Ancient book intelligent identification and restoration method and device based on machine learning

Abstract

The invention provides an ancient book intelligent identification and restoration method and device based on machine learning, and relates to the technical field of ancient book identification and restoration. According to the method, the image features of each to-be-recognized area in the paper image of the to-be-repaired ancient book are extracted, the intelligent recognition model of the ancient book is input, intelligent recognition is carried out, multiple recognition results of each to-be-recognized area are obtained, and then context semantic analysis and intelligent matching are carried out based on the multiple recognition results in combination with the ancient book database. According to the method, the to-be-repaired ancient book is automatically identified, the repair scheme is automatically generated, the professional requirement of a repairer is reduced, the repair time of the repairer is shortened, and the identification and repair efficiency of the ancient book is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ancient book recognition and restoration, and in particular to an ancient book intelligent recognition and restoration method and device based on machine learning. Background Art

[0002] Ancient books are important resources for academic research and are also important carriers of cultural inheritance and development. However, as time goes by, these precious ancient books and documents are generally facing severe challenges such as aging and damage, and are in urgent need of effective protection and restoration. The identification and restoration of ancient books has therefore become a vital task, which not only requires a high degree of professionalism, but also implies respect for history and cultural inheritance.

[0003] In the traditional field of ancient book identification and restoration, manual operation has always dominated. This method relies on the restorer's profound professional background and rich practical experience. The restorer needs to have extensive knowledge of history, literature, art and even papermaking technology in order to accurately identify the age, material, style and even the cause of damage of the ancient book. On this basis, they also need to consult a large amount of literature and compare and analyze to determine the best restoration plan. This process is not only time-consuming and laborious, but also greatly tests the restorer's personal ability and patience.

[0004] As time goes by, the inefficiency of traditional manual restoration methods for ancient books has become increasingly prominent. Faced with a large number of ancient book collections in various conditions, it is difficult for only a few experienced restorers to meet the growing demand for restoration. In addition, the high degree of precision and strong irreversibility of ancient book restoration work itself have further prolonged the entire restoration cycle. The inefficient manual restoration method of ancient books not only affects the effective use of ancient book resources, but also poses a potential threat to cultural inheritance and development. Summary of the invention

[0005] The present invention provides an ancient book intelligent recognition and restoration method and device based on machine learning, which can realize automatic recognition of ancient books and automatic generation of restoration plans, reduce the professional requirements of restorers, reduce the restoration time of restorers, and improve the recognition and restoration efficiency of ancient books.

[0006] In a first aspect, the present invention provides a method for intelligent identification and restoration of ancient books based on machine learning, the method comprising: obtaining a paper image of the ancient book to be restored; performing image segmentation on the paper image of the ancient book to be restored, and determining a plurality of areas to be identified; performing feature extraction on the plurality of areas to be identified, and determining image features of each area to be identified; the image features include color features, texture features, shape features, and size features; based on the image features of each area to be identified, and an intelligent recognition model of ancient books, determining a plurality of recognition results for each area to be identified; based on the plurality of recognition results for each area to be identified, and an ancient book database, performing semantic analysis and intelligent matching analysis on contextual information of the ancient book to be restored, and determining a restoration plan for the ancient book to be restored; and based on the restoration plan for the ancient book to be restored, repairing the ancient book to be restored.

[0007] In one possible implementation, the paper image of the ancient book to be restored is segmented to determine a plurality of areas to be identified, including: edge detection is performed on the paper image to determine a plurality of edge points in the paper image; based on the plurality of edge points, a plurality of text areas in the paper image are determined; image segmentation is performed based on the plurality of text areas to obtain a plurality of areas to be identified; each area to be identified includes one or more text areas, and a non-text area adjacent to the text area.

[0008] In a possible implementation, feature extraction is performed on multiple areas to be identified to determine the image features of each area to be identified, including: calculating the color histogram of each area to be identified, and determining the color features of each identification area based on the color histogram of each area to be identified; determining the grayscale co-occurrence matrix of each area to be identified, and calculating the texture features of each area to be identified based on the grayscale co-occurrence matrix of each area to be identified, the texture features including contrast, homogeneity and correlation; determining the text contours of each text area based on the edge points of each text area in each area to be identified; determining the shape features of each area to be identified based on the text contours of each text area; calculating the size features of each text based on the text contours of each text area in each area to be identified; determining the size features of the non-text area based on the contours of the non-text area of ​​each area to be identified; determining the size features of each area to be identified based on the size features of each text and the size features of the non-text area; performing feature fusion based on the color features, texture features, shape features and size features of each area to be identified to obtain the image features of each area to be identified.

[0009] In a possible implementation, multiple recognition results for each area to be recognized are determined based on the image features of each area to be recognized and an intelligent recognition model for ancient books, including: determining the age of the ancient book to be restored based on the image features of each area to be recognized and an age recognition module of the intelligent recognition model for ancient books; screening multiple glyph recognition modules of the intelligent recognition model for ancient books based on the age of the ancient book to be restored to obtain a target glyph recognition module corresponding to the age of the ancient book to be restored; performing glyph recognition based on the image features of each area to be recognized and the target glyph recognition module to obtain multiple recognition results for each area to be recognized; wherein the multiple recognition results are multiple recognition results whose probabilities are greater than a set value among the recognition results of the area to be recognized.

[0010] In a possible implementation, before determining multiple recognition results for each area to be recognized based on the image features of each area to be recognized and the ancient book intelligent recognition model, it also includes: acquiring image features and text data of multiple ancient books of determined ages and glyphs; generating a first training sample based on the image data and age of each ancient book; generating a second training sample for each age based on the image features and text data of each ancient book in each age; performing neural network training based on the first training samples to obtain an age recognition module of the ancient book intelligent recognition model; and performing neural network training based on the second training samples of each age to obtain a glyph recognition module for each age.

