OCR image fine segmentation system based on edge detection

Abstract

The invention relates to the technical field of OCR image fine segmentation, in particular to an OCR image fine segmentation system based on edge detection. According to the technical scheme, the OCR image fine segmentation system based on edge detection comprises an edge detection module and an edge refining and connecting module; a traditional OCR image fine segmentation system generally adopts a relatively simple image method such as global thresholding or region growing, edge information is not fully utilized, meanwhile, a segmentation method based on characteristics such as pixel colors and textures is generally adopted, the adaptive capacity to a complex background is relatively weak, and compared with the traditional OCR image fine segmentation system, the method provided by the invention has the advantages that the segmentation efficiency is greatly improved; through the edge detection module and the edge refining and connecting module, the edge information in the image can be accurately identified and refined, and meanwhile, the text region can be more accurately positioned and segmented by utilizing the edge detection result and combining the region growing algorithm and the contour analyzer.

Description

Technical Field

[0001] The present invention relates to the technical field of OCR image fine segmentation, and particularly to an OCR image fine segmentation system based on edge detection. Background Art

[0002] An OCR image fine segmentation system is an image processing system specifically used for finely segmenting an image containing text for subsequent character recognition and information extraction. Through a series of complex algorithms and techniques, it accurately separates the text area in the image from the background or other non-text areas, and further divides the text area into individual characters or words, providing an image processing system that can finely segment an image containing text for subsequent character recognition and information extraction.

[0003] Traditional OCR image fine segmentation systems generally use relatively simple image methods such as global thresholding or region growing, which do not make full use of edge information. At the same time, they generally use segmentation methods based on features such as pixel color and texture, and have weak adaptability to complex backgrounds. Summary of the Invention

[0004] In order to overcome the problems that traditional OCR image fine segmentation systems generally use relatively simple image methods such as global thresholding or region growing, which do not make full use of edge information, and at the same time generally use segmentation methods based on features such as pixel color and texture, and have weak adaptability to complex backgrounds.

[0005] The technical solution of the present invention is: an OCR image fine segmentation system based on edge detection, characterized in that it includes:

[0006] An image acquisition module: responsible for acquiring the OCR image to be processed;

[0007] A preprocessing module: preliminarily processes the OCR image to be processed collected by the image acquisition module to improve the image quality;

[0008] An edge detection module: detects the edge information in the OCR image;

[0009] An edge thinning and connection module: thins the edge information detected by the edge detection module to remove redundant pixel points;

[0010] A text area segmentation module: based on edge detection, separates the text area in the OCR image from the background or other non-text areas;

[0011] A feature extraction and recognition module: extracts the features of the text area and then performs character recognition;

[0012] Post - processing and Output Module: Post - process the recognition results and output the recognition results.

[0013] Preferably, the image acquisition module includes a scanner and an interface circuit. The scanner is used to capture the OCR image to be processed, and the interface circuit is used to transmit the OCR image data to a computer or a processing unit. The image acquisition module is used to obtain the OCR image data to be processed. The image acquisition module converts the text content on a paper document, an electronic screen or other visual media into a digital image signal through a high - precision camera, scanner or dedicated image capture device and transmits it to the image processing unit.

[0014] Preferably, the pre - processing module includes a filter, a binarization unit and an image enhancer. The filter is used to remove the noise in the OCR image. The binarization unit is used to convert the OCR image into a binary image for subsequent edge detection. The image enhancer is used to enhance the image features through contrast adjustment and sharpening. The pre - processing module is used to optimize the OCR image to be processed. The pre - processing module first applies a variety of filter techniques, including Gaussian filtering and median filtering, to remove the noise and defects in the OCR image. Through image binarization, a color or grayscale image is converted into a binary image containing only two colors, black and white, to simplify the subsequent edge detection process. At the same time, the pre - processing module also performs enhancement processing on the binary image to highlight the text features.

[0015] Preferably, the edge detection module includes an edge detection algorithm, a gradient calculator, a non - maximum suppression unit and a double - threshold processor. The gradient calculator is used to calculate the gradient magnitude and direction of the OCR image to determine the position of the edge; the non - maximum suppression unit is used to retain the local maximum in the gradient direction and suppress non - edge points; the double - threshold processor is used to screen edge points by setting two thresholds, a high threshold and a low threshold, to reduce false edges. The edge detection module uses the Canny algorithm, Sobel operator or Laplacian operator to perform pixel - by - pixel analysis on the OCR image, calculate the gradient magnitude and direction of each pixel, and retain the local maximum in the gradient direction through non - maximum suppression technology to suppress non - edge points, thereby refining the edge. At the same time, combined with the double - threshold processing strategy, two thresholds, a high threshold and a low threshold, are set to screen and confirm the gradient magnitude.

