module, an index identification sub-module and a data display sub-module;

the question type recognition module comprises a question type recognition sub-module and a characteristic optimization sub-module;

the comment feedback module comprises an AR feedback sub-module and an interface integration sub-module.

And an answer sheet digitizing module:

the answer sheet scanning sub-module uses a scanner or camera equipment to convert the paper answer sheet into a digital image, so as to ensure high resolution and definition. The image quality adjustment sub-module performs quality adjustment on the scanned image, including removing noise, adjusting brightness and contrast, to ensure image sharpness. The image normalization sub-module normalizes the image, including cutting edges and rotation correction, so that the answer sheet image is easier to analyze in subsequent processing.

Text region segmentation module:

the image graying sub-module converts the answer sheet image into a gray image, reduces color information and is convenient for text region segmentation. The edge detection sub-module uses an edge detection algorithm (e.g., sobel or Canny) to detect edges of text regions for subsequent segmentation. The morphological operation submodule performs morphological operations such as corrosion and expansion to further process the edge image, eliminate noise and improve the text segmentation effect. The region growing sub-module combines adjacent pixels into a text region through a region growing algorithm, so that texts in the answer sheet are segmented.

And an image optimization module:

the gray level histogram submodule analyzes the gray level histogram of the answer sheet image, and determines parameters such as a threshold value for subsequent processing according to the histogram information. The threshold determination submodule determines an appropriate threshold to separate the image into text and background based on the gray level histogram. The binarization processing sub-module binarizes the image to better separate text and background. The image enhancer module uses filtering, enhancement and other technologies to improve the definition of the text and ensure that the text is clear and recognizable.

Handwriting feature analysis module:

the image preprocessing sub-module performs appropriate preprocessing such as denoising, smoothing and enhancement to prepare the image for subsequent analysis. The pixel clustering sub-module uses a clustering algorithm to divide the handwriting in the answer volume into different chunks for subsequent feature extraction. The feature extraction submodule extracts the features of each handwriting chunk, such as the shape, the size, the curve and the like, so as to help identify the handwritten characters. The data arrangement sub-module arranges and constructs the characteristic information so as to facilitate subsequent answer sheet comparison.

Answer sheet comparison module:

and the answer sheet retrieval sub-module retrieves standard answers or reference answers matched with the answer sheets according to the question information. The gray level comparison sub-module is used for gray level comparison of the images of the student answer sheets and the standard answers so as to determine whether the images are correct or not. The index identification submodule identifies and records each index in the scoring criteria for subsequent scoring. The data display sub-module displays the comparison result and the scoring standard to the commentators in an easy-to-read mode, so that scoring decision making is facilitated for the commentators.

The question type identification module:

the question recognition sub-module uses a convolutional neural network or other suitable machine learning algorithm to classify and recognize the questions in each answer sheet. The feature optimization sub-module uses a deep learning technique, such as an automatic encoder, to optimize the topic recognition result and improve the classification accuracy.

And the comment feedback module is used for:

the AR feedback sub-module provides visual feedback for the commentator by using the augmented reality technology, and the recognition result is superimposed on the answer sheet image, so that the commentator can browse the answer sheet content more easily. The interface integration submodule creates a user interface, integrates the question type recognition result, the AR feedback and the scoring function, so that a commentator can conveniently check the answer content, input the score and store the commentary result.

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims (10)

1. An artificial evaluation error investigation method based on an image gray level calculation method is characterized by comprising the following steps:

scanning the student answer sheet into a digital image by using a laser scanning technology, and generating a digital image of the student answer sheet;

based on the digital image of the student answer sheet, adopting a Sobel edge detection algorithm and morphological processing to segment and extract a text region, and generating a text image with gray features;

based on the text image with gray features, performing contrast and definition optimization on the image by adopting an Otsu's self-adaptive threshold method, and generating an optimized image;

based on the optimized image, performing handwriting feature analysis of student writing by using a K-means clustering algorithm to generate handwriting feature information;

Based on the handwriting characteristic information, comparing the gray distribution mode of the answer sheet with that of a standard answer sheet by adopting a structural similarity index algorithm to generate a comparison result and an analysis report;

based on the comparison result and the analysis report, a convolutional neural network is adopted to identify and classify the answer questions based on the comparison result and the corresponding analysis report, and an AR technology is utilized to provide visual feedback for a commentator to generate an answer question identification result and a commentator feedback.