[0011] In a possible implementation, semantic analysis and intelligent matching analysis are performed on the context information of the ancient book to be repaired based on multiple recognition results of each area to be recognized and a database of ancient books, and a repair plan for the ancient book to be repaired is determined, including: determining the textual semantics of each recognition result in each area to be recognized based on multiple recognition results of each area to be recognized and a database of ancient books; performing contextual semantic association based on the textual semantics of each recognition result in each area to be recognized, and determining multiple semantic schemes for the ancient book to be repaired; calculating evaluation indicators for each semantic scheme based on multiple semantic schemes for the ancient book to be repaired, the evaluation indicators including semantic coherence, historical accuracy, logical rationality and text readability; determining the repair plan for the ancient book to be repaired based on the evaluation indicators of each semantic scheme.

[0012] In a possible implementation, based on multiple recognition results of each area to be identified and a database of ancient books, semantic analysis and intelligent matching analysis are performed on the contextual information of the ancient book to be repaired to determine the repair plan for the ancient book to be repaired, and the method also includes: based on the repair plan, preliminarily generating a repaired interface of the ancient book to be repaired, and displaying the repaired interface to the repair personnel; receiving the repair personnel's annotation information on the repaired interface; the annotation information includes text modifications, semantic corrections and / or text completions on the repaired interface; based on the annotation information, re-performing semantic analysis and matching analysis, and regenerating the repair plan for the ancient book to be repaired.

[0013] In a possible implementation, the ancient book to be repaired is repaired based on the restoration plan of the ancient book to be repaired, including: determining the background color and text color of the ancient book to be repaired based on the age of the ancient book to be repaired and the current color of the ancient book to be repaired; determining the glyph data of the ancient book to be repaired based on the restoration plan of the ancient book to be repaired; performing digital rendering based on the glyph data, as well as the background color and text color, to obtain a digitized interface of the ancient book to be repaired; performing feature extraction based on the digitized interface of the ancient book to be repaired to obtain image features of each area to be identified in the digitized interface; determining the restoration method and restoration content of each area to be identified based on the image features of each area to be identified in the paper image and the image features of each area to be identified in the digitized interface.

[0014] In a possible implementation, the method also includes: obtaining images before and after restoration of the manually restored ancient book; determining a restoration plan estimated for manual restoration based on the images before restoration, an ancient book intelligent recognition model, and an ancient book database; performing comparison based on the restoration plan and the images after restoration to determine the difference between manual restoration and intelligent recognition of each area in the manually restored ancient book; generating restoration quality inspection information based on the difference between manual restoration and intelligent recognition of each area, the restoration quality inspection information including multiple disputed areas where there are disputes in manual restoration, and the disputed content of each disputed area; and presenting the restoration quality inspection information to the restoration personnel.

[0015] In a second aspect, an embodiment of the present invention provides an ancient book intelligent recognition and restoration device based on machine learning, the device comprising: a communication module and a processing module, the communication module being used to obtain a paper image of the ancient book to be restored; the processing module being used to perform image segmentation on the paper image of the ancient book to be restored, and determine a plurality of areas to be recognized; perform feature extraction on the plurality of areas to be recognized, and determine the image features of each area to be recognized; the image features comprising color features, texture features, shape features and size features; based on the image features of each area to be recognized, and an ancient book intelligent recognition model, determine a plurality of recognition results for each area to be recognized; based on the plurality of recognition results for each area to be recognized, and an ancient book database, perform semantic analysis and intelligent matching analysis on the context information of the ancient book to be restored, and determine a restoration plan for the ancient book to be restored; and based on the restoration plan for the ancient book to be restored, repair the ancient book to be restored.

[0016] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, the memory storing a computer program, the processor being used to call and run the computer program stored in the memory to perform the steps of the method described in the first aspect and any possible implementation method of the first aspect.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and wherein when the computer program is executed by a processor, the steps of the method described in the first aspect and any possible implementation method of the first aspect are implemented.

[0018] The present invention provides an ancient book intelligent recognition and restoration method and device based on machine learning. The present invention extracts image features of each to-be-recognized area in a paper image of the ancient book to be restored, inputs the features into an ancient book intelligent recognition model, performs intelligent recognition, and obtains multiple recognition results of each to-be-recognized area. Then, based on the multiple recognition results, context semantic analysis and intelligent matching are performed in combination with an ancient book database to determine a restoration plan for the ancient book to be restored, and the ancient book to be restored is restored, thereby realizing automatic recognition of the ancient book to be restored and automatic generation of the restoration plan, reducing the professional requirements of restorers, reducing the restoration time of restorers, and improving the recognition and restoration efficiency of ancient books. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0020] Figure 1 It is a flow chart of an ancient book intelligent recognition and restoration method based on machine learning provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of an ancient book intelligent recognition and restoration device based on machine learning provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0022] In the description of the present invention, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, "at least one" and "plurality" refer to two or more. The words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not limit them to be different.

[0023] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.

[0024] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include other steps or modules that are not listed, or may optionally include other steps or modules that are inherent to these processes, methods, products or devices.

[0025] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be described through specific embodiments in conjunction with the accompanying drawings of the present invention.

[0026] As described in the background art, the current traditional manual restoration methods for ancient books have the problem of low efficiency.

[0027] To solve the above technical problems, Figure 1 As shown, the embodiment of the present invention provides an ancient book intelligent recognition and restoration method based on machine learning. The method includes steps S101-S106.

[0028] S101, obtaining a paper image of the ancient book to be restored.

[0029] In some embodiments, the paper image of the ancient book to be restored is an image of a page of the ancient book obtained by a digitizing device. The paper image contains complete information such as text, illustrations, seals, and physical damage features (such as wormholes, stains, folds, etc.).

[0030] The embodiments of the present invention can use a high-resolution scanner (≥600dpi) or a multispectral imaging system (such as infrared / ultraviolet imaging) to capture ink or watermarks that are invisible under visible light. Use non-local means filtering (Non-Local Means) or a deep learning denoising model (such as DnCNN) to eliminate dust and scanning noise. Use histogram equalization or Retinex algorithm to improve contrast and highlight faded text. Use perspective transformation (Perspective Transformation) to correct page bending deformation.