[0016] Preferably, the edge thinning and connection module includes a thinning algorithm and an edge connector. The edge connector is used to connect edge points into a contour according to the gradient direction. The edge thinning and connection module is used to detect the edges in the OCR image using image processing technology. The edge corresponds to the region with the highest brightness change in the OCR image. The edge specifically includes the contour of the text. The detected edges are thinned to remove redundant edge information and retain key edge features. At the same time, the thinned edges are connected to form a complete edge contour.

[0017] Preferably, when the edge detection module and the edge thinning and connection module are working, they include the following steps:

[0018] S101: The edge detection module first uses a gradient calculator to calculate the gradient magnitude and direction of the OCR image. The non-maximum suppressor retains the local maximum in the gradient direction and suppresses non-edge points, thereby thinning the edges;

[0019] S102: The double-threshold processor screens edge points by setting two thresholds, a high threshold and a low threshold, to reduce false edges;

[0020] S103: The edge thinning and connection module thins the detected edges, removes redundant edge information, and retains key edge features;

[0021] S104: After the thinning process, the edge connector connects the edge points into a contour according to the gradient direction to form a complete edge contour;

[0022] S105: In the edge detection module, a confidence evaluation is performed on the detected edge points. The confidence is determined based on the gradient magnitude, the result after non-maximum suppression, and the passing of the high threshold in the double-threshold processing;

[0023] S106: In the text region segmentation module, the threshold for region growing is dynamically adjusted according to the confidence of the edge points;

[0024] S107: For edge points with high confidence, the similarity threshold for region growing is lowered, and for edge points with low confidence, the similarity threshold for region growing is raised;

[0025] S108: During the contour analysis process, if the contour shape of a certain region does not match the expected text region shape, it is fed back to the edge detection module to re-evaluate the edge confidence of that region;

[0026] S109: According to the feedback result, the edge detection module adjusts its parameters and algorithms.

[0027] Preferably, the text region segmentation module includes a region growing algorithm and a contour analyzer. The region growing algorithm is used to start from a seed point and gradually expand according to the similarity of pixels to form a text region. The contour analyzer is used to analyze the shape and position of the edge contour to determine the range of the text region. The text region segmentation module is used to utilize the result of edge detection, combine image analysis techniques, locate the text region in the OCR image, separate the located text region from the OCR image, remove the interference of non-text regions, and further segment the text region into individual characters.

[0028] Preferably, the feature extraction and recognition module includes a feature extractor and a character recognizer. The feature extractor is used to extract the contour features and stroke features of characters, and the character recognizer is used to classify and recognize the extracted features based on machine learning or deep learning algorithms to obtain the text content. The feature extraction and recognition module is used to extract features from the segmented characters, convert the character images into numerical representations for the classifier to process, and at the same time use the classifier to classify the extracted character features to recognize the corresponding character categories. The classifier is a pre-trained model.

[0029] Preferably, when the feature extraction and recognition module works, it includes the following steps:

[0030] S201: The feature extraction and recognition module first receives the output from the text area segmentation module, that is, the already segmented single character images;

[0031] S202: The feature extractor first extracts the contour features of the character image. The extracted contour features include the external contour and internal holes of the character;

[0032] S203: The feature extractor simultaneously extracts other features of the character, including the histogram of oriented gradients and local binary patterns;

[0033] S204: The character recognizer selects a classifier to classify the extracted features based on support vector machine or random forest technology;

[0034] S205: The recognized character categories are output to the post-processing and output module for further verification, format adjustment, display, and storage.

[0035] Preferably, the post-processing and output module includes an error corrector, a format adjuster, and an output device. The error corrector is used to correct the recognition results based on a language model or context information. The format adjuster is used to adjust the format of the output text according to user requirements. The output device is used to display and store the recognition results. The post-processing and output module is used to post-process the recognized results and at the same time output the post-processed recognition results in text form for the user to perform subsequent processing.

