A method and platform for automatic extraction and annotation of image data

Abstract

The present application provides an automatic extraction and annotation method and platform for image data, which relates to the field of image annotation technology. The method reads the identification of a target image and performs size recognition of the content, configures an initial resolution for global image acquisition, establishes a global recognition result, performs text recognition on the image, calls a geometric figure segmentation unit in an annotation network to perform image segmentation of the target image based on the size recognition result and establishes an associated identification, synchronizes it to an annotation network recognition layer to perform automatic recognition and annotation, establishes an annotation result and generates an annotation label, and annotates and displays the annotation result and the annotation label, thereby solving the problems of insufficient processing accuracy of complex images, poor user interaction and personalized adjustment, and lack of flexibility and adaptability in image annotation, and achieving the effect of improving the processing response capability, accuracy, efficiency, and system adaptability and flexibility of image annotation.

Description

Technical Field

[0001] The present invention relates to the technical field of image annotation, and in particular to a method and platform for automatic extraction and annotation of image data. Background Art

[0002] Computer-aided design and other technologies have become indispensable tools in the field of modern design and engineering. CAD software helps users to carry out two-dimensional and three-dimensional design, drawing and annotation through a graphical interface, and is widely used in industries such as architectural design, mechanical manufacturing, and electronic engineering. With the increasing complexity of engineering projects and the development of diversified needs, CAD technology has been constantly innovating, from early static graphic design to today's intelligent and automated annotation systems, and the technological evolution has accelerated. Although traditional CAD tools have certain advantages in design accuracy and complex project processing, they still have some limitations in terms of the flexibility and automation of drawing viewing, modification and annotation. Specifically, image annotation is a process of assigning labels to specific content in an image. These labels can describe objects in the image and other feature information. Existing image annotation technologies include manual annotation and automatic annotation. Among them, manual annotation usually requires a lot of manual input, which is very time-consuming and costly; while the automatic annotation technology attached to the software has improved efficiency, it still has some limitations, such as low annotation accuracy, inability to process complex images, and lack of flexibility to adapt to the specific needs of different users.

[0003] In summary, existing image annotation technologies often have technical problems such as insufficient processing accuracy for complex images, lack of necessary user interaction and personalized adjustment, lack of flexibility and adaptability to cope with long-term annotation performance, resulting in reduced annotation efficiency and reliability. Summary of the invention

[0004] The present application provides a method and platform for automatic extraction and annotation of image data, which is used to solve the technical problems existing in the prior art of image annotation technology, namely, insufficient processing accuracy of complex images, lack of necessary user interaction and personalized adjustment, lack of flexibility and adaptability to cope with long-term annotation performance, resulting in reduced annotation efficiency and reliability.

[0005] In view of the above problems, the present application provides a method and platform for automatic extraction and annotation of image data.

[0006] In a first aspect, the present application provides a method for automatically extracting and annotating image data, the method comprising:

[0007] The image identifier of the target image is read, and the size of the image content of the target image is identified. The initial resolution is configured with the size identification result of the image content, and the global acquisition of the target image is performed using the initial resolution to establish a global recognition result; the text recognition is performed on the target image, and the annotation network is initialized according to the text recognition result, the image identifier, and the global recognition result. The initialization content includes the annotation strategy and the recognition bias; the geometric figure segmentation unit in the annotation network is called to perform image segmentation of the target image based on the size recognition result, and an associated identifier of the image segmentation result is established; the associated identifier and the image segmentation result are synchronized to the recognition layer of the annotation network, and automatic recognition and annotation are performed. The recognition layer is initialized by the recognition bias and the annotation strategy; the annotation result is established, and the annotation label of the annotation result is generated. The annotation label includes the main structure label, the auxiliary reference line label, and the feature type label. The annotation label is a floating index label used to adjust the annotation layer of the annotation result; the annotation result and the annotation label are annotated and displayed.

[0008] In a second aspect, the present application provides an image data automatic extraction and annotation platform, the platform comprising:

[0009] The image recognition module is used to read the image identifier of the target image, and perform size recognition of the image content of the target image, configure the initial resolution based on the size recognition result of the image content, use the initial resolution to perform global acquisition of the target image, and establish a global recognition result; the initialization module is used to perform text recognition in the target image, and initialize the annotation network according to the text recognition result, image identifier, and global recognition result. The initialization content includes annotation strategy and recognition bias; the image segmentation module is used to call the geometric figure segmentation unit in the annotation network, perform image segmentation of the target image based on the size recognition result, and establish an associated identifier of the image segmentation result; the annotation module is used to synchronize the associated identifier and image segmentation result to the recognition layer of the annotation network, perform automatic recognition and annotation, and the recognition layer is initialized by the recognition bias and annotation strategy; the annotation adjustment module is used to establish the annotation result and generate the annotation label of the annotation result, the annotation label includes the main structure label, the auxiliary reference line label, and the feature type label. The annotation label is a floating index label used to adjust the annotation layer of the annotation result; the annotation display module is used to annotate and display the annotation result and the annotation label.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The present application provides a method for automatically extracting and annotating image data. The method comprises the following steps: reading an image identifier of a target image, performing size recognition on the target image content, configuring an initial resolution based on the size recognition result of the image content, performing global acquisition of the target image using the initial resolution, and establishing a global recognition result; performing text recognition on the target image, initializing an annotation network based on the text recognition result, the image identifier, and the global recognition result, wherein the initialization content includes an annotation strategy and an identification bias; calling a geometric figure segmentation unit in the annotation network, performing image segmentation of the target image based on the size recognition result, and establishing an associated identifier of the image segmentation result; synchronizing the associated identifier and the image segmentation result to the recognition layer of the annotation network, performing automatic recognition and Annotation, the recognition layer is initialized by identifying bias and annotation strategy; the annotation result is established, and the annotation label of the annotation result is generated, the annotation label includes the main structure label, the auxiliary reference line label, and the feature type label. The annotation label is a floating index label, which is used to adjust the annotation layer of the annotation result; the annotation result and the annotation label are annotated and displayed, which solves the technical problems of insufficient processing accuracy of complex images, lack of necessary user interaction and personalized adjustment, lack of flexibility and adaptability to cope with long-term annotation performance, resulting in reduced annotation efficiency and reliability, and achieves the technical effect of improving the processing response capability, accuracy, efficiency and system adaptability and flexibility of image annotation, thereby meeting the personalized needs of different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of the process of automatically extracting and annotating image data is provided for this application.