[0031] S102, performing image segmentation on the paper image of the ancient book to be restored, and determining a plurality of areas to be identified.

[0032] As a possible implementation manner, step S102 may be specifically implemented as steps S1021 - S1023 .

[0033] S1021. Perform edge detection on the paper image to determine a plurality of edge points in the paper image.

[0034] In some embodiments, edge detection is a technique in image processing for identifying points in an image where brightness changes significantly, namely edge points. These edge points usually mark the boundaries between different objects or regions in the image.

[0035] Exemplarily, embodiments of the present invention may use edge detection algorithms, such as the Canny edge detector, to process paper images. The Canny algorithm detects edges by calculating the amplitude and direction of the image gradient and then applying non-maximum suppression and double threshold techniques. The algorithm parameters (such as the standard deviation, high threshold, and low threshold of the Gaussian filter) are adjusted to adapt to the specific characteristics of ancient book images, such as blurring caused by aging of paper, stains, and the diversity of handwritten text. The edge detection results are output to obtain a series of edge points that roughly outline the boundaries of the text, pattern, and background in the image.

[0036] S1022: Determine multiple text areas in the paper image based on multiple edge points.

[0037] In some embodiments, the text area is a portion of the image that contains text. In an ancient book image, the text area is usually represented by a series of connected or adjacent edge points, forming a regular or irregular arrangement.

[0038] Exemplarily, embodiments of the present invention may apply morphological operations (such as dilation and erosion) to connect adjacent edge points to form a more coherent text area outline. Use contour detection algorithms (such as line detection based on Hough transform) to identify straight line structures in the image, which often correspond to the edges of text lines. Possible text areas are screened out based on the geometric features of the contours (such as length, width, direction) and contextual information (such as the spacing between adjacent contours). Automatically identified text areas may need to be manually adjusted or verified to ensure accuracy.

[0039] S1023: Perform image segmentation based on the multiple text regions to obtain multiple regions to be identified.

[0040] In some embodiments, each area to be identified includes one or more text areas, and non-text areas adjacent to the text areas.

[0041] In some embodiments, the region to be identified is a region that is identified as requiring separate analysis or processing during the image segmentation process. In an ancient book image, the region to be identified may include one or more text regions, and non-text regions adjacent to these text regions (such as patterns, decorations, stains, etc.).

[0042] Exemplarily, an embodiment of the present invention can define an extended boundary that includes the identified text areas to include possible non-text context information (such as punctuation marks, decorative lines, etc.). Use a region-based image segmentation algorithm (such as region growing, watershed algorithm) or a graph-based segmentation algorithm (such as minimum cut / maximum flow algorithm) to further subdivide the image within the extended boundary to obtain a more refined area to be identified. Consider factors such as the arrangement of the text area (such as horizontal, vertical, tilted), font size, line spacing, etc., and adjust the parameters of the segmentation algorithm to ensure the accuracy and completeness of the area to be identified. Output the segmentation results to obtain multiple areas to be identified, each of which contains one or more text areas, as well as non-text areas adjacent to these text areas.

[0043] S103: extract features from multiple to-be-recognized regions to determine image features of each to-be-recognized region.

[0044] In some embodiments, the image features include color features, texture features, shape features, and size features.

[0045] As a possible implementation manner, step S103 may be specifically implemented as steps S1031 - S1035 .

[0046] S1031 , calculating a color histogram of each to-be-recognized area, and determining a color feature of each recognition area based on the color histogram of each to-be-recognized area.

[0047] In some embodiments, color histogram: a statistical chart showing the frequency or quantity of different colors in an image. Color feature: a property extracted from a color histogram and used to describe the color distribution of an image.

[0048] Exemplarily, the embodiment of the present invention can convert each area to be identified from the RGB color space to a space more suitable for color feature extraction, such as HSV or Lab space. In the selected color space, the color histogram of each area to be identified is calculated. The entire color space can be divided into several color intervals (bins), and then the number of pixels in each interval is counted. The color histogram is normalized to eliminate the influence of different area sizes on the color features. The main statistics of the color histogram (such as mean, variance, median, etc.) are extracted as color features.

[0049] S1032, determining the gray level co-occurrence matrix of each region to be identified, and calculating the texture features of each region to be identified based on the gray level co-occurrence matrix of each region to be identified, where the texture features include contrast, homogeneity, and correlation.

[0050] In some embodiments, Gray Level Co-occurrence Matrix (GLCM): a matrix describing the spatial relationship between gray levels in an image. Texture features: properties extracted from the gray level co-occurrence matrix used to describe image texture, including contrast, homogeneity, and correlation.

[0051] Exemplarily, the embodiment of the present invention can convert each area to be identified into a grayscale image. The grayscale co-occurrence matrix of each grayscale image is calculated, considering different directions (such as 0 degrees, 45 degrees, 90 degrees, 135 degrees) and distance parameters. Texture features such as contrast, homogeneity and correlation are extracted from the grayscale co-occurrence matrix. Contrast reflects the speed of local grayscale changes in the image, homogeneity reflects the consistency of local grayscale in the image, and correlation reflects the linear correlation between grayscale levels.

[0052] S1033, determining the text contour of each text area based on the edge points of each text area in each area to be recognized; and determining the shape features of each area to be recognized based on the text contour of each text area.

[0053] In some embodiments, text contour: the boundary line of a text area, used to describe the shape of the text. Shape feature: an attribute that describes the shape of an object in an image.

[0054] Exemplarily, the embodiment of the present invention can perform edge detection on the text area in each area to be identified to obtain the edge points of the text. Use a contour tracking algorithm (such as tracking the contour after the Zhang-Suen thinning algorithm) to determine the contour of the text area. Based on the contour points, calculate shape features, such as the perimeter, area, convex hull, circumscribed rectangle, etc. of the contour. These features can be used to describe the shape and size of the text area.

[0055] S1034. Calculate the size features of each character based on the character contours of each text area in each area to be identified; determine the size features of the non-text area based on the contours of the non-text area in each area to be identified; determine the size features of each area to be identified based on the size features of each character and the size features of the non-text area.

