Picture book image recognition method and system combined with character information

The picture book recognition method integrates text and image features using OCR and a preset library for efficient and accurate classification by reducing irrelevant information interference.

CN119068500APending Publication Date: 2024-12-03LUCA (HANGZHOU) INTELLIGENT TECHNOLOGY CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411050263.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing methods for picture book recognition are based on image feature recognition, which often fail to accurately describe the diverse and complex characteristics of picture books due to their visual and textual content, leading to suboptimal recognition results.

Method used

A picture book recognition method and system that combines text and image features using OCR technology to extract text data, fuse it with image features, and utilize a multimodal representation for recognition, leveraging a preset feature library for efficient matching and classification.

Benefits of technology

The method improves recognition accuracy and efficiency by integrating text and image features, providing a reliable basis for matching and classification, and adaptively focusing on key areas to reduce irrelevant information interference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068500A_ABST
    Figure CN119068500A_ABST
Patent Text Reader

Abstract

The invention provides a picture book image recognition method and system combined with character information, and relates to the field of picture book image recognition. The method comprises the following steps: acquiring a picture book image to be recognized, and extracting character data in the picture book image to be recognized through OCR (Optical Character Recognition) to obtain picture book character data; performing feature extraction on the picture book image to be recognized to obtain image features; fusing the picture book character data and the image features to obtain multi-modal representation of the picture book; comparing the multi-modal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; and performing picture book image recognition on the picture book image to be recognized according to the picture book matching result and the multi-modal representation of the picture book to obtain a picture book recognition result. According to the technical scheme, the picture book image recognition capability is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of picture book image recognition, and in particular to a picture book image recognition method and system combined with text information. Background Art

[0002] With the continuous development and progress of society, picture books have become an indispensable part of children's enlightenment education. Picture books not only contain colorful pictures, but also contain simple and easy-to-understand texts, which help to cultivate children's imagination and creativity, and improve children's language expression and aesthetic abilities.

[0003] At present, there are many kinds of picture books circulating in the market. In order to facilitate the management and use of picture books, picture books need to be identified and classified. Traditional picture book recognition methods are mainly based on image feature recognition, that is, by extracting the features of picture book images, such as color, texture, shape, etc., and then matching the extracted features with the preset picture book feature library to achieve picture book recognition. However, due to the diversity and complexity of picture book images, it is often difficult to achieve ideal results by relying solely on image features for recognition. Summary of the invention

[0004] The present application provides a picture book image recognition method and system combined with text information to improve the recognition capability of picture book images.

[0005] In a first aspect, the present application provides a picture book image recognition method combined with text information, the method comprising: Acquire a picture book image to be recognized, and extract text data in the picture book image to be recognized by OCR to obtain picture book text data; Extracting features of the picture book image to be identified to obtain image features; Fusing the picture book text data and the image features to obtain a multimodal representation of the picture book; Comparing the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; Picture book image recognition is performed on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

[0006] In the above technical solution, the text data in the picture book image is extracted by OCR technology to obtain the picture book text data, and the picture book text data is fused with the extracted image features to form a multimodal representation of the picture book, which makes full use of the rich text information contained in the picture book image and makes up for the deficiency of relying solely on image features for recognition. The text information of the picture book often carries key semantic content and complements the image information. The combination of the two can more comprehensively and accurately describe the characteristics of the picture book, providing a more reliable basis for subsequent matching and recognition. Secondly, the multimodal representation of the picture book is compared with the preset picture book feature library to obtain the picture book matching result. With the help of the existing picture book feature library, picture book samples similar to the picture book to be identified can be quickly found, without the need to perform complete recognition processing on each picture book sample, which greatly improves the recognition efficiency. At the same time, through the matching results, the category and attributes of the picture book to be identified can be preliminarily determined, providing powerful prior information for subsequent fine recognition. Finally, the picture book image to be identified is recognized according to the picture book matching results and the multimodal representation of the picture book to obtain the final picture book recognition result. This step integrates the prior information of the matching results and the rich features of the multimodal representation. Through further processing of the attention mechanism and recognition model, it can adaptively focus on the key areas and features of the picture book, reduce the interference of irrelevant information, and improve the accuracy and robustness of recognition.

[0007] In a second aspect of the present application, a picture book image recognition system combined with text information is provided, the system comprising: A data acquisition module is used to acquire a picture book image to be recognized, and extract text data in the picture book image to be recognized through OCR to obtain picture book text data; A feature extraction module is used to extract features of the picture book image to be identified to obtain image features; A multimodal representation module, used for fusing the picture book text data and the image features to obtain a multimodal representation of the picture book; A matching module, used to compare the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; The picture book recognition module is used to perform picture book image recognition on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

[0008] In a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above method steps.

[0009] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the above method.

