PDF (Portable Document Format) scanning copy identification method and equipment and medium
Through multi-model collaborative recognition and image classification processing, the problem of low recognition accuracy of OCR technology in complex layouts and different font PDF files is solved, achieving higher recognition accuracy and text recognition effect.
Patent Information
- Application Number
- CN202411051353.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-19
AI Technical Summary
The existing OCR technology is insufficient in identifying PDF files with complex layouts and different fonts, resulting in low recognition accuracy.
The multi-model collaborative recognition method is used to split the PDF file into plain text, complex layout and natural pictures, and input the most suitable recognition model separately, and improve the recognition accuracy through image classification and correction processing.
It significantly improves the recognition accuracy and text recognition effect of complex documents, overcomes the limitations of single scene recognition, and improves the recognition accuracy of PDF scanned documents.
Smart Images

Figure CN118982838A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a PDF scan recognition method, device and medium. Background Art
[0002] With the rise of the digital age, a large number of paper documents are gradually being converted into electronic form, among which PDF files are widely popular due to their cross-platform compatibility and stability of content presentation. However, when scanning and generating PDF files, especially when containing text in image form, it is difficult to edit and retrieve the text content.
[0003] In order to overcome this problem, Optical Character Recognition (OCR) technology came into being. OCR technology can convert text in an image into a computer-editable text format through the principles of image processing and pattern recognition. Due to the variety of data types in PDF files, OCR technology is not versatile enough in recognizing complex layouts and different fonts, resulting in a low accuracy rate for the recognized file content. Summary of the invention
[0004] The embodiments of the present application provide a PDF scan recognition method, device and medium for solving the following technical problems: Due to the variety of data types in PDF files, OCR technology is not versatile enough in recognizing complex layouts and different fonts, resulting in low accuracy of the recognized file content.
[0005] The present application embodiment adopts the following technical solutions:
[0006] The embodiment of the present application provides a PDF scan recognition method. The method comprises: splitting a PDF file to be recognized to obtain multiple page images, and encoding and marking the multiple page images; classifying the multiple page images based on the page content; wherein the image categories include pure text images, complex layout images, and natural images; inputting the classified page images into corresponding information recognition models respectively, and outputting corresponding information content through multi-model collaboration; wherein the information content at least includes the recognized data corresponding to the page image and the position of the recognized data in the page image; based on the marked codes and information content, combining the information contents corresponding to the multiple page images to obtain the word document corresponding to the PDF file to be recognized.
[0007] The embodiment of the present application improves the accuracy of PDF scan recognition by processing different types of PDF scan files and using multiple models for collaborative recognition. In view of the limitations of single-scene text recognition in the prior art, the embodiment of the present application introduces an image classification mechanism to split the scans into multiple types such as plain text, complex layouts, and natural scenes, and adopts the most suitable recognition strategy for different types of images, thereby improving the accuracy of text recognition. In addition, the embodiment of the present application selects the optimal recognition model for each element in the image, realizes collaborative work between models, and significantly improves the recognition effect of complex documents.
[0008] In one implementation of the present application, a PDF file to be identified is split into pages to obtain multiple page images, and the multiple page images are encoded and labeled, specifically including: splitting the PDF file to be identified into pages, and converting the split PDF into image format; performing image correction processing on the PDF file converted into image format; wherein the image correction processing includes at least one of image denoising, image deformation correction, image shadow and highlight correction, image watermark removal, and image moiré removal; based on the PDF file after the correction processing, multiple page images are obtained; the page order of the PDF file to be identified is determined, the multiple page images are sorted based on the page order, and the encoding and labeling are re-performed.
[0009] In one implementation of the present application, before classifying multiple page images based on page content, the method also includes: training a preset neural network model based on the acquired image sample data and the image categories corresponding to the image sample data to obtain an image classification model; classifying multiple page images based on the page content, specifically including: inputting the page images into the image classification model to output the image categories corresponding to the multiple page images through the image classification model.