[0056] In some embodiments, size feature: an attribute that describes the size of an image or an object in the image.

[0057] Exemplarily, the embodiment of the present invention can calculate the average height, width, aspect ratio and other dimensional features of the text in each area to be identified based on the text contour. For non-text areas, the area, perimeter and other dimensional features are also calculated based on their contours. The dimensional features of the text area and the non-text area are combined to determine the dimensional features of the entire area to be identified. This may include the maximum / minimum size, average size, size distribution, etc. of the area.

[0058] S1035 , performing feature fusion based on the color features, texture features, shape features, and size features of each area to be identified, to obtain image features of each area to be identified.

[0059] Exemplarily, the embodiments of the present invention may fuse the color features, texture features, shape features, and size features extracted above. This can be achieved by simple concatenation, that is, connecting all feature vectors into a long vector. Before feature fusion, each feature may need to be normalized to ensure that they are comparable in value range. The fused feature vector will be used as the image feature of each area to be identified for subsequent machine learning model recognition and analysis.

[0060] S104, determining multiple recognition results for each area to be recognized based on the image features of each area to be recognized and the ancient book intelligent recognition model.

[0061] In some embodiments, the ancient book intelligent recognition model includes a time recognition module and a character recognition module. The time recognition module is used to recognize the time of the ancient book. Each time corresponds to a character recognition module. The character recognition module is used to recognize the estimated text of the corresponding time.

[0062] In some embodiments, the recognition result includes a candidate list of characters in the area to be recognized, the pronunciation and meaning of the characters, etc.

[0063] As a possible implementation manner, step S104 may be specifically implemented as steps S1041 - S1043 .

[0064] S1041. Determine the age of the ancient book to be restored based on the image features of each area to be identified and the age recognition module of the ancient book intelligent recognition model.

[0065] Exemplarily, the embodiment of the present invention can input the image features of each area to be identified into the age recognition module of the ancient book intelligent recognition model. The age recognition module may be trained using a classification algorithm (such as a support vector machine, random forest, convolutional neural network, etc.) to identify the feature differences of ancient books of different ages. According to the output of the age recognition module, combined with the comprehensive judgment of multiple areas to be identified, the approximate age of the ancient book to be repaired is determined. It may be necessary to consider majority voting, weighted average or other fusion strategies to improve the accuracy of the judgment. The age information of the ancient book to be repaired is output for subsequent screening by the glyph recognition module.

[0066] S1042. Based on the age of the ancient book to be restored, multiple glyph recognition modules of the ancient book intelligent recognition model are screened to obtain a target glyph recognition module corresponding to the age of the ancient book to be restored.

[0067] For example, the embodiment of the present invention can select a module that matches or is closest to the target age from multiple glyph recognition modules of the ancient book intelligent recognition model according to the age information of the ancient book to be restored. The glyph recognition module may be specially trained for ancient books of different ages or styles to adapt to the changes in text features in different historical periods. The selected target glyph recognition module is output for subsequent glyph recognition tasks.

[0068] S1043: Performing glyph recognition based on the image features of each area to be recognized and a target glyph recognition module to obtain a plurality of recognition results for each area to be recognized.

[0069] The multiple recognition results are multiple recognition results whose probabilities are greater than a set value among the recognition results of the to-be-recognized area.

[0070] Exemplarily, the embodiment of the present invention can input the image features of each area to be recognized into the target glyph recognition module. The target glyph recognition module may be trained using a character recognition algorithm (such as optical character recognition OCR, deep learning model, etc.) to accurately recognize the text in ancient books. According to the output of the target glyph recognition module, multiple recognition results and their corresponding probabilities for each area to be recognized are obtained. A set value is applied as a screening criterion, and only recognition results with a probability greater than the set value are retained. This helps to reduce misrecognition and improve recognition accuracy. The filtered recognition results are output, and these results can be used for subsequent tasks such as ancient book restoration, content understanding or digital preservation.

[0071] S105 , based on the multiple recognition results of each area to be recognized and the ancient book database, semantic analysis and intelligent matching analysis are performed on the context information of the ancient book to be restored, and a restoration plan for the ancient book to be restored is determined.

[0072] In some embodiments, the ancient book database stores a large amount of ancient book information (such as text, patterns, layouts, historical backgrounds, etc.) By understanding the meaning and context of the text or image content, information similar to or matching the ancient book to be restored is searched in the database.

[0073] As a possible implementation manner, step S105 may be specifically implemented as steps S1051 - S1054 .

[0074] S1051, based on multiple recognition results of each area to be recognized and the ancient book database, determine the textual semantics of each recognition result in each area to be recognized.

[0075] For example, the embodiments of the present invention can use OCR (optical character recognition) technology or deep learning models (such as convolutional neural network CNN) to perform text recognition on ancient book images to obtain preliminary recognition results. Then, combined with the vocabulary and semantic information in the ancient book database, each recognition result is semantically parsed to determine its corresponding text semantics.

[0076] S1052: Based on the text semantics of each recognition result of each area to be recognized, contextual semantic association is performed to determine multiple semantic schemes of the ancient book to be restored.

[0077] Exemplarily, after determining the semantics of each recognition result, the embodiment of the present invention can use natural language processing technology (such as semantic role labeling, dependency syntax analysis, etc.) to analyze the semantic relationship between the recognition results and establish contextual semantic associations. Based on these associations, multiple possible semantic schemes are generated, each of which represents a reasonable interpretation of the content of the ancient book.

[0078] S1053. Based on multiple semantic schemes of the ancient book to be restored, calculate the evaluation index of each semantic scheme.

[0079] In some embodiments, the evaluation criteria include semantic coherence, historical accuracy, logical rationality, and text readability.

[0080] Among them, semantic coherence: evaluates the logical relationship and coherence between the texts within the semantic scheme to ensure that the content is smooth and without contradictions. The embodiment of the present invention can grade the semantic coherence, for example: Excellent: the logical relationship between the contents is close, the connection is natural, and there is no abruptness. Good: the logical relationship between the contents is basically clear, but there may be individual unnatural connections. Average: the logical relationship between the contents is not clear enough, and there are many unnatural or abrupt connections. Poor: the logical relationship between the contents is confusing and a coherent narrative cannot be formed.