[0010] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. The present invention extracts text data from picture book images through OCR technology, obtains picture book text data, and fuses it with the extracted image features to form a multimodal representation of the picture book, making full use of the rich text information contained in the picture book image, and making up for the deficiency of relying solely on image features for recognition. The text information of the picture book often carries key semantic content, which complements the image information. The combination of the two can more comprehensively and accurately describe the characteristics of the picture book, providing a more reliable basis for subsequent matching and recognition. Secondly, the multimodal representation of the picture book is compared with the preset picture book feature library to obtain the picture book matching result. With the help of the existing picture book feature library, picture book samples similar to the picture book to be identified can be quickly found, without the need to perform complete recognition processing on each picture book sample, which greatly improves the recognition efficiency. At the same time, through the matching results, the category and attributes of the picture book to be identified can be preliminarily determined, providing powerful prior information for subsequent fine recognition. Finally, the picture book image to be identified is recognized according to the picture book matching results and the multimodal representation of the picture book to obtain the final picture book recognition result. This step integrates the prior information of the matching results and the rich features of the multimodal representation. Through further processing of the attention mechanism and recognition model, it can adaptively focus on the key areas and features of the picture book, reduce the interference of irrelevant information, and improve the accuracy and robustness of recognition.

[0011] 2. This application uses a preset text detection model to locate the text area of ​​the picture book image, which can accurately determine the area containing text information in the picture book image. The text detection model is trained and optimized on a large amount of picture book image data through a deep learning algorithm, and learns the visual characteristics and spatial distribution patterns of the text area. The model can automatically identify the text area in the picture book image, and mark the position and range of the text in the form of borders, masks, etc. Compared with traditional rule-based or manually designed text positioning methods, the text detection model has higher accuracy, robustness and generalization ability, can adapt to picture book images of different styles, formats and languages, accurately locate the text area, and provide reliable location information for subsequent text cutting and extraction.

[0012] 3. The picture book image to be identified in this application is preprocessed to obtain a preliminary picture book image, which can effectively eliminate interference factors such as noise and distortion in the picture book image and improve the quality and recognizability of the image. The preprocessing process usually includes operations such as image denoising, correction, and normalization. By smoothing, sharpening, and adjusting the contrast of the picture book image, high-frequency noise, blur, distortion and other factors that affect feature extraction in the image are removed. At the same time, the preprocessing can also perform scale adjustment, rotation correction, color correction and other operations on the picture book image to make the image meet the input requirements of the feature extraction algorithm and improve the accuracy and robustness of feature extraction. The preliminary picture book image obtained after preprocessing has a higher signal-to-noise ratio, clarity and consistency, providing a good image foundation for subsequent multi-level feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a picture book image recognition method combined with text information provided in an embodiment of the present application; Figure 2 An architecture diagram of a picture book image recognition system combined with text information provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0014] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0015] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" 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 "for example" or "for example" is intended to present related concepts in a specific way.

[0016] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0017] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0018] With the rapid development of society and the continuous progress of education, picture books have become an indispensable and important part of children's enlightenment education. Picture books attract children's attention with their unique charm. They not only contain colorful, vivid and interesting pictures, but also contain simple, easy-to-understand and catchy texts. They can effectively cultivate children's imagination and creativity, improve children's language expression and aesthetic abilities, and play a vital role in children's growth and development.

[0019] Nowadays, with the booming picture book market, there are many types of picture books and huge quantities. In order to better manage and utilize these picture book resources, an efficient and accurate picture book recognition and classification method is urgently needed. Traditional picture book recognition methods are mainly based on image feature recognition, that is, by extracting visual features of picture book images, such as color, texture, shape, etc., and then matching the extracted features with the pre-built picture book feature library, so as to realize automatic recognition and classification of picture books. However, due to the extremely high diversity and complexity of picture book images, it is often difficult to achieve ideal results by relying solely on image features for recognition. On the one hand, picture books of different styles and themes are very different visually, and a single image feature is difficult to fully describe the characteristics of picture books. On the other hand, in addition to pictures, picture books often contain a lot of text information. If these text information are ignored, the important semantic content of the picture book will be lost, affecting the accuracy of recognition. Therefore, how to make full use of the information of both picture book images and text to achieve more comprehensive and accurate picture book recognition has become an urgent problem to be solved.

[0020] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0021] Based on the above background technology, please refer to Figure 1 , Figure 1 A flowchart of a method for picture book image recognition combined with text information provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in an embodiment of the present application, the method can be applied on a server, but can also be applied to electronic devices such as a server. A method for picture book image recognition combined with text information includes the following steps: S101, obtaining a picture book image to be recognized, and extracting text data in the picture book image to be recognized by OCR to obtain picture book text data; Specifically, in order to obtain the text information of the picture book, it is first necessary to obtain the picture book image to be recognized, and then extract the text data in the picture book image to be recognized through OCR (optical character recognition) technology to obtain the picture book text data.

[0022] Specifically, the digital image of the picture book can be obtained through a scanner, a digital camera or other equipment, or the picture book image can be directly obtained from an electronic picture book file. After the picture book image is obtained, the OCR technology is used to detect and identify the text area in the image. OCR technology can automatically locate and extract the text information in the image by analyzing the pixel distribution and features in the image, and convert the text in the image into a text format that can be edited and processed by a computer to obtain the picture book text data.

[0023] In the process of extracting picture book text data, some preprocessing technologies can be used, such as image denoising, tilt correction, character segmentation, etc., to improve the accuracy of OCR recognition. At the same time, for special cases such as artistic fonts, handwriting, irregular layout, etc. that may appear in picture book images, more adaptable OCR algorithms can be selected, such as OCR models based on deep learning, to meet the text recognition needs of different picture book styles.