[0010] In one implementation of the present application, the classified page images are respectively input into corresponding information recognition models, and the corresponding information content is output through multi-model collaboration, specifically including: when the page image is a plain text image, removing the header and footer in the plain text image; performing layout analysis on the plain text image to determine the column division of the text in the plain text image; inputting the image parts corresponding to each column into a preset general model in turn, and outputting the recognized text information corresponding to each column through the preset general model.
[0011] In one implementation of the present application, the classified page images are respectively input into corresponding information recognition models, and corresponding information content is output through multi-model collaboration, specifically including: when the page image is a complex layout image, the header and footer in the complex layout image are removed; the complex layout image is analyzed to determine the data type in the complex layout image; wherein the data type includes at least one of text data, title data, table data, seal data and image data; based on the pixel point information corresponding to the data of each data type, the positions corresponding to different types of data in the complex layout are determined; the data of each data type are respectively input into the corresponding preset different data type recognition models to output the corresponding recognized data through the preset different data type recognition models; based on the position and the recognized data, the recognized information content corresponding to the complex layout image is obtained.
[0012] In one implementation of the present application, the classified page images are respectively input into corresponding information recognition models, and the corresponding information content is output through multi-model collaboration, specifically including: when the page image is a natural image, the header and footer in the natural image are removed; text information is detected for the natural image, and when there is no text in the natural image, the natural image is not recognized; when there is text in the natural image, content information detection is performed on the PDF file to be recognized to determine whether the text in the natural image is text to be recognized or text that does not need to be recognized; when it is text to be recognized, the natural image is input into a preset natural scene recognition model, and the recognized information content is output through the preset natural scene recognition model.
[0013] In one implementation of the present application, when there is text in a natural image, content information detection is performed on a PDF file to be identified to determine whether the text in the natural image is text to be identified or text that does not need to be identified, specifically including: determining the format of the text in the natural image, comparing the format with a preset format template, and obtaining a first detection result based on the comparison result; extracting content from the PDF file to be identified, determining text content adjacent to the natural image, performing keyword extraction on the text content, comparing the extracted keywords with preset keywords, and obtaining a second detection result based on the comparison result; when the first detection result and / or the second detection result is text to be identified, the natural image is identified.
[0014] In one implementation of the present application, the information contents of multiple page images are combined based on the marked codes and information contents to obtain a word document corresponding to the PDF file to be identified, specifically including: arranging the identified information contents corresponding to each page image based on the position of the data in the page image; sorting the identified pages after the content arrangement based on the marked codes; and combining the sorted and identified pages to obtain a word document corresponding to the PDF file to be identified.
[0015] An embodiment of the present application provides a PDF scan recognition device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: split the PDF file to be recognized into pages to obtain multiple page images, and encode and annotate the multiple page images; classify the multiple page images based on page content; wherein the image categories include plain text images, complex layout images, and natural images; input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration; wherein the information content at least includes recognized data corresponding to the page image and the position of the recognized data in the page image; based on the annotated codes and information content, combine the information contents corresponding to the multiple page images to obtain a word document corresponding to the PDF file to be recognized.
[0016] A non-volatile computer storage medium provided in an embodiment of the present application stores computer executable instructions, wherein the computer executable instructions are configured to: split a PDF file to be identified into pages to obtain a plurality of page images, and encode and annotate the plurality of page images; classify the plurality of page images based on page content; wherein the image categories include pure text images, complex layout images, and natural images; input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration; wherein the information content includes at least identified data corresponding to the page image and the position of the identified data in the page image; and combine the information contents corresponding to the plurality of page images based on the annotated codes and information contents to obtain a word document corresponding to the PDF file to be identified.
[0017] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application improve the accuracy of PDF scan recognition by processing different types of PDF scan files and using multiple models for collaborative recognition. In view of the limitations of single-scene text recognition in the prior art, the embodiments of the present application introduce an image classification mechanism to split the scans into multiple types such as plain text, complex layouts, natural scenes, etc., and adopt the most suitable recognition strategy for different types of images, thereby improving the accuracy of text recognition. In addition, the embodiments of the present application select the optimal recognition model for each element in the image, realize the collaborative work between models, and significantly improve the recognition effect of complex documents. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.