[0081] Exemplarily, the embodiments of the present invention can use text coherence assessment tools in the field of natural language processing, such as methods based on vector space models, language models, or graph neural networks, to score the coherence of semantic schemes. These tools evaluate the coherence of content by analyzing features such as vocabulary, syntax, and semantics in the text.

[0082] Alternatively, the embodiment of the present invention may also periodically invite a certain number of experts or personnel with relevant background knowledge to read and understand the semantic scheme, and then evaluate and score it based on the logical relationship between the contents, whether the connection is natural, etc.

[0083] Historical accuracy: Combined with the historical background information in the ancient book database, evaluate whether the semantic scheme is consistent with the historical background and historical facts of the ancient book. The embodiment of the present invention can classify historical accuracy, for example: Completely accurate: The content is completely consistent with the historical data and there are no errors. Basically accurate: The content is generally consistent with the historical data, but there may be errors in individual details. There are errors: There are obvious errors between the content and the historical data, but it does not affect the overall understanding. Serious errors: There are serious discrepancies between the content and the historical data, which misleads the reader's understanding.

[0084] For example, the embodiment of the present invention can compare the content in the semantic scheme with historical data, relevant documents or expert interpretations in the ancient book database to evaluate its historical accuracy. Alternatively, the embodiment of the present invention can also invite historians or ancient book experts to review the semantic scheme and judge the accuracy of the content based on their professional knowledge and experience.

[0085] Logical rationality: Analyze the logical structure in the semantic scheme to ensure that the content conforms to the logical rules and has no irrationality. The embodiment of the present invention can classify logical rationality, for example: Completely reasonable: The content logic is clear and there is no contradiction. Basically reasonable: The content logic is generally reasonable, but there may be logical problems in individual details. There are logical problems: There are obvious logical contradictions or irrationalities in the content. Logical confusion: The content logic is seriously confused and cannot form a reasonable narrative.

[0086] For example, the embodiment of the present invention can perform logical reasoning analysis on the content of the semantic scheme to check whether there are logical contradictions or illogical places. Alternatively, the embodiment of the present invention can also invite logic experts or personnel with logical reasoning ability to evaluate the semantic scheme to determine its logical rationality.

[0087] Text readability: evaluates whether the textual expression of the semantic scheme is clear, easy to understand, and conforms to reading habits. The embodiment of the present invention can grade the text readability, for example: Very easy to read: the vocabulary is simple and easy to understand, the sentence structure is clear, and it can be understood without much effort. Relatively easy to read: the vocabulary and sentence structure are relatively simple, but it may take some attention to fully understand. Somewhat difficult: the vocabulary and sentence structure are relatively complex, requiring a certain level of reading ability and attention to understand. Difficult to read: the vocabulary is uncommon, the sentence structure is complex, and it is very difficult to understand.

[0088] Exemplarily, the embodiments of the present invention can use text readability evaluation tools to score the semantic scheme. These tools usually evaluate the readability of the text based on features such as vocabulary difficulty, sentence length, and grammatical complexity. Alternatively, the embodiments of the present invention can also invite a certain number of readers to take a reading test on the semantic scheme, and then collect their feedback to evaluate its readability.

[0089] S1054. Determine a restoration plan for the ancient book to be restored based on the evaluation index of each semantic plan.

[0090] For example, the embodiment of the present invention can select the semantic scheme with the highest score as the final restoration scheme based on the evaluation index score of each semantic scheme, taking into account the weight and importance of each index. If multiple schemes have similar scores, the decision can be made by further combining expert opinions and ancient book restoration experience.

[0091] Optionally, after step S1054, the embodiment of the present invention regenerates the repair plan by interacting with the repair personnel, such as steps S1055-S1057.

[0092] S1055. Based on the restoration plan, a restoration interface of the ancient book to be restored is preliminarily generated, and the restoration interface is displayed to the restoration personnel.

[0093] In some embodiments, the post-repair interface is a visualization effect of the restored ancient book generated by simulating the restoration plan based on a preliminary restoration plan through technical means (such as image processing, text editing, etc.).

[0094] For example, the embodiment of the present invention can combine the text, typesetting and other information in the recognition results with the styles in the ancient book database based on the initially determined restoration plan by using image processing software or a text editor to generate a restored ancient book interface. The generated restored interface is displayed to the restorer through an electronic screen or printing for review and annotation.

[0095] S1056. Receive the restoration personnel's comments on the restored interface.

[0096] In some embodiments, the annotation information includes text modification, semantic correction and / or text completion of the repaired interface.

[0097] For example, in the embodiment of the present invention, when viewing the restored interface, the restorer can use annotation tools (such as electronic pens, mice, etc.) to directly perform operations such as text modification, semantic correction, or text completion on the interface. The system automatically records the restorer's annotation information and saves it in an editable file format for subsequent processing.

[0098] S1057. Based on the annotation information, re-perform semantic analysis and matching analysis to regenerate a restoration plan for the ancient book to be restored.

[0099] Exemplarily, the embodiments of the present invention can input the annotation information of the restorer into the machine learning algorithm and process it as new input data. The algorithm re-analyzes the content of the restored ancient books based on the annotation information, including semantic understanding, context association, logical verification, etc. At the same time, combined with the information in the ancient book database, the restored ancient book content is matched and analyzed to verify its historical accuracy, logical rationality and text readability. According to the results of the re-analysis, the restoration plan is revised and optimized to generate a more accurate restoration plan.

[0100] S106. Repairing the ancient book to be repaired based on the repair plan of the ancient book to be repaired.

[0101] As a possible implementation manner, step S106 may be specifically implemented as steps S1061 - S1065 .

[0102] S1061. Determine the background color and text color of the ancient book to be restored based on the age of the ancient book to be restored and the current color of the ancient book to be restored.