[0036] Advantages of the present invention:

[0037] Compared with traditional OCR image fine segmentation systems that generally use relatively simple image methods such as global thresholding or region growing, and do not make full use of edge information, and generally use segmentation methods based on features such as pixel color and texture, with weak adaptability to complex backgrounds, this OCR image fine segmentation system can accurately identify and refine the edge information in the image through an edge detection module and an edge refinement and connection module. At the same time, using the edge detection results and combining with the region growing algorithm and contour analyzer, it can more accurately locate and segment the text region. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 FIG. shows a schematic diagram of the framework of the OCR image fine segmentation system based on edge detection of the present invention;

[0039] Figure 2 FIG. shows a schematic diagram of the working process of the edge detection module and the edge refinement and connection module of the OCR image fine segmentation system based on edge detection of the present invention;

[0040] Figure 3 FIG. shows a schematic diagram of the process of the feature extraction and recognition module of the OCR image fine segmentation system based on edge detection of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0041] The present invention will be further described below with reference to the drawings and embodiments.

[0042] Please refer to Figures 1-3 , the present invention provides an embodiment: an OCR image fine segmentation system based on edge detection, including:

[0043] Image acquisition module: responsible for acquiring the OCR image to be processed;

[0044] Preprocessing module: preliminarily processes the acquired original image to improve the image quality;

[0045] Edge detection module: detects the edge information in the image;

[0046] Edge refinement and connection module: refines the detected edges, removes redundant pixel points, and makes the edges clearer and more accurate;

[0047] Text region segmentation module: based on edge detection, separates the text region in the image from the background or other non-text regions;

[0048] Feature extraction and recognition module: extracts the features of the text region and then performs character recognition;

[0049] Post-processing and output module: post-processes the recognition results and then outputs the recognition results in an appropriate form.

[0050] The image acquisition module includes a scanner and an interface circuit. The scanner is used to capture the original image, and the interface circuit is used to transfer the image data to a computer or a processing unit. The image acquisition module is used to efficiently and accurately obtain the original image data to be processed. The image acquisition module converts the text content on paper documents, electronic screens, or other visual media into digital image signals through high-precision cameras, scanners, or dedicated image capture devices. These image signals are then transmitted to the image processing unit, providing basic data support for subsequent steps such as image preprocessing and edge detection. The performance of the image acquisition module directly affects the overall recognition effect and accuracy of the OCR system.

[0051] Preferably, the preprocessing module includes a filter, a binarization unit, and an image enhancer. The filter is used to remove noise in the image. The binarization unit is used to convert the image into a binary image, that is, an image with only two colors, black and white, which is convenient for subsequent edge detection. The image enhancer is used to enhance the image features through contrast adjustment and sharpening. The preprocessing module is used to perform a series of optimization processes on the original image to improve the image quality and recognition accuracy. The preprocessing module first applies various filter techniques, including Gaussian filtering and median filtering, to effectively remove noise and defects in the image. Through image binarization, the color or grayscale image is converted into a binary image containing only two colors, black and white, simplifying the subsequent edge detection process. At the same time, the preprocessing module also performs enhancement processing on the image to highlight the text features and make the characters more clearly distinguishable.

[0052] Preferably, the edge detection module includes an edge detection algorithm, a gradient calculator, a non-maximum suppressor, and a double-threshold processor. The gradient calculator is used to calculate the gradient magnitude and direction of the image to determine the position of the edge. The non-maximum suppressor is used to retain the local maximum in the gradient direction and suppress non-edge points. The double-threshold processor is used to further screen edge points by setting two thresholds, a high threshold and a low threshold, to reduce false edges. The edge detection module is responsible for accurately identifying the edge information in the image. The edge detection module uses the Canny algorithm, Sobel operator, or Laplacian operator to perform pixel-by-pixel analysis on the image, calculate the gradient magnitude and direction of each pixel, and retain the local maximum in the gradient direction through non-maximum suppression technology to suppress non-edge points, thereby refining the edge. At the same time, combined with the double-threshold processing strategy, two thresholds, a high threshold and a low threshold, are set to screen and confirm the gradient magnitude, further improving the accuracy and robustness of edge detection. The output result of the edge detection module provides important edge feature information for subsequent text region segmentation and character recognition.

[0053] Preferably, the edge refinement and connection module includes a refinement algorithm and an edge connector. The edge connector is used to connect edge points into contours according to the gradient direction. The edge refinement and connection module is used to detect edges in the image using image processing technology. The edges correspond to areas in the image where the brightness changes significantly, including the contours of text. The detected edges are refined to remove redundant edge information and retain key edge features. At the same time, the refined edges are connected to form a complete edge contour.