[0019] In the figure:
[0020] Figure 1 A flow chart of a PDF scan recognition method provided in an embodiment of the present application;
[0021] Figure 2 A flowchart of a PDF scan recognition process provided in an embodiment of the present application;
[0022] Figure 3 A schematic diagram of the structure of a PDF scan recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0023] Embodiments of the present application provide a PDF scan recognition method, device, and medium.
[0024] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.
[0025] The technical solution proposed in the embodiment of the present invention is described in detail below with reference to the accompanying drawings.
[0026] Figure 1A flow chart of a PDF scan recognition method provided in an embodiment of the present application, such as Figure 1 As shown, the PDF scan recognition method includes the following steps:
[0027] S101, splitting the PDF file to be recognized into pages to obtain multiple page images, and encoding and marking the multiple page images.
[0028] In one embodiment of the present application, the PDF file to be identified is de-paged, and the de-paged PDF is converted into an image format. The PDF file converted into an image format is subjected to image correction processing; wherein the image correction processing includes at least one of image denoising, image deformation correction, image shadow and highlight correction, image watermark removal, and image moiré removal. Based on the PDF file after the correction processing, multiple page images are obtained. The page order of the PDF file to be identified is determined, and the multiple page images are sorted based on the page order, and re-encoded and annotated.
[0029] Specifically, the input scanned PDF file is preprocessed by first splitting the PDF file into pages, and performing operations such as denoising, deformation correction, shadow and highlight correction, watermark removal, and moiré removal on the image, thereby improving the quality of the input data.
[0030] Furthermore, first, the PDF file is split into separate pages, and each page is extracted as an independent unit. Then, these PDF pages are converted into image formats, such as JPEG, PNG, etc. The image correction processing in the embodiment of the present application is to improve the image quality, for example, image denoising, removing noise in the image; image deformation correction, correcting the distortion caused by scanning or shooting; image shadow highlight correction, adjusting the dark or bright areas in the image; image watermark removal, removing the watermark in the image; image moiré removal, eliminating the moiré caused by the interaction between the camera sensor and the periodic structure such as the screen, etc.
[0031] Furthermore, after the image correction is completed, the converted and processed multiple page images are sorted according to the page order of the original PDF file and re-encoded and annotated.
[0032] S102: Classify multiple page images based on page content.
[0033] In one embodiment of the present application, a preset neural network model is trained based on the acquired image sample data and the image categories corresponding to the image sample data to obtain an image classification model. Based on the page content, multiple page images are classified, specifically including: inputting the page images into the image classification model to output the image categories corresponding to the multiple page images through the image classification model.
[0034] Specifically, image sample data is obtained, wherein the obtained image sample data is a picture that has been marked (i.e., its category is known), and the prepared image sample data and its corresponding image category are input into a preset neural network model. The model minimizes the difference between the predicted category and the actual category by adjusting its internal parameters (weights and biases). After the training results meet the requirements, the image classification model is obtained.
[0035] Furthermore, the page images are input into the trained image classification model, and the model will output the predicted category corresponding to each page image. The image categories include pure text images, complex layout images, and natural images.
[0036] S103: Input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration.
[0037] In one embodiment of the present application, when the page image is a plain text image, the header and footer in the plain text image are removed. The plain text image is analyzed for layout to determine the column division of the text in the plain text image. The image parts corresponding to each column are sequentially input into the preset general model, and the recognized information content corresponding to each column is output through the preset general model.
[0038] The information content at least includes the recognized data corresponding to the page image and the position of the recognized data in the page image.
[0039] Specifically, the header and footer areas in the image are identified through image analysis technology, and they are removed from the image through operations such as cropping. Furthermore, the layout analysis of the plain text image is performed to determine the column division of the text in the plain text image. For example, the edge detection algorithm is used to identify the dividing line between text blocks, the image is projected horizontally or vertically, and the changes in the projection curve are analyzed to determine the column division of the text. Suppose the text in a plain text image is divided into two columns. Through layout analysis, the dividing line between the two columns is identified, and the image is divided into two areas containing their respective text contents.