[0103] For example, the embodiment of the present invention can preliminarily determine the possible background color and text color range based on the age information of the ancient book to be restored and the historical color data in the ancient book database. Then, combined with the current color information of the ancient book to be restored, the background color and text color can be further accurately determined through a color matching algorithm.

[0104] S1062. Determine the glyph data of the ancient book to be restored based on the restoration plan of the ancient book to be restored.

[0105] For example, the embodiment of the present invention can retrieve corresponding glyph data from the ancient book database according to the age and style of the ancient book to be restored. If there is no directly matching glyph data in the database, a machine learning algorithm can be used to transfer or generate the style of the text in the ancient book to obtain glyph data that conforms to the style of the ancient book.

[0106] S1063, performing digital rendering based on the glyph data, as well as the background color and the text color, to obtain a digital interface of the ancient book to be restored.

[0107] For example, the embodiment of the present invention can use digital rendering technology to combine the determined background color, text color and font data to generate a digital interface of the ancient book to be restored. During the rendering process, it is necessary to ensure that the contrast between the text and the background is moderate for subsequent recognition and restoration.

[0108] S1064, performing feature extraction based on the digitized interface of the ancient book to be restored, and obtaining image features of each area to be identified in the digitized interface.

[0109] Exemplarily, the embodiment of the present invention can use a feature extraction algorithm to extract image features of each area to be identified from the digitized interface. These features may include color features, texture features, shape features, etc. The purpose of feature extraction is to subsequently match the features in the paper image to determine the restoration method and restoration content.

[0110] S1065: Determine the restoration method and restoration content of each area to be identified based on the image features of each area to be identified in the paper image and the image features of each area to be identified in the digitized interface.

[0111] Exemplarily, the embodiment of the present invention can match the image features of each area to be identified in the paper image with the corresponding image features in the digitized interface. According to the matching results, the specific conditions of the area to be repaired are analyzed, such as blurred, missing, or contaminated text. According to the analysis results, the repair method of each area to be identified is determined, such as digital repair, manual repair, etc., and the specific repair content is clarified, such as replacing blurred text, completing missing text, etc.

[0112] The present invention provides an ancient book intelligent recognition and restoration method based on machine learning. The method extracts image features of each to-be-recognized area in a paper image of the ancient book to be restored, inputs the features into an ancient book intelligent recognition model, performs intelligent recognition, and obtains multiple recognition results of each to-be-recognized area. Then, based on the multiple recognition results, context semantic analysis and intelligent matching are performed in combination with an ancient book database to determine a restoration plan for the ancient book to be restored, and the ancient book to be restored is restored. Automatic recognition of the ancient book to be restored and automatic generation of the restoration plan are achieved, the professional requirements of the restorer are reduced, the restoration time of the restorer is reduced, and the recognition and restoration efficiency of the ancient books is improved.

[0113] Optionally, the ancient book intelligent identification and restoration method based on machine learning provided in the embodiment of the present invention further includes steps S201-S205 before step S104.

[0114] S201, obtaining image features and text data of a plurality of ancient books of determined age and font shape.

[0115] Exemplarily, the embodiments of the present invention can collect ancient book samples of known age and fonts, which should cover multiple historical periods to ensure the comprehensiveness of the training. Each ancient book sample is scanned or photographed at high resolution to obtain its image data. The features of the ancient book image are extracted using image processing techniques such as edge detection and texture analysis. The text data in the ancient book is obtained through OCR (optical character recognition) technology or manual transcription.

[0116] S202: Generate a first training sample based on the image data and age of each ancient book.

[0117] In some embodiments, the first training sample is a training data set consisting of image data of multiple ancient books and corresponding age information, which is used to train the age recognition module of the ancient book intelligent recognition model.

[0118] For example, the embodiment of the present invention can combine the image data of each ancient book and its corresponding age information into a training sample, and ensure that the ages of the ancient books in the training sample are evenly distributed to avoid the model's preference for a specific age.

[0119] S203: Generate second training samples of each era based on the image features and text data of each ancient book in each era.

[0120] In some embodiments, the second training samples are a training data set consisting of image features of ancient books and corresponding text data, which are classified by era and used to train the character recognition modules of each era.

[0121] For example, the embodiment of the present invention can generate training samples for ancient book samples in each era according to their image features and text data, ensuring that the number of training samples in each era is sufficient to improve the accuracy of the character recognition module.

[0122] S204: Based on the first training sample, neural network training is performed to obtain a time recognition module of the ancient book intelligent recognition model.

[0123] For example, in the embodiment of the present invention, a suitable neural network architecture, such as a convolutional neural network (CNN), can be selected to train the age recognition module. The first training sample is input into the neural network, and forward propagation and back propagation are performed, and the network parameters are adjusted to minimize the loss function. Through multiple iterations of training, until the network converges, a trained age recognition module is obtained.

[0124] S205 , performing neural network training based on the second training samples of each decade to obtain a character recognition module of each decade.

[0125] Exemplarily, the embodiment of the present invention can select a suitable neural network architecture, such as a recurrent neural network (RNN) or a convolutional neural network (CNN), for each decade to train the glyph recognition module of that decade. The second training sample of the corresponding decade is input into the neural network, and forward propagation and back propagation are performed, and the network parameters are adjusted to minimize the loss function. Through multiple iterations of training, until the glyph recognition modules of each decade converge, the trained glyph recognition modules of each decade are obtained.

[0126] In this way, the embodiment of the present invention can pre-train the ancient book intelligent recognition model, realize the intelligent recognition of ancient books, and improve the recognition and repair efficiency of ancient books. In addition, the present invention trains ancient books of different ages separately, effectively distinguishes the glyphs and characters of different historical periods, and improves the accuracy of intelligent recognition of ancient books.

[0127] Optionally, the ancient book intelligent identification and restoration method based on machine learning provided in an embodiment of the present invention further includes steps S301-S305.

[0128] S301, obtaining a restored image and a pre-restoration image of an artificially restored ancient book.