[0024] By extracting picture book text data through OCR technology, we can obtain key text information contained in picture books, such as story content, character dialogue, background description, etc. These text information are closely related to the theme and plot of the picture book and are an important part of the picture book content. The extracted picture book text data can be used for subsequent picture book content analysis, keyword extraction, theme classification and other tasks, providing important semantic information for picture book recognition and understanding.

[0025] Based on the above embodiment, as an optional embodiment, extracting the text data in the picture book image to be recognized by OCR to obtain the picture book text data includes: The text area of ​​the picture book image is located by using a preset text detection model to determine the text area to be extracted; S201, cutting text data in the picture book image to be recognized according to the text area to be extracted to obtain a picture book text image; Specifically, a text detection algorithm based on deep learning, such as the EAST (Efficient and Accurate Scene Text Detection) model, is used to locate the text area of ​​the picture book image through a preset text detection model. First, the picture book image to be identified is input into the EAST model, and the model extracts the multi-scale and multi-directional features of the picture book image through a convolutional neural network. Then, the regression branch and the classification branch are applied to the feature map to predict the bounding box coordinates of the text area and the confidence score of the text area, respectively. Next, the predicted bounding box is processed by non-maximum suppression (NMS) to filter out overlapping and redundant bounding boxes to obtain the final text area positioning result. The EAST model can efficiently and accurately realize the text area positioning of picture book images through end-to-end training and multi-task learning.

[0026] Based on the above embodiment, as an optional embodiment, the step of cutting the text data in the picture book image to be recognized according to the text area to be extracted to obtain the picture book text image includes: S301, determining the coordinates of the area to be cropped in the picture book image to be recognized according to the text area to be extracted; Specifically, the process of determining the coordinates of the area to be cropped in the picture book image to be identified based on the text area to be extracted is as follows: First, obtain the bounding box information of each text area in the text area positioning result, including the pixel coordinates, width and height of the upper left corner and the lower right corner. Then, according to the bounding box information, calculate the actual position and size of each text area in the picture book image. Specifically, multiply the coordinate value of the bounding box by the width and height of the picture book image to obtain the pixel-level coordinates of the text area in the original image. Next, considering that the text area may have a certain tilt or rotation, the bounding box needs to be corrected and aligned. By calculating the main direction of the bounding box, rotate it to the horizontal or vertical direction to ensure that the text area is aligned with the edge of the image. Finally, the corrected bounding box coordinates are used as the coordinates of the area to be cropped to provide a reference for subsequent text image cropping.

[0027] In the process of determining the coordinates of the area to be cropped, some special cases and boundary conditions need to be considered. For example, when the text area is at the edge of the image, the coordinate values ​​need to be truncated and corrected to ensure that the cropped area does not exceed the image range. At the same time, for smaller or sparse text areas, a more complete and continuous text image can be obtained by expanding the bounding box or merging adjacent areas. In addition, the coordinates of the area to be cropped can be fine-tuned and optimized according to the layout and font characteristics of the picture book to improve the quality and readability of the text image.

[0028] S302 : Crop the text data in the picture book image to be recognized according to the coordinates of the area to be cropped to obtain the picture book text image.

[0029] Specifically, the process of cropping the text data in the picture book image to be recognized according to the coordinates of the area to be cropped is as follows: First, determine the starting position and size of the cropping on the picture book image according to the coordinates of the area to be cropped. The coordinates of the upper left corner of the area to be cropped are used as the starting point of the cropping, and the width and height of the area are used as the cropping size. Then, use the cropping function in the image processing library, such as the cv2.rectangle() function of OpenCV, to crop the picture book image. The cropping function extracts the image data of the specified area according to the starting position and size parameters to generate a new image object. Next, the size of the cropped text image is normalized and scaled to a uniform size, such as a fixed width and a proportional height. Size normalization can eliminate the recognition differences caused by text images of different sizes and improve the generalization ability of the recognition model. Finally, the normalized text image is saved as an independent image file as the output result of the picture book text image.

[0030] S202, extracting text from the picture book text image through the OCR to obtain the picture book text data.

[0031] Specifically, the text extraction from the picture book text image through OCR adopts an OCR recognition model based on deep learning, such as the CRNN (Convolutional Recurrent Neural Network) model. First, the picture book text image is input into the convolution layer of the CRNN model, and the local and global features of the text image are extracted through multi-layer convolution and pooling operations. The convolution layer can automatically learn the feature representations such as character shape and stroke structure in the text image. Then, the feature map output by the convolution layer is input into the recurrent layer, and the feature sequence is encoded and decoded through the bidirectional LSTM (Long Short-Term Memory) network. The recurrent layer can capture the contextual dependency between characters in the text image and learn the temporal dynamic features of the character sequence. Next, the attention mechanism is applied to the output of the recurrent layer to adaptively focus on the key areas in the text image and improve the recognition accuracy. Finally, the output of the recurrent layer is matched and optimized with the real character label through the CTC (Connectionist Temporal Classification) loss function to achieve end-to-end text recognition training. Through the application of the CRNN model, the picture book text image can be efficiently and accurately converted into the corresponding text sequence.

[0032] S102, extracting features of the picture book image to be identified to obtain image features; Specifically, after obtaining the picture book image to be identified, in order to make full use of the visual information contained in the picture book image, it is necessary to extract the features of the picture book image to be identified and obtain image features. Image features are a compact and effective way to describe the visual content of an image. They can describe the characteristics of the image in terms of color, texture, shape, etc., and are an important basis for image recognition and classification.