2. The method for artificially evaluating paper errors based on the image gray level calculation method of claim 1, wherein the step of scanning the paper of the student into a digital image by using a laser scanning technique, and generating the digital image of the paper of the student comprises the following steps:

based on the initialization of a laser scanner, placing the answer sheet flat and fixing, and obtaining the fixed-position information of the answer sheet;

starting a laser scanner to perform linear scanning based on the answer sheet fixing in-place information, and generating a preliminary digital answer sheet image;

based on the preliminary digital answer sheet image, adopting a color balance and calibration method to adjust the image quality, and generating a color balance digital answer sheet image;

And based on the color balanced digital answer sheet image, performing image size and resolution standardization to generate a digital image of the student answer sheet.

3. The method for artificially evaluating a paper for a mistake based on an image gray level calculation method according to claim 1, wherein the steps of dividing and extracting a text region based on the digital image of the paper for student answer by using a Sobel edge detection algorithm and morphological processing, and generating a text image with gray level characteristics are specifically as follows:

converting the image by using an image graying method based on the digital image of the student answer sheet to generate a gray answer sheet image;

based on the gray-scale answer sheet image, applying a Sobel edge detection algorithm to generate an edge intensity image;

generating a text region image of enhanced contrast using morphological operations based on the edge intensity image;

and executing a region growing method based on the text region image with enhanced contrast, and generating a text image with gray scale characteristics.

4. The method for artificially evaluating a survey error based on an image gray level calculation method according to claim 1, wherein the steps of optimizing the contrast and the sharpness of the image by using Otsu's adaptive thresholding based on the text image with gray level features, and generating the optimized image are specifically as follows:

Calculating a gray histogram based on the text image with gray features, and generating gray histogram distribution;

based on the gray histogram distribution, determining an optimal threshold value by applying an Otsu's algorithm, and obtaining an optimal binarization threshold value;

based on the optimal binarization threshold value, performing binarization processing on the image to generate a binarized text image;

and based on the binarized text image, enhancing the image by using a histogram equalization technology, and generating an optimized image.

5. The method for manually evaluating a survey error based on an image gray level calculation method of claim 1, wherein the step of performing handwriting feature analysis of student writing by using a K-means clustering algorithm based on the optimized image, and generating handwriting feature information comprises the following steps:

denoising and smoothing by adopting an image preprocessing technology based on the optimized image to generate a clean and optimized image;

based on the clean and optimized image, performing pixel clustering by adopting a K-means clustering algorithm to generate a handwriting feature clustering result;

based on the handwriting feature clustering result, extracting key handwriting information by adopting a feature extraction technology to generate handwriting feature information;

Based on the handwriting characteristic information, the format is unified by adopting a data sorting technology, and formatted handwriting characteristic information is generated.

6. The method for artificially evaluating and surveying errors based on the image gray level calculation method as set forth in claim 1, wherein the steps of comparing gray level distribution patterns of the answer sheet with standard answer sheets by adopting a structural similarity index algorithm based on the handwriting characteristic information, and generating a comparison result and an analysis report are specifically as follows:

based on the formatted handwriting characteristic information, carrying out standard answer sheet retrieval by utilizing a database query technology, and generating a standard answer sheet image;

based on the standard answer sheet image, adopting a structural similarity index algorithm to perform gray pattern comparison to generate comparison data;

based on the comparison data, performing key index identification by adopting a data analysis technology, and generating an analysis report;

based on the analysis report, a visualization technology is adopted for data display, and a comparison result and an analysis report are generated.