[0129] For example, the embodiments of the present invention can collect images before and after manual restoration from ancient book restoration projects to ensure that the images are clear, complete and easy to compare. The images are pre-processed, such as denoising, contrast enhancement, etc., to improve the accuracy of subsequent analysis.

[0130] S302: Determine a restoration plan based on the pre-restoration image, the ancient book intelligent recognition model, and the ancient book database.

[0131] For example, the embodiments of the present invention can use the ancient book intelligent recognition model to analyze the image before restoration, identify the text content, glyph features, etc. Combined with the historical information and style features in the ancient book database, one or more possible restoration plans are generated. These plans will serve as a benchmark for comparison with the manual restoration results.

[0132] S303: Based on the restoration plan and the restored image, a comparison is performed to determine the difference between the manual restoration and the intelligent recognition of each area in the manual restoration of the ancient book.

[0133] In some embodiments, difference degree: an indicator for measuring the degree of inconsistency between a manual repair result and an intelligent recognition repair solution, which is usually calculated by comparing the differences between the two in terms of text content, font shape, layout, etc.

[0134] For example, the embodiments of the present invention can compare the generated restoration scheme with the manually restored image, focusing on the differences in text content, font, layout, etc. These differences are quantified using image processing technology and machine learning algorithms to calculate a difference index. The difference can be obtained by calculating similarity, error rate or other suitable metrics.

[0135] S304: Generate restoration quality inspection information based on the difference between manual restoration and intelligent recognition in each area.

[0136] In some embodiments, the repair quality inspection information includes manually repairing multiple disputed areas where there are disputes, and the disputed content of each disputed area.

[0137] In some embodiments, disputed areas: areas where there are significant differences between manual repair and intelligent recognition, which may require further review or repair. Disputed content: differences in specific text, fonts, or layout within the disputed area.

[0138] For example, the embodiment of the present invention can identify disputed areas with significant differences based on the difference index. Perform a detailed analysis on each disputed area and extract disputed content, such as text errors, inconsistent fonts, chaotic layout, etc. Organize the disputed areas and disputed content into a repair quality inspection information report for subsequent review and processing.

[0139] S305: Present restoration quality inspection information to restoration personnel.

[0140] For example, the embodiments of the present invention can present the restoration quality inspection information to the restoration personnel through electronic screens, printed reports or online platforms, etc., to ensure that the information is clear and intuitive, so that the restoration personnel can understand and deal with the disputed areas. Necessary explanations and guidance are provided to help the restoration personnel understand the disputed content and its possible causes.

[0141] In this way, the embodiment of the present invention can verify and quality-check the manually restored ancient books through intelligent recognition and semantic analysis of ancient books, and present the disputed areas and response content to the restorers for them to browse and finally confirm, thereby improving the accuracy of ancient book recognition and restoration.

[0142] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0143] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0144] Figure 2 The schematic diagram of the structure of an ancient book intelligent identification and restoration device based on machine learning provided by an embodiment of the present invention is shown. The restoration device 400 includes a communication module 401 and a processing module 402 .

[0145] The communication module 401 is used to obtain the paper image of the ancient book to be restored.

[0146] The processing module 402 is used to perform image segmentation on the paper image of the ancient book to be repaired, and determine multiple areas to be identified; perform feature extraction on the multiple areas to be identified, and determine the image features of each area to be identified; the image features include color features, texture features, shape features and size features; based on the image features of each area to be identified and the ancient book intelligent recognition model, determine multiple recognition results for each area to be identified; based on the multiple recognition results of each area to be identified and the ancient book database, perform semantic analysis and intelligent matching analysis on the context information of the ancient book to be repaired, and determine the restoration plan for the ancient book to be repaired; based on the restoration plan of the ancient book to be repaired, repair the ancient book to be repaired.

[0147] Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, the electronic device 500 includes: a processor 501, a memory 502, and a computer program 503 stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program 503, the steps in the above-mentioned method embodiments are implemented, for example Figure 1 Alternatively, when the processor 501 executes the computer program 503, the functions of the modules / units in the above-mentioned device embodiments are realized, for example, Figure 2 The functions of the communication module 401 and the processing module 402 are shown.

[0148] Exemplarily, the computer program 503 may be divided into one or more modules / units, which are stored in the memory 502 and executed by the processor 501 to implement the present invention. The one or more modules / units may be capable of implementing a particular Figure 3 A series of computer program instruction segments with certain functions, which are used to describe the execution process of the computer program 503 in the electronic device 500. For example, the computer program 503 can be divided into Figure 2 A communication module 401 and a processing module 402 are shown.

[0149] The processor 501 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0150] The memory 502 may be an internal storage unit of the electronic device 500, such as a hard disk or memory of the electronic device 500. The memory 502 may also be an external storage device of the electronic device 500, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 500. Further, the memory 502 may also include both an internal storage unit of the electronic device 500 and an external storage device. The memory 502 is used to store the computer program and other programs and data required by the terminal. The memory 502 may also be used to temporarily store data that has been output or is to be output.

[0151] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for intelligent identification and restoration of ancient books based on machine learning, characterized in that: include: Obtain paper images of ancient books to be restored; Performing image segmentation on the paper image of the ancient book to be restored to determine a plurality of areas to be identified; Extracting features from the multiple to-be-identified regions to determine image features of each to-be-identified region; the image features include color features, texture features, shape features, and size features; Based on the image features of each area to be identified and the ancient book intelligent recognition model, multiple recognition results of each area to be identified are determined; Based on the multiple recognition results of each area to be recognized and the ancient book database, semantic analysis and intelligent matching analysis are performed on the context information of the ancient book to be restored to determine the restoration plan of the ancient book to be restored; Based on the restoration plan of the ancient book to be restored, the ancient book to be restored is restored.

2. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: The step of segmenting the paper image of the ancient book to be restored to determine a plurality of areas to be identified includes: Performing edge detection on the paper image to determine a plurality of edge points in the paper image; Determine a plurality of text areas in the paper image based on the plurality of edge points; Image segmentation is performed based on the multiple text regions to obtain multiple regions to be identified; each region to be identified includes one or more text regions and non-text regions adjacent to the text regions.

3. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: The extracting features of the plurality of to-be-identified regions to determine the image features of each to-be-identified region includes: Calculating the color histogram of each area to be identified, and determining the color characteristics of each identification area based on the color histogram of each area to be identified; Determine a gray level co-occurrence matrix of each region to be identified, and calculate texture features of each region to be identified based on the gray level co-occurrence matrix of each region to be identified, wherein the texture features include contrast, homogeneity and correlation; Based on the edge points of each text area in each area to be identified, determine the text contour of each text area; based on the text contour of each text area, determine the shape features of each area to be identified; Calculate the size features of each character based on the character contours of each character area in each area to be identified; determine the size features of the non-character area based on the contours of the non-character area in each area to be identified; determine the size features of each area to be identified based on the size features of each character and the size features of the non-character area; Based on the color features, texture features, shape features and size features of each area to be identified, feature fusion is performed to obtain the image features of each area to be identified.

4. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: The method of determining multiple recognition results of each area to be recognized based on the image features of each area to be recognized and the ancient book intelligent recognition model includes: Determine the age of the ancient book to be restored based on the image features of each area to be identified and the age recognition module of the ancient book intelligent recognition model; Based on the age of the ancient book to be restored, multiple glyph recognition modules of the ancient book intelligent recognition model are screened to obtain a target glyph recognition module corresponding to the age of the ancient book to be restored; Based on the image features of each area to be recognized and the target character recognition module, character recognition is performed to obtain multiple recognition results for each area to be recognized; wherein the multiple recognition results are multiple recognition results whose probabilities are greater than a set value among the recognition results of the area to be recognized.

5. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: Before determining multiple recognition results of each area to be recognized based on the image features of each area to be recognized and the ancient book intelligent recognition model, the method further includes: Obtain image features and text data of multiple ancient books with confirmed ages and font shapes; Generate a first training sample based on the image data and age of each ancient book; Based on the image features and text data of each ancient book in each era, a second training sample of each era is generated; Based on the first training sample, neural network training is performed to obtain a time recognition module of the ancient book intelligent recognition model; Based on the second training samples of each decade, neural network training is performed to obtain the glyph recognition modules of each decade.

6. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: Based on the multiple recognition results of each area to be recognized and the ancient book database, semantic analysis and intelligent matching analysis are performed on the context information of the ancient book to be restored to determine the restoration plan of the ancient book to be restored, including: Determine the text semantics of each recognition result in each area to be recognized based on multiple recognition results in each area to be recognized and the ancient book database; Based on the text semantics of each recognition result in each area to be recognized, contextual semantic association is performed to determine multiple semantic schemes for the ancient books to be restored; Based on multiple semantic schemes of the ancient books to be restored, calculating evaluation indicators for each semantic scheme, the evaluation indicators include semantic coherence, historical accuracy, logical rationality and text readability; Based on the evaluation indicators of each semantic scheme, the restoration plan for the ancient books to be restored is determined.

7. The ancient book intelligent identification and restoration method based on machine learning according to claim 6 is characterized in that: Based on the multiple recognition results of each area to be recognized and the ancient book database, the context information of the ancient book to be restored is subjected to semantic analysis and intelligent matching analysis to determine the restoration plan of the ancient book to be restored, and further includes: Based on the restoration plan, a restoration interface of the ancient book to be restored is preliminarily generated, and the restoration interface is displayed to the restoration personnel; Receiving annotation information on the repaired interface from the repair personnel; the annotation information includes text modification, semantic correction and / or text completion on the repaired interface; Based on the annotation information, semantic analysis and matching analysis are re-performed to regenerate a repair plan for the ancient book to be repaired.

8. The ancient book intelligent identification and restoration method based on machine learning according to claim 1 is characterized in that: The repair scheme based on the ancient book to be repaired, repairing the ancient book to be repaired, includes: Determining the background color and text color of the ancient book to be restored based on the age of the ancient book to be restored and the current color of the ancient book to be restored; Determining the glyph data of the ancient book to be restored based on the restoration plan of the ancient book to be restored; Based on the glyph data, as well as the background color and the text color, digital rendering is performed to obtain a digital interface of the ancient book to be restored; Based on the digitized interface of the ancient book to be restored, feature extraction is performed to obtain image features of each area to be identified in the digitized interface; Based on the image features of each area to be identified in the paper image and the image features of each area to be identified in the digitized interface, the restoration method and restoration content of each area to be identified are determined.

9. The ancient book intelligent identification and restoration method based on machine learning according to any one of claims 1 to 8, characterized in that: The method further comprises: Obtaining images of artificially restored ancient books after restoration and before restoration; Determine a restoration plan of manual restoration estimation based on the pre-restoration image, the ancient book intelligent recognition model, and the ancient book database; Based on the restoration scheme and the restored image, a comparison is performed to determine the difference between the manual restoration and the intelligent recognition of each area in the manual restoration of the ancient book; Generate restoration quality inspection information based on the difference between manual restoration and intelligent recognition in each area, wherein the restoration quality inspection information includes multiple disputed areas where manual restoration is disputed, and the disputed content of each disputed area; The restoration quality inspection information is presented to the restoration personnel.

10. An ancient book intelligent recognition and restoration device based on machine learning, characterized in that: include: A communication module, used to obtain a paper image of the ancient book to be restored; A processing module is used to perform image segmentation on the paper image of the ancient book to be restored to determine a plurality of regions to be identified; perform feature extraction on the plurality of regions to be identified to determine image features of each region to be identified; the image features include color features, texture features, shape features and size features; Based on the image features of each area to be identified and the ancient book intelligent recognition model, multiple recognition results of each area to be identified are determined; based on the multiple recognition results of each area to be identified and the ancient book database, semantic analysis and intelligent matching analysis are performed on the context information of the ancient book to be restored to determine the restoration plan of the ancient book to be restored; Based on the restoration plan of the ancient book to be restored, the ancient book to be restored is restored.

Images