[0033] Based on the above embodiment, as an optional embodiment, the feature extraction of the picture book image to be identified to obtain the image feature includes: S401, preprocessing the picture book image to be identified to obtain a preliminary picture book image; Specifically, the process of preprocessing the picture book image to be identified and obtaining the preliminary picture book image is as follows: First, the picture book image is normalized, and images of different sizes and resolutions are scaled to a uniform size, such as scaling the image to a fixed width and height. Size normalization can eliminate the scale differences between different images, which is convenient for subsequent feature extraction and comparison. Then, the normalized image is converted to a color space, and the image is converted from the RGB color space to a grayscale space or other color space suitable for feature extraction, such as HSV, LAB, etc. Color space conversion can reduce the color information redundancy of the image and highlight the texture and structural features of the image. Next, the converted image is denoised and smoothed, and algorithms such as median filtering and Gaussian filtering are used to remove high-frequency noise and burrs in the image, making the image smoother and continuous. Denoising and smoothing can improve the signal-to-noise ratio of the image and enhance the visual quality of the image. Finally, the smoothed image is edge detected and enhanced, and algorithms such as Canny and Sobel are used to extract edge and contour information in the image and highlight the geometric structure and shape features of the image. Edge enhancement can provide more clear and rich image information for subsequent feature extraction.

[0034] S402, performing multi-level feature extraction on the preliminary picture book image to obtain multi-scale image features; Specifically, the process of extracting multi-level features from the preliminary picture book image and obtaining multi-scale image features is as follows: First, the preliminary picture book image is extracted using a convolutional neural network (CNN). The preliminary picture book image is input into a pre-trained CNN model, such as VGG, ResNet, etc., and the hierarchical features of the image are extracted through layer-by-layer convolution and pooling operations. Feature maps of different scales and abstract levels can be obtained in different convolutional layers and pooling layers of CNN, such as shallow texture feature maps, middle-layer shape feature maps, and deep-layer semantic feature maps. Then, multi-scale analysis and fusion are performed on the feature maps extracted by CNN. Using techniques such as spatial pyramid pooling (SPP) or multi-scale pooling, feature maps of different scales are pooled and combined to generate feature representations at multiple scales. Multi-scale analysis can capture the structure and content information of the image at different scales and improve the robustness and generalization ability of the features. Next, the multi-scale features are encoded and normalized. Feature encoding methods such as Fisher Vector and VLAD are used to convert multi-scale features into compact and discriminative feature vectors. Feature encoding can remove feature redundancy and noise, and improve feature representation efficiency and discrimination. Finally, the encoded multi-scale features are connected or fused to obtain the final multi-scale image features. Through feature connection or fusion, feature information at different levels and scales can be integrated into a unified feature representation to fully describe the visual content of the picture book image.

[0035] S403: Fusing the multi-scale image features to obtain the image features.

[0036] Specifically, the process of fusing multi-scale image features to obtain image features is as follows: First, regularize the multi-scale image features. Since features of different scales and levels may have different value ranges and distributions, the features need to be normalized and standardized so that they can be fused under the same scale and distribution. Common regularization methods include maximum and minimum value normalization, Z-score standardization, etc., which map features to a unified range and distribution through linear transformation. Then, the regularized multi-scale image features are fused using feature combination methods. Common feature combination methods include feature splicing, feature weighting, feature pooling, etc. Feature splicing is to directly connect feature vectors of different scales into a long vector, feature weighting is to add features of different scales after adding different weight coefficients, and feature pooling is to perform pooling operations such as maximum value and average value on features of different scales. Through feature combination, multi-scale image features can be fused into a unified feature representation. Next, the fused features are subjected to dimensionality reduction and compression. Using dimensionality reduction methods such as principal component analysis (PCA) and linear discriminant analysis (LDA), high-dimensional fusion features are mapped to low-dimensional subspaces to remove linear correlation and redundant information between features. Dimensionality reduction can improve the compactness and computational efficiency of features while retaining the discriminability and information content of features. Finally, the fusion features after dimensionality reduction are used as the final image features for subsequent image analysis and application tasks.

[0037] S103, fusing the picture book text data and the image features to obtain a multimodal representation of the picture book; Specifically, after obtaining the picture book text data and image features, in order to make full use of the multi-source information of the picture book, it is necessary to fuse the picture book text data and image features to obtain a multimodal representation of the picture book. Multimodal representation refers to the integration of information from different modalities (such as text, image) to form a more comprehensive and compact content representation method. By fusing the text information and visual information of the picture book, the semantic content and visual characteristics of the picture book can be better portrayed, and the accuracy of picture book recognition and understanding can be improved.

[0038] Specifically, based on feature-level fusion, text features and image features are directly concatenated or connected to form a unified feature vector. For example, the picture book text data is represented as a bag-of-words model or word vector, and the image features are represented as a color histogram or CNN features, and then they are connected into a long vector as a multimodal representation of the picture book.