[0040] Furthermore, after determining the columns of the text, the image portion corresponding to each column is input into a preset general model, such as an OCR model, for text recognition. These models are trained to recognize text in images and convert them into editable text format.
[0041] In one embodiment of the present application, when the page image is a complex layout image, the header and footer in the complex layout image are removed. The complex layout image is subjected to layout analysis to determine the data type in the complex layout image; wherein the data type includes at least one of text data, title data, table data, seal data, and image data. Based on the pixel point information corresponding to the data of each data type, the positions corresponding to different types of data in the complex layout are determined. The data of each data type is respectively input into the corresponding preset different data type recognition model, so as to output the corresponding recognized data through the preset different data type recognition model. Based on the position and the recognized data, the recognized information content corresponding to the complex layout image is obtained.
[0042] Specifically, when the page image is a complex layout image, the header and footer are identified, and these areas are removed by cropping or masking. The complex layout image is analyzed to determine the data type in the complex layout image. For example, a machine learning or deep learning model is applied to identify the data type of each area, such as text, title, table, seal, image, etc. The pixel points corresponding to each data type in the image are parsed, the bounding box or outline of each data type in the image is marked, and the coordinate information of these bounding boxes or outlines is recorded.
[0043] Furthermore, different types of data are input into preset recognition models for different data types, for example, using an OCR model to recognize text data, applying a table recognition algorithm to extract table data, and using image processing techniques such as color matching and shape recognition to recognize stamps and image data.
[0044] Furthermore, based on the location information of the data type and the recognized data content, all data are integrated together to form the complete recognized information content corresponding to the complex layout image. Specifically, based on the location information of the data type, the original layout is reconstructed in the final document, and the recognized data content is inserted into the corresponding position. Based on the recognized text, table, seal and image data, the layout and content of the original file are reconstructed in the electronic document, and the user can perform further editing and processing in the electronic document.
[0045] In one embodiment of the present application, when the page image is a natural image, the header and footer in the natural image are removed. Text information detection is performed on the natural image, and when there is no text in the natural image, the natural image is not recognized. When there is text in the natural image, content information detection is performed on the PDF file to be recognized to determine whether the text in the natural image is text to be recognized or text that does not need to be recognized. In the case of text to be recognized, the natural image is input into a preset natural scene recognition model, and the recognized information content is output through the preset natural scene recognition model.
[0046] Specifically, when the page image is a natural image, the header and footer are identified, and these areas are removed by cropping or masking. When processing natural images, it is detected whether there is text in the image. After confirming that there is no text in the image, the natural image is not recognized. After confirming that there is text in the image, it is determined whether the text is content that needs to be recognized. For example, decisions can be made based on context, a predefined set of rules, or user input.
[0047] Furthermore, if it is determined that the text in the image is content that needs to be recognized, the image is input into a pre-trained natural scene recognition model, which analyzes the text area in the image and converts it into readable text information.
[0048] In one embodiment of the present application, the format of the text in the natural image is determined, the format is compared with a preset format template, and a first detection result is obtained based on the comparison result. Content is extracted from the PDF file to be identified, the text content adjacent to the natural image is determined, keywords are extracted from the text content, the extracted keywords are compared with preset keywords, and a second detection result is obtained based on the comparison result. If the first detection result and / or the second detection result is the text to be identified, the natural image is identified.
[0049] Specifically, identify whether there is text in the natural picture and determine the format of the text. The format of the text may include font, font size, color, arrangement, such as horizontal, vertical or tilted. Compare the extracted text format with the preset format template. The preset format template contains a series of expected format specifications, which are used to determine whether the recognized text meets the requirements or standards of the text to be recognized. Based on the comparison result with the preset format template, a first detection result is obtained. If the extracted text format completely matches the preset format template, then the first detection result is "yes", indicating that the text in the natural picture needs to be recognized.
[0050] Furthermore, content extraction is performed on the PDF file containing the natural image, especially the text content adjacent to or related to the natural image. Keyword extraction is performed on the extracted text content, and these keywords are compared with a preset keyword list. The preset keyword list contains words related to the image, for example: as shown in the figure below, as shown in the figure, etc. Based on the comparison result of the keywords, a second detection result is obtained. If the text content adjacent to the natural image contains the keyword "as shown in the figure", then the second detection result is "yes", indicating that this text may be related to the text in the natural image, and the text in the natural image needs to be identified.