7. The method for artificially evaluating paper errors based on the image gray level calculation method according to claim 1, wherein based on the comparison result and the analysis report, a convolutional neural network is adopted to identify and classify paper questions based on the comparison result and the corresponding analysis report, an AR technology is utilized to provide visual feedback for a paper evaluator, and the steps of generating the paper answer identification result and the paper evaluation feedback are specifically as follows:

Based on the comparison result and the analysis report, performing topic identification by adopting a convolutional neural network to generate a topic classification result;

based on the topic classification result, performing feature optimization by adopting a deep learning technology, and generating an optimized topic recognition result;

based on the optimized topic identification result, performing visual feedback by adopting an augmented reality technology to generate AR visual feedback;

based on the AR visual feedback, interface integration is performed by adopting a user interface design technology, and answer sheet recognition results and comment feedback are generated.

8. The manual paper assessment error investigation system based on the image gray level calculation method is characterized by comprising a paper digitizing module, a text region segmentation module, an image optimization module, a handwriting feature analysis module, a paper comparison module, a question identification module and a paper assessment feedback module, wherein the manual paper assessment error investigation system based on the image gray level calculation method is used for executing the manual paper assessment error investigation method based on the image gray level calculation method of any one of claims 1-7.

9. The manual examination paper investigation system based on the image gray level calculation method according to claim 8, wherein the examination paper digitization module is used for digitizing the examination paper based on the laser scanning technology, then performing image quality processing by using an image color balance and calibration method, performing image size and resolution standardization, and generating a digital image of the student examination paper;

The text region segmentation module converts a color image into a gray image by using an image graying method based on a digital image of a student answer sheet, draws image edge information by using a Sobel edge detection algorithm, enhances the contrast between a text and a background by using morphological operation, extracts a main text region by using a region growing method, and generates a text image with gray characteristics;

the image optimization module calculates the gray distribution state of the image by adopting a gray histogram on the basis of the text image with gray characteristics, determines an optimal segmentation threshold by using an Otsu's self-adaptive threshold method, and carries out binarization processing on the image to obtain an optimized image;

the handwriting feature analysis module is used for carrying out image cleaning and smoothing by applying an image preprocessing technology based on the optimized image, carrying out feature mining and clustering on the student writing handwriting by adopting a K-means clustering algorithm, and extracting key handwriting information by utilizing a feature extraction technology to obtain formatted handwriting feature information;

the answer sheet comparison module is used for searching out standard answer sheets based on the formatted handwriting characteristic information by adopting a database query technology, comparing the gray patterns of the answer sheets and the standard answer sheets through a structural similarity index algorithm, and carrying out key index recognition by adopting a data analysis technology to produce comparison results and analysis reports;

The problem type recognition module performs student answer problem type recognition by using a convolutional neural network based on a comparison result and an analysis report, classifies answer content, performs feature optimization by using a deep learning technology, and generates an optimized problem type recognition result;

and the evaluation feedback module is used for providing visual feedback based on the optimized question type recognition result by using an augmented reality technology, and performing result presentation and feedback to form an answer sheet question type recognition result and an evaluation feedback.

10. The manual paper assessment error investigation system based on the image gray level calculation method according to claim 8, wherein the paper answer digitizing module comprises a paper answer scanning sub-module, an image quality adjusting sub-module and an image standardization sub-module;

the text region segmentation module comprises an image graying sub-module, an edge detection sub-module, a morphological operation sub-module and a region growing sub-module;

the image optimization module comprises a gray level histogram sub-module, a threshold value determination sub-module, a binarization processing sub-module and an image enhancement sub-module;

the handwriting feature analysis module comprises an image preprocessing sub-module, a pixel clustering sub-module, a feature extraction sub-module and a data arrangement sub-module;

The answer sheet comparison module comprises an answer sheet retrieval sub-module, a gray comparison sub-module, an index identification sub-module and a data display sub-module;

the question type recognition module comprises a question type recognition sub-module and a characteristic optimization sub-module;

the evaluation feedback module comprises an AR feedback sub-module and an interface integration sub-module.