[0039] Based on the above embodiment, as an optional embodiment, the fusing of the picture book text data and the image features to obtain a multimodal representation of the picture book includes: S501, spatially aligning the picture book text data and the image features to obtain target picture book text data and target image features; Specifically, the process of spatially aligning the picture book text data and image features to obtain the target picture book text data and target image features is as follows: First, the picture book text data is marked and the coordinates are extracted. Use text detection and recognition algorithms, such as CTPN, EAST, etc., to locate and identify the text area in the picture book image, and obtain the position coordinates and text content of each text block. The position coordinates include the bounding box coordinates, center point coordinates, etc. of the text block, which are used to describe the spatial position of the text in the image. Then, the image features are spatially mapped and coordinates are transformed. According to the extraction method and scale of the image features, the image features are mapped to the same spatial coordinate system as the picture book image. For image features extracted based on convolutional neural networks, it is necessary to consider the receptive field and downsampling rate of the network, and map the coordinates on the feature map back to the original image coordinate system. For manually extracted image features, it is necessary to perform coordinate transformation and alignment according to the position and scale of the feature points. Next, the picture book text data and image features are spatially matched and aligned. According to the position coordinates of the text block, the corresponding area and position are found on the image features, and the spatial correspondence between the text and the image is established. Indicators such as the intersection over union (IoU) and Euclidean distance can be used to measure the overlap and similarity between text blocks and image areas, and the best matching image features can be selected as the corresponding features of the text. The aligned text data and image features form the spatially corresponding target picture book text data and target image features. Finally, the target picture book text data and target image features are spatially encoded and represented. The aligned text coordinates and image coordinates are normalized and quantized to generate a unified spatial encoding. Spatial encoding can use discrete grid coordinates, continuous position vectors, and other forms to represent the spatial position relationship between text and images. The spatially encoded target picture book text data and target image features can be used as input for subsequent multimodal fusion to model the spatial interaction and semantic association between text and images.

[0040] S502, establishing a semantic association between the target picture book text data and the target image features to obtain a multimodal representation of the picture book.

[0041] Specifically, the process of establishing the semantic association between the target picture book text data and the target image features and obtaining the multimodal representation of the picture book is as follows: First, the target picture book text data is semantically encoded and represented. The picture book text data is converted into a semantic vector representation using technologies such as word embedding and language model. Word embedding can map each word to a low-dimensional semantic space to capture the semantic similarity and analogy between words. The language model can model the text sequence and learn the contextual semantics and long-distance dependencies of the text. The semantically encoded target picture book text data represents the semantic content of the text in the form of a vector. Then, the target image features are semantically encoded and represented. The image features are converted into a semantic vector representation using technologies such as image semantic embedding and scene graph generation. Image semantic embedding can map the image to the same semantic space as the text and learn the semantic correspondence between the image and the text. Scene graph generation can extract semantic elements such as objects, attributes, and relationships from the image to construct a structured semantic representation of the image. The semantically encoded target image features represent the semantic content of the image in the form of a vector. Next, the attention mechanism is used to establish the semantic association between the target picture book text data and the target image features. Using multi-head attention, cross-attention and other mechanisms, the correlation and dependency between text semantic vectors and image semantic vectors are calculated. The attention mechanism can adaptively learn the importance and correlation strength of text and images in different semantic dimensions, and achieve alignment and fusion at the semantic level. Through attention aggregation, a multimodal context representation that integrates the semantics of text and image can be obtained. Finally, the multimodal fusion network is used to further fuse and refine the multimodal context representation after attention aggregation to obtain the final multimodal representation of the picture book. The multimodal fusion network can use structures such as gated recurrent units (GRU) and long short-term memory networks (LSTM) to perform sequence modeling and feature transformation on the multimodal context representation and extract cross-modal high-level semantic representations. The output of the fusion network is the multimodal representation of the picture book, which encodes the semantic content and modal interaction information of the picture book in a compact vector form.

[0042] In the process of semantic association, some optimization strategies targeting the characteristics of picture books can also be introduced. For example, considering the child-orientedness and educational significance of picture books, different semantic association methods can be used for different types of picture books, such as story picture books focusing on plots and character relationships, and popular science picture books focusing on knowledge points and explanations. At the same time, the granularity and depth of semantic association can be adaptively adjusted in combination with the target audience and reading level of the picture books, such as using shallow keyword matching for picture books for young children and deep semantic reasoning for picture books for older children. In addition, external knowledge bases and common sense reasoning can be introduced to provide background knowledge and reasoning ability for semantic association, thereby improving the accuracy and comprehensiveness of semantic understanding of picture books.

[0043] S104, comparing the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; Specifically, after obtaining the multimodal representation of the picture book, in order to realize the rapid recognition and classification of the picture book, it is necessary to compare the multimodal representation of the picture book with the preset picture book feature library to obtain the picture book matching result. The picture book feature library is a pre-built picture book feature data set that contains a large number of multimodal feature representations of known picture books, as well as the corresponding picture book labels and attribute information. By comparing the multimodal representation of the picture book to be identified with the picture book features in the feature library, picture books similar to the picture book to be identified can be quickly found, thereby realizing the recognition and classification of the picture book.

[0044] Specifically, a variety of similarity measurement methods can be used to compare the multimodal representation of a picture book with a preset picture book feature library. A common method is based on vector distance measurement, that is, the multimodal representation of the picture book and the picture book features in the feature library are represented as high-dimensional vectors, and then the distance between them is calculated, such as Euclidean distance, cosine similarity, etc. The smaller the distance, the more similar the multimodal representations of the two picture books are, and the closer their content and style are.