[0051] Further, based on the first detection result and the second detection result, it is comprehensively determined whether the text in the natural image is the text to be recognized. If any detection result is "yes", or both are "yes", the text in the natural image is recognized.
[0052] S104: Based on the marked codes and information contents, the information contents corresponding to the multiple page images are combined to obtain a word document corresponding to the PDF file to be identified.
[0053] In one embodiment of the present application, the recognized information contents corresponding to each page image are arranged based on the position of the data in the page image. The recognized pages after the content arrangement are sorted based on the marked codes. The sorted and recognized pages are combined to obtain the word document corresponding to the PDF file to be recognized.
[0054] Specifically, according to the layout analysis results and the text recognition results, the entire document is reconstructed to form a word document.
[0055] Figure 2 This is a flowchart of a PDF scan recognition process provided by an embodiment of the present application. Figure 2 As shown in the figure, the PDF scan recognition process includes:
[0056] The first step is to preprocess the input scanned PDF file. First, the PDF file is split into pages, and operations such as image denoising, deformation correction, shadow and highlight correction, watermark removal, and moiré removal are performed to improve the quality of the input data.
[0057] In the second step, the images in the document are encoded in sequence, and the images are rotated, corrected and classified into plain text, complex layouts, natural scenes, etc.
[0058] The third step is to analyze the layout of pure text images, remove headers and footers, determine the text columns, use a general model to recognize the text, and return the recognition results in the correct order.
[0059] The fourth step is to analyze the layout of complex images, remove headers and footers, and find the locations of titles, texts, images, tables, seals, etc. in the images. These are then sent to different models for text recognition and the recognition results are output with the correct locations and order.
[0060] The fifth step is to determine whether natural images contain text. By analyzing the entire document, determine whether recognition is required. Then, the image to be recognized is put into the natural scene for model processing and the correct recognition result is returned.
[0061] The sixth step is to reconstruct the entire document according to the layout analysis results and text recognition results to form a word document.
[0062] Figure 3 This is a schematic diagram of the structure of a PDF scan recognition device provided in an embodiment of the present application. Figure 3 As shown, the PDF scan recognition device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: split the PDF file to be recognized into pages to obtain multiple page images, and encode and annotate the multiple page images; classify the multiple page images based on the page content; wherein the image categories include plain text images, complex layout images, and natural images; input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration; wherein the information content at least includes recognized data corresponding to the page image and the position of the recognized data in the page image; based on the annotated code and information content, combine the information content corresponding to the multiple page images to obtain the word document corresponding to the PDF file to be recognized.
[0063] The embodiment of the present application also provides a non-volatile computer storage medium, storing computer executable instructions, wherein the computer executable instructions are configured to: split a PDF file to be identified into pages to obtain multiple page images, and encode and annotate the multiple page images; classify the multiple page images based on page content; wherein the image categories include pure text images, complex layout images, and natural images; input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration; wherein the information content at least includes recognized data corresponding to the page image and the position of the recognized data in the page image; based on the annotated codes and information content, combine the information contents corresponding to the multiple page images to obtain a word document corresponding to the PDF file to be identified.
[0064] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0065] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various changes and variations. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.
Claims
1. A PDF scan recognition method, characterized in that: The method comprises: Splitting the PDF file to be identified into pages to obtain multiple page images, and encoding and marking the multiple page images; Based on the page content, the plurality of page images are classified; wherein the image categories include pure text images, complex layout images, and natural images; Input the classified page images into corresponding information recognition models respectively, and output corresponding information content through multi-model collaboration; wherein the information content at least includes the recognized data corresponding to the page image and the position of the recognized data in the page image; Based on the marked codes and the information content, the information contents corresponding to the multiple page images are combined to obtain the word document corresponding to the PDF file to be identified.