[0054] Preferably, when the edge detection module and the edge refinement and connection module are linked, the following steps are included:

[0055] S101: The edge detection module first uses a gradient calculator to calculate the gradient amplitude and direction of the image, and a non-maximum suppressor retains the local maximum value in the gradient direction and suppresses non-edge points, thereby refining the edge;

[0056] S102: The dual threshold processor further screens edge points and reduces false edges by setting two high and low thresholds;

[0057] S103: The edge refinement and connection module refines the detected edges, removes redundant edge information, and retains key edge features;

[0058] S104: After the thinning process, the edge connector connects the edge points into contours according to the gradient direction to form a complete edge contour;

[0059] S105: In the edge detection module, the confidence level of the detected edge points is evaluated, and the confidence level is determined based on the gradient amplitude, the result after non-maximum suppression, and the high threshold passing condition in the dual threshold processing;

[0060] S106: In the text region segmentation module, dynamically adjust the threshold of region growth according to the confidence of the edge point;

[0061] S107: For edge points with high confidence, lower the similarity threshold of region growing, and for edge points with low confidence, increase the similarity threshold;

[0062] S108: During the contour analysis process, if the contour shape of a certain area does not match the expected text area shape, the contour shape is fed back to the edge detection module to re-evaluate the edge confidence of the area;

[0063] S109: Based on the feedback result, the edge detection module can adjust its parameters or algorithms to more accurately detect the edge of the area.

[0064] Preferably, the text area segmentation module includes a region growing algorithm and a contour analyzer. The region growing algorithm is used to start from a seed point and gradually expand according to the similarity of pixels to form a text area. The contour analyzer is used to analyze the shape and position of the edge contour to determine the range of the text area. The text area segmentation module is used to utilize the result of edge detection, combine image analysis techniques to locate the text area in the image, separate the located text area from the image, remove the interference of non-text areas such as the background, and further segment the text area into individual characters.

[0065] Preferably, the feature extraction and recognition module includes a feature extractor and a character recognizer. The feature extractor is used to extract the contour features and stroke features of characters. The character recognizer is used to classify and recognize the extracted features based on machine learning or deep learning algorithms to obtain the text content. The feature extraction and recognition module is used to extract features from the segmented characters, convert the character image into a numerical representation that can be processed by a classifier, and at the same time use the classifier to classify the extracted character features to identify the corresponding character categories. The classifier can be a pre-trained model or a custom-trained model.

[0066] Preferably, when the feature extraction and recognition module works, it includes the following steps:

[0067] S201: The feature extraction and recognition module first receives the output from the text area segmentation module, that is, the already segmented individual character images;

[0068] S202: The feature extractor first extracts the contour features of the character image. The contour is an important description of the character shape and can be obtained through an edge detection algorithm or a contour tracking algorithm. The extracted contour features include the external contour and internal holes of the character;

[0069] S203: The feature extractor simultaneously extracts other features of the character, including the histogram of oriented gradients and local binary patterns;

[0070] S204: The character recognizer selects a suitable classifier to classify the extracted features based on support vector machine or random forest technology;

[0071] S205: The recognized character categories are output to the post-processing and output module for further verification, format adjustment, display, and storage.

[0072] Preferably, the post - processing and output module includes an error corrector, a format adjuster, and an output device. The error corrector is used to correct the recognition result based on a language model or context information. The format adjuster is used to adjust the format of the output text according to user requirements. The output device is used to display and save the recognition result. The post - processing and output module is used to post - process the recognized result and output the post - processed recognition result in text form for subsequent user processing.