[0045] Based on the above embodiment, as an optional embodiment, the multimodal representation of the picture book is compared with a preset picture book feature library to obtain a picture book matching result, including: S601, calculating the similarity between the multimodal representation of the picture book and the template features in the picture book feature library to obtain the similarity value between the multimodal representation of the picture book and each template feature in the picture book feature library; Specifically, the multimodal representation of the picture book is calculated with the template features in the picture book feature library to obtain the similarity value of the multimodal representation of the picture book and each template feature in the picture book feature library as follows: First, the template features in the picture book feature library are multimodally represented. The picture book feature library is a pre-built knowledge base that contains a large number of template picture books and their corresponding multimodal feature representations. The template picture book is a typical picture book sample selected according to standards such as content, style, and theme, covering various types and characteristics in the picture book field. For each template picture book, the same multimodal fusion method as the current picture book is used to extract its text data and image features, and a multimodal representation is constructed, which is stored in the picture book feature library as a template feature. Then, a suitable similarity measurement method is selected to calculate the similarity between the multimodal representation of the picture book and the template features in the picture book feature library. Commonly used similarity measurement methods include cosine similarity, Euclidean distance, Jaccard similarity, etc., which are selected according to the characteristics of multimodal representation and application requirements. Taking cosine similarity as an example, the multimodal representation of the picture book and each template feature are represented as vectors, and the cosine value of the angle between them is calculated to obtain the similarity score. The higher the similarity score, the closer the picture book and the template are in the semantic space, and the higher the matching degree. Next, all template features in the picture book feature library are traversed, and the similarity is calculated with the multimodal representation of the picture book respectively to obtain a set of similarity values. The similarity value represents the matching degree between the picture book and each template picture book, which can be used for subsequent sorting, screening and recommendation. In order to improve the matching efficiency, inverted index, hash index and other technologies can be used to speed up the similarity calculation process and avoid exhaustive comparison of large-scale feature libraries. Finally, the calculated similarity value is compared with the preset threshold to filter out the template features with the highest matching degree with the picture book. The setting of the threshold can be adjusted according to the application scenario and recommendation requirements to balance the matching accuracy and recall rate. Template features exceeding the threshold are regarded as highly matching with the picture book and can be used as candidate results for further sorting and recommendation. Template features that do not exceed the threshold are filtered out to reduce irrelevant noise and interference.

[0046] S602: taking the template features whose similarity values ​​exceed a preset threshold as the picture book matching results.

[0047] Specifically, the process of taking the template features whose similarity values ​​exceed the preset threshold as the picture book matching results is as follows: First, the similarity values ​​calculated previously are sorted in descending order, and the similarity values ​​are arranged from high to low. After sorting, the template features with higher similarity values ​​are located at the front of the list, indicating that the matching degree with the current picture book is higher; the template features with lower similarity values ​​are located at the back of the list, indicating that the matching degree with the current picture book is lower. Through sorting, the optimal matching result can be easily identified, and an orderly processing order is also provided for subsequent threshold screening. Then, a similarity threshold is set to determine whether the matching degree between the template feature and the picture book meets the expected standard. The setting of the threshold needs to comprehensively consider the matching accuracy and recall rate, as well as factors such as application scenarios and user needs. A higher threshold can improve the accuracy of the matching results, but may reduce the number of matching results; a lower threshold can increase the number of matching results, but may introduce some less relevant features. The selection of the threshold needs to balance between precision and recall rate, and be adjusted and optimized according to the actual situation. Next, the sorted similarity value list is traversed, and the similarity value is compared with the preset threshold. For each template feature, determine whether its similarity value exceeds the threshold. If the similarity value is greater than or equal to the threshold, the template feature is added to the picture book matching result set, indicating that the feature has a high degree of match with the current picture book and can be used as a recommended candidate. If the similarity value is less than the threshold, the template feature is removed and not included in the matching result to ensure the matching quality and relevance. Repeat this process until all template features are traversed to form the final picture book matching result set. Finally, the screened picture book matching results are sorted according to the size of the similarity value, and the most relevant and matching template features are placed in front. The sorted matching results can be directly presented to the user as a basis for picture book recommendations and reading guidance. Users can choose picture books of interest to read and explore based on the relevance of the matching results and their personal interests. At the same time, the picture book matching results can also be used as input for subsequent personalized recommendations and intelligent reading services to provide users with continuous and thoughtful reading support.

[0048] S105 , performing picture book image recognition on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

[0049] Specifically, according to the picture book matching results and the multimodal representation of the picture book, an end-to-end recognition model based on deep learning is used for picture book image recognition, such as a combination model of convolutional neural network (CNN) and recurrent neural network (RNN). First, the multimodal representation of the picture book is used as the input of CNN, and the high-level visual features of the picture book image are extracted through multi-layer convolution and pooling operations. Then, the extracted visual features are spliced ​​with the word vector representation of the picture book text to form a mixed feature vector containing visual and text information. Next, the mixed feature vector is input into the RNN, and the picture book content is serialized and modeled and learned through the loop structure of the recurrent neural network. RNN can capture the contextual dependency of the picture book content and learn the temporal dynamic characteristics of the picture book story. Finally, in the output layer of the RNN, multiple classifiers and regressors are set, corresponding to different recognition tasks, such as picture book category classification, keyword extraction, storyline generation, etc. Through end-to-end joint training, the parameters of CNN and RNN are optimized to achieve end-to-end learning and prediction of the picture book image recognition model.