2. A PDF scan recognition method according to claim 1, characterized in that: The step of splitting the PDF file to be identified into pages to obtain a plurality of page images, and encoding and marking the plurality of page images specifically includes: Split the PDF file to be recognized into pages, and convert the split PDF files into image format; Performing image correction processing on the PDF file converted into image format; wherein the image correction processing includes at least one of image denoising, image deformation correction, image shadow and highlight correction, image watermark removal, and image moiré removal; Based on the corrected PDF file, a plurality of page images are obtained; The page sequence of the PDF file to be identified is determined, the plurality of page images are sorted based on the page sequence, and the page images are re-encoded and annotated.
3. A PDF scan recognition method according to claim 1, characterized in that: Before classifying the plurality of page images based on page content, the method further includes: Based on the acquired image sample data and the image category corresponding to the image sample data, training a preset neural network model to obtain an image classification model; The classifying of the plurality of page images based on page content specifically includes: The page image is input into the image classification model so that the image classification model outputs a plurality of image categories corresponding to the page images respectively.
4. A PDF scan recognition method according to claim 1, characterized in that: The classified page images are input into corresponding information recognition models respectively, and corresponding information content is output through multi-model collaboration, specifically including: In the case where the page image is the plain text image, removing the header and footer in the plain text image; Performing layout analysis on the plain text image to determine the column division of the text in the plain text image; The picture parts corresponding to each column are sequentially input into a preset general model, and the recognized information contents corresponding to each column are output through the preset general model.
5. A PDF scan recognition method according to claim 1, characterized in that: The classified page images are input into corresponding information recognition models respectively, and corresponding information content is output through multi-model collaboration, specifically including: In the case where the page image is the complex layout image, removing the header and footer in the complex layout image; Performing layout analysis on the complex layout image to determine the data type in the complex layout image; wherein the data type includes at least one of text data, title data, table data, seal data and image data; Based on the pixel point information corresponding to the data of each data type, respectively, the positions corresponding to the different types of data in the complex layout are determined; Inputting data of each data type into corresponding preset different data type recognition models respectively, so as to output corresponding recognized data through the preset different data type recognition models; Based on the position and the identified data, the identified information content corresponding to the complex layout image is obtained.
6. A PDF scan recognition method according to claim 1, characterized in that: The classified page images are input into corresponding information recognition models respectively, and corresponding information content is output through multi-model collaboration, specifically including: When the page image is the natural image, removing the header and footer in the natural image; Performing text information detection on the natural image, and not recognizing the natural image when no text exists in the natural image; In the case where there are words in the natural image, content information detection is performed on the PDF file to be recognized to determine whether the words in the natural image are words to be recognized or words that do not need to be recognized; In the case of the text to be recognized, the natural image is input into a preset natural scene recognition model, and the recognized information content is output through the preset natural scene recognition model.
7. A PDF scan recognition method according to claim 6, characterized in that: In the case where there are words in the natural image, performing content information detection on the PDF file to be recognized to determine whether the words in the natural image are words to be recognized or words that do not need to be recognized specifically includes: Determine the format of the text in the natural image, compare the format with a preset format template, and obtain a first detection result based on the comparison result; Extracting content from the PDF file to be identified, determining text content adjacent to the natural image, extracting keywords from the text content, comparing the extracted keywords with preset keywords, and obtaining a second detection result based on the comparison result; When the first detection result and / or the second detection result is the text to be recognized, the natural image is recognized.
8. A PDF scan recognition method according to claim 1, characterized in that: The method of combining the information contents of the plurality of page images based on the marked codes and the information contents to obtain a word document corresponding to the PDF file to be identified specifically includes: Arranging the identified information contents corresponding to the page images respectively based on the positions of the data in the page images; sorting the identified pages after the content arrangement based on the annotated codes; The identified pages after sorting are combined to obtain a word document corresponding to the PDF file to be identified.
9. A PDF scan recognition device, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 8.
10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Document scanning PDF (Portable Document Format) file-oriented parallel identification processing method
CN115063667A
Image recognition method, image auditing method and electronic equipment
CN115471841A
Captions detector
US20080077847A1
Cited By
Adaptive document understanding
CN115730032A
Methods, systems, and computer program products for adaptive document understanding
CN115730032B