[0073] Example 1

[0074] Application of OCR Image Fine Segmentation System Based on Edge Detection in Complex Documents

[0075] S301: Use a high - resolution scanner to collect document images containing complex backgrounds (such as a mixture of charts, pictures, and text), ensuring image clarity and detail retention;

[0076] S302: Apply Gaussian filtering to remove image noise, use the Otsu binarization method to automatically determine the threshold, convert the image into a binary image, and enhance the contrast between text and background through contrast enhancement and sharpening processing;

[0077] S303: Adopt the Canny edge detection algorithm to calculate the gradient magnitude and direction of the image, perform non - maximum suppression to reduce non - edge pixels, and set high and low double thresholds to accurately screen edge points;

[0078] S304: Apply the Zhang - Suen thinning algorithm to thin the edges, use an edge connector to connect broken edges according to the gradient direction to form a complete contour;

[0079] S305: Use the region - growing algorithm to expand from seed points to identify text regions, and a contour analyzer analyzes the edge contour to accurately define the text range;

[0080] S306: Extract the contour features, direction gradient histogram, and local binary pattern of characters, and use a pre - trained convolutional neural network as a character recognizer to classify the features;

[0081] S307: The error corrector corrects recognition errors based on a language model, the format adjuster adjusts the output format according to user requirements, and the output device saves the final result in text file form or displays it on the screen;

[0082] S308: Comparing experimental data, the F1 - score of the traditional OCR method on complex - background documents is 72.3%, and the F1 - score of the system in this example is increased to 86.7% on the same test set.

[0083] Example 2

[0084] Application of OCR Image Fine Segmentation System Based on Edge Detection in Multilingual Mixed Documents

[0085] S401: Use a high-precision camera to capture a document containing a mixture of multiple languages (such as English, Chinese, Arabic), ensuring image quality;

[0086] S402: Adopt median filtering to reduce salt-and-pepper noise in the image, apply an adaptive binarization method to process text under different lighting conditions, and use image enhancement techniques to improve the readability of the text;

[0087] S403: Use the Sobel operator for edge detection, calculate the gradient, perform non-maximum suppression, thin the edges, and perform double-threshold processing to reduce false edges and improve the accuracy of edge detection;

[0088] S404: Apply the Guo-Hall thinning algorithm to further thin the edges. The edge connector uses gradient information to connect the edges to form a coherent text contour;

[0089] S405: Combine morphological operations, optimize the region growing algorithm, accurately segment the multilingual text regions, and the contour analyzer identifies and distinguishes text blocks in different languages;

[0090] S406: Extract the stroke features, HOG features, and LBP features of the characters, and use a custom model trained by a deep learning framework for character recognition, supporting multilingual recognition;

[0091] S407: The error corrector corrects the recognition results by combining multilingual dictionaries and context information. The format adjuster supports multiple output formats, including PDF and Word, and the output device displays and saves the final recognition results;

[0092] S408: Comparing the experimental data, the F1-score of the traditional OCR method on multilingual mixed documents is 68.5%, and the F1-score of the system in this embodiment reaches 84.2% on the same test set.

[0093] The above has described the embodiments of the present invention in detail with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the gist of the present invention.

Claims

1. OCR image fine segmentation system based on edge detection, characterized by: Included are: Image acquisition module: responsible for acquiring the OCR image to be processed; Preprocessing module: performs preliminary processing on the OCR image to be processed collected in the image acquisition module to improve the image quality; Edge detection module: detects edge information in OCR images; Edge refinement and connection module: refines the edge information detected by the edge detection module and removes redundant pixels; Text region segmentation module: Based on edge detection, the text region in the OCR image is segmented from the background or other non-text regions; Feature extraction and recognition module: extract the features of the text area and then perform character recognition; Post-processing and output module: post-process the recognition results and output the recognition results.

2. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The image acquisition module includes a scanner and an interface circuit. The scanner is used to capture the OCR image to be processed. The interface circuit is used to transmit the OCR image data to a computer or a processing unit. The image acquisition module is used to obtain the OCR image data to be processed. The image acquisition module converts the text content on a paper document, an electronic screen or other visual media into a digital image signal and transmits it to the image processing unit through a high-precision camera, a scanner or a dedicated image capture device.

3. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The preprocessing module includes a filter, a binarization unit and an image enhancer. The filter is used to remove noise from the OCR image. The binarization unit is used to convert the OCR image into a binary image to facilitate subsequent edge detection. The image enhancer is used to enhance image features through contrast adjustment and sharpening. The preprocessing module is used to optimize the OCR image to be processed. The preprocessing module first applies a variety of filter technologies, including Gaussian filtering and median filtering, to remove noise and defects from the OCR image. Through image binarization, the color or grayscale image is converted into a binary image containing only black and white colors, simplifying the subsequent edge detection process. At the same time, the preprocessing module also enhances the binary image to highlight the text features.

4. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The edge detection module includes an edge detection algorithm, a gradient calculator, a non-maximum suppressor and a dual threshold processor. The gradient calculator is used to calculate the gradient amplitude and direction of the OCR image and determine the position of the edge; the non-maximum suppressor is used to retain the local maximum value in the gradient direction and suppress non-edge points; the dual threshold processor is used to screen edge points and reduce pseudo edges by setting two high and low thresholds; the edge detection module uses the Canny algorithm, Sobel operator or Laplacian operator to analyze the OCR image pixel by pixel, calculate the gradient amplitude and direction of each pixel, and retain the local maximum value in the gradient direction through non-maximum suppression technology, suppress non-edge points, thereby refining the edge. At the same time, combined with the dual threshold processing strategy, two high and low thresholds are set to screen and confirm the gradient amplitude.

5. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The edge refinement and connection module includes a refinement algorithm and an edge connector. The edge connector is used to connect edge points into contours according to the gradient direction. The edge refinement and connection module is used to detect edges in OCR images using image processing technology. The edges correspond to the areas with the highest brightness changes in the OCR images. The edges specifically include the contours of text. The detected edges are refined to remove redundant edge information and retain key edge features. At the same time, the refined edges are connected to form a complete edge contour.

6. The OCR image fine segmentation system based on edge detection according to claim 5, characterized in that: When the edge detection module and the edge refinement and connection module are working, the following steps are included: S101: The edge detection module first uses a gradient calculator to calculate the gradient amplitude and direction of the OCR image, and a non-maximum suppressor retains the local maximum value in the gradient direction and suppresses non-edge points, thereby refining the edge; S102: The dual threshold processor sets two high and low thresholds to filter edge points and reduce false edges; S103: The edge refinement and connection module refines the detected edges, removes redundant edge information, and retains key edge features; S104: After the thinning process, the edge connector connects the edge points into contours according to the gradient direction to form a complete edge contour; S105: In the edge detection module, the confidence level of the detected edge points is evaluated, and the confidence level is determined based on the gradient amplitude, the result after non-maximum suppression, and the high threshold passing condition in the dual threshold processing; S106: In the text region segmentation module, dynamically adjust the threshold of region growth according to the confidence of the edge point; S107: For edge points with high confidence, lower the similarity threshold of region growing, and for edge points with low confidence, increase the similarity threshold of region growing; S108: During the contour analysis process, if the contour shape of a certain area does not match the expected text area shape, feedback is given to the edge detection module to re-evaluate the edge confidence of the area; S109: According to the feedback results, the edge detection module adjusts its parameters and algorithms.

7. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The text region segmentation module includes a region growing algorithm and a contour analyzer. The region growing algorithm is used to start from a seed point and gradually expand according to the similarity of pixels to form a text region. The contour analyzer is used to analyze the shape and position of the edge contour and determine the scope of the text region. The text region segmentation module is used to use the results of edge detection combined with image analysis technology to locate the text region in the OCR image, separate the located text region from the OCR image, remove the interference of non-text areas, and further segment the text region into single characters.

8. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The feature extraction and recognition module includes a feature extractor and a character recognizer. The feature extractor is used to extract the contour features and stroke features of the characters. The character recognizer is used to classify and recognize the extracted features based on machine learning or deep learning algorithms to obtain text content. The feature extraction and recognition module is used to extract features from the segmented characters and convert the character images into numerical representations for processing by the classifier. At the same time, the classifier is used to classify the extracted character features and identify the corresponding character categories. The classifier is a pre-trained model.

9. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The feature extraction and recognition module includes the following steps when working: S201: The feature extraction and recognition module first receives the output from the text region segmentation module, that is, the segmented single character image; S202: The feature extractor first extracts contour features from the character image, where the extracted contour features include the outer contour and inner holes of the character; S203: The feature extractor simultaneously extracts other features of the character, including a histogram of oriented gradients and a local binary pattern; S204: The character recognizer selects a classifier to classify the extracted features based on a support vector machine or a random forest technique; S205: The identified character categories are output to the post-processing and output module for further verification, format adjustment, display and storage.

10. The OCR image fine segmentation system based on edge detection according to claim 1, characterized in that: The post-processing and output module includes an error corrector, a format adjuster and an output device. The error corrector is used to correct the recognition results based on the language model or context information. The format adjuster is used to adjust the format of the output text according to user needs. The output device is used to display and save the recognition results. The post-processing and output module is used to post-process the recognition results and output the post-processed recognition results in text form for subsequent processing by the user.

Images