[0050] In the picture book image recognition process, attention mechanism and knowledge fusion technology are also introduced to further improve the recognition performance. Through the attention mechanism, the recognition model can adaptively focus on the key areas and keywords in the picture book image and text, highlight important information, and reduce irrelevant interference. At the same time, the picture book matching results are used as prior knowledge and fused with the output of the recognition model. Specifically, the picture book category, style and other information in the matching results are encoded into vector representations, which are concatenated or element-wise multiplied with the output vector of the recognition model to form a knowledge-enhanced recognition result. Through knowledge fusion, the recognition model can make full use of the useful information provided by the matching results to improve the accuracy and consistency of recognition.

[0051] Based on the above embodiment, as an optional embodiment, performing picture book image recognition on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result includes: S701, performing attention mechanism processing on the picture book matching result and the multimodal representation of the picture book to obtain target picture book features; Specifically, first, the picture book matching result and the multimodal representation of the picture book are feature fused. The picture book matching result is a set of template features that are highly matched with the picture book image to be identified by screening through a similarity threshold, and contains feature information in multiple aspects such as content, style, and theme. The multimodal representation of the picture book is a unified representation obtained by semantically associating the text data and image features of the picture book, which comprehensively depicts the semantic connotation and modal interaction information of the picture book. By fusing the two, the complementarity of the matching results and the multimodal representation can be fully utilized to obtain a more comprehensive and rich picture book feature representation. The fusion method can use operations such as splicing and weighted averaging to merge feature vectors from different sources into a unified feature representation. Then, the attention mechanism is introduced to evaluate the importance and assign weights to the fused picture book features. The attention mechanism can use classic models such as multi-head attention and self-attention to adaptively learn the importance of different features by calculating the correlation and dependency between features. Specifically, the attention mechanism assigns an attention weight to each feature, indicating the importance of the feature for picture book image recognition. The calculation of weights can be based on factors such as similarity and contextual relationships between features, and can be normalized and transformed through functions such as softmax. The size of the attention weight reflects the saliency and discriminability of the feature. The higher the weight, the greater the impact of the feature on the recognition result. Next, the picture book features are weighted and aggregated according to the attention weight to obtain the target picture book features. Each feature is multiplied by its corresponding attention weight to obtain a weighted feature representation. Then, the weighted features are aggregated, usually using operations such as addition or concatenation, to obtain a compact feature vector, which is the target picture book feature. The target picture book feature concentrates on the core content and key attributes of the picture book image to be identified. Through the importance evaluation and weight allocation of the attention mechanism, the feature information most relevant to the recognition task is highlighted, and the interference of minor or noise features is suppressed. Finally, the target picture book feature is output as the input feature for subsequent picture book image recognition. The target picture book feature is represented in the form of a compact vector, which contains rich semantic information and discriminative ability, and can be directly used for recognition tasks such as feature matching, similarity calculation, and classification decision. At the same time, the target picture book features can be further combined with other feature extraction and fusion methods, such as convolutional neural networks, recurrent neural networks, etc., to build a more powerful and robust picture book image recognition model.

[0052] S702: Input the target picture book feature into a preset picture book recognition model to obtain the picture book recognition result.

[0053] Specifically, the process of inputting the target picture book features into the preset picture book recognition model to obtain the picture book recognition result is as follows: First, load the preset picture book recognition model. The picture book recognition model is an intelligent model specially designed and trained for the picture book image recognition task, and usually adopts a deep learning architecture, such as a convolutional neural network, a recurrent neural network, an attention mechanism, etc. The input of the model is the feature representation of the picture book image, and the output is the semantic annotation result of the picture book image. The training process of the picture book recognition model requires a large amount of picture book image data and corresponding annotation information. Through supervised learning, semi-supervised learning, etc., the model learns the feature representation and semantic mapping relationship of the picture book image. The trained picture book recognition model is stored in the form of a parameter file and can be directly loaded into the memory for subsequent reasoning and prediction. Then, the target picture book features are input into the picture book recognition model. The target picture book features are compact feature vectors obtained by processing the attention mechanism, which concentrate on the core content and discriminant attributes of the picture book image to be recognized. The target picture book features are passed to the picture book recognition model as the input data of the model. The picture book recognition model processes and transforms the input features layer by layer, extracts high-level semantic features through operations such as convolution, pooling, and attention, and performs feature fusion and discriminant analysis. The internal structure and parameter settings of the model determine the effect of feature extraction and recognition, and repeated training and optimization are needed to improve the performance and generalization ability of the model. Next, the picture book recognition model infers and predicts the target picture book features to obtain the picture book recognition results. The output layer of the model usually uses a softmax function or other classifier to map high-level semantic features to predefined category labels or attribute values. The picture book recognition results can be semantic annotations in multiple aspects, such as object detection results, scene recognition results, text extraction results, sentiment analysis results, etc. Each recognition result is expressed in the form of a label, probability value, or attribute value, reflecting the content and characteristics of the picture book image in different semantic dimensions. The reasoning process of the model is usually carried out in real time. Through the forward propagation algorithm, the input features are converted and processed layer by layer to finally obtain the output results. Finally, the picture book recognition results are parsed and applied. The output of the picture book recognition model is structured semantic annotation, which needs to be parsed and used according to specific application scenarios and needs. For example, object detection results can be used for automatic annotation and retrieval of picture book content; scene recognition results can be used for theme classification and style analysis of picture books; text extraction results can be used for text generation and speech synthesis of picture books; sentiment analysis results can be used for sentiment recommendation of picture books and user experience optimization, etc. By integrating and associating picture book recognition results with other information, such as picture book metadata and user portraits, more personalized and intelligent picture book recommendation and reading guidance services can be achieved.

[0054] On the other hand, the present invention also provides a picture book image recognition system combined with text information, such as Figure 2 , the system comprises: The data acquisition module 1 is used to acquire a picture book image to be recognized, and extract text data in the picture book image to be recognized by OCR to obtain picture book text data; Feature extraction module 2, used for extracting features of the picture book image to be identified to obtain image features; A multimodal representation module 3, used for fusing the picture book text data and the image features to obtain a multimodal representation of the picture book; A matching module 4 is used to compare the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; The picture book recognition module 5 is used to perform picture book image recognition on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

[0055] Please refer to Figure 3 The application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301 , at least one network interface 304 , a user interface 303 , a memory 305 , and at least one communication bus 302 .

[0056] The communication bus 302 is used to realize the connection and communication between these components.

[0057] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0058] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0059] Among them, the processor 301 may include one or more processing cores. The processor 301 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field~Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 301 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301, and it can be implemented separately through a chip.

[0060] Among them, the memory 305 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read~Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may optionally be at least one storage system located away from the aforementioned processor 301. Refer to Figure 3 The memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a flocculant addition analysis method.

[0061] exist Figure 3In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing the road assessment method in the memory 305. When executed by one or more processors 301, the electronic device 300 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application. In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0062] In the several implementations provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of the system or unit can be electrical or other forms.

[0063] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0064] The present application also provides a computer storage medium that can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figure 1 The road assessment method of the embodiment shown in the figure can be found in the specific implementation process. Figure 1 The specific description of the illustrated embodiment will not be repeated here.

[0065] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0066] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory, including a number of instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0067] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the specification and the practice disclosure, those skilled in the art will easily think of other embodiments of the present disclosure.

[0068] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A picture book image recognition method combined with text information, characterized in that: The method comprises: Obtaining a picture book image to be recognized, and extracting text data from the picture book image to be recognized through OCR to obtain picture book text data; Extracting features of the picture book image to be identified to obtain image features; fusing the picture book text data and the image features to obtain a multimodal representation of the picture book; Comparing the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; Picture book image recognition is performed on the picture book image to be recognized according to the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

2. The method according to claim 1, characterized in that The extracting text data from the picture book image to be recognized by OCR to obtain picture book text data includes: Use the preset text detection model to locate the text area of ​​the picture book image and determine the text area to be extracted; cutting the text data in the picture book image to be recognized according to the text area to be extracted to obtain a picture book text image; The OCR is used to extract text from the picture book text image to obtain the picture book text data.

3. The method according to claim 2, characterized in that The step of cutting the text data in the picture book image to be recognized according to the text area to be extracted to obtain the picture book text image includes: Determining the coordinates of the area to be cropped in the picture book image to be recognized according to the text area to be extracted; The text data in the picture book image to be recognized is cropped according to the coordinates of the area to be cropped to obtain the picture book text image.

4. The method according to claim 1, wherein The step of extracting features from the picture book image to be identified to obtain image features includes: Preprocessing the picture book image to be identified to obtain a preliminary picture book image; Performing multi-level feature extraction on the preliminary picture book image to obtain multi-scale image features; The multi-scale image features are subjected to feature fusion to obtain the image features.

5. The method according to claim 1, wherein The fusing of the picture book text data and the image features to obtain a multimodal representation of the picture book includes: spatially aligning the picture book text data and the image features to obtain target picture book text data and target image features; A semantic association is established between the target picture book text data and the target image features to obtain a multimodal representation of the picture book.

6. The method according to claim 1, characterized in that The step of comparing the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result includes: Calculating similarity between the multimodal representation of the picture book and the template features in the picture book feature library to obtain similarity values ​​between the multimodal representation of the picture book and each template feature in the picture book feature library; The template features whose similarity values ​​exceed a preset threshold are used as the picture book matching results.

7. The method according to claim 1, characterized in that The performing picture book image recognition on the picture book image to be recognized based on the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result includes: Performing attention mechanism processing on the picture book matching result and the multimodal representation of the picture book to obtain target picture book features; The target picture book features are input into a preset picture book recognition model to obtain the picture book recognition result.

8. A picture book image recognition system combined with text information, characterized in that: include: A data acquisition module is used to acquire a picture book image to be recognized, and extract text data from the picture book image to be recognized through OCR to obtain picture book text data; A feature extraction module is used to extract features of the picture book image to be identified to obtain image features; a multimodal representation module, configured to fuse the picture book text data with the image features to obtain a multimodal representation of the picture book; A matching module, configured to compare the multimodal representation of the picture book with a preset picture book feature library to obtain a picture book matching result; The picture book recognition module is used to perform picture book image recognition on the picture book image to be recognized based on the picture book matching result and the multimodal representation of the picture book to obtain a picture book recognition result.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executed by a method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The electronic device comprises a processor, a memory and a transceiver, wherein the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Self-adaptive OCR (optical character recognition) method, system, equipment and medium

    CN120997840A