[0013] Figure 2 A schematic diagram of the structure of an automatic image data extraction and annotation platform is provided for this application.

[0014] Explanation of the reference numerals: image recognition module 11 , initialization module 12 , image segmentation module 13 , annotation module 14 , annotation adjustment module 15 , annotation display module 16 . DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a method for automatically extracting and annotating image data, the method comprising:

[0017] The image identifier of the target image is read, and the size of the image content of the target image is recognized. The initial resolution is configured based on the size recognition result of the image content. The initial resolution is used to perform global acquisition of the target image and establish a global recognition result.

[0018] Specifically, the target image (i.e., CAD drawing file) is first read and its image identifier is identified. The image identifier includes metadata of the image, such as file name, creation date, image source, etc. Then, the size recognition of the image content is performed. The key to this step is to analyze the physical size and pixel resolution of the image, so as to determine the richness of the image details and its appropriate processing resolution. For example, a large image with rich details may need to set a higher initial resolution to ensure that the details are not lost in subsequent processing. Then, according to the size recognition result of the image content, a suitable initial resolution is configured. This step is to adjust the input resolution of the image processing system based on the recognition result to ensure that the image can be processed at the most appropriate resolution in the subsequent global acquisition. The correct configuration of the initial resolution is crucial to improving the accuracy and efficiency of image processing. If it is set too low, important details may be missed, and if it is set too high, unnecessary computing load may be caused. After the initial resolution is configured, the target image is globally acquired using this resolution. This process involves scanning and analyzing the entire image content to establish a global recognition result. The global recognition result refers to the preliminary information about the main content and background extracted from the image, including but not limited to the main color area of ​​the image, possible object boundaries, and other significant visual features. This global recognition result provides basic data and reference for subsequent more detailed image processing and annotation, ensuring that each step is optimized based on the most comprehensive image information, thereby greatly improving the accuracy and effectiveness of automatic image annotation. Through the above steps, image data can be automatically extracted effectively, the processing flow can be significantly optimized, and the quality of the processing results can be improved.

[0019] Perform text recognition on the target image, and initialize the annotation network based on the text recognition results, image identification, and global recognition results. The initialization content includes annotation strategy and recognition bias.

[0020] Optionally, the target image is further subjected to text recognition within the image. This step first detects and analyzes the target image, and then extracts the text content in the image. By identifying various fonts and text layout formats, key text information in the image, such as date, name, title or any significant text description, can be extracted. After recognition, the text recognition result is obtained, which is used together with the previous image identification and global recognition results to initialize the annotation network. The combination of this information provides comprehensive background knowledge for the initialization of the annotation network, ensuring that the annotation system can operate more accurately on the specific image content. Among them, the initialization of the annotation network refers to configuring and preparing a special neural network or algorithm system so that it can start the automatic annotation task of the image. This step involves multiple key links, including setting network parameters, defining annotation strategies, adjusting recognition biases, and loading necessary pre-trained models and weights according to specific image data and task requirements. When initializing the annotation network, important steps include setting annotation strategies and recognition biases. Annotation strategies refer to defining how the network identifies and annotates various elements in the image, such as deciding which objects should be annotated, how to set the priority of annotations, and how to handle multiple objects in the image. Recognition bias is a preference setting for the network, such as preferring to label certain types of objects or emphasizing certain features under certain conditions, which helps to adjust the network's output to meet specific application requirements.

[0021] For example, in a specific embodiment, the annotation strategy can be dynamic, which automatically adjusts the annotation rules based on the drawing type (such as architectural drawings, mechanical drawings, electronic circuit diagrams, etc.). The system can automatically identify the drawing type and apply different annotation templates or annotation strategies, which do not rely on fixed annotation rules, but are adaptive to the characteristics of the drawings. At the same time, the annotation strategy also has a self-optimization mechanism. The system can adjust the annotation rules and model parameters by analyzing historical annotation records and user feedback. For example, if the user manually adjusts the annotation position, the system can automatically record the adjustment and optimize future annotation operations in similar scenarios. In addition, the annotation strategy also involves semantic analysis and multi-source data fusion. Specifically, the element names and annotations of the graphics in the CAD drawings are analyzed by combining natural language processing (NLP) with geometric recognition, and intelligently associated with geometric features, such as being able to automatically identify and annotate complex physical structures such as "left wall", "main beam", and "bearing". At the same time, standard data from external databases, such as building specifications and mechanical standards, are introduced in the drawing annotation process, and combined with geometric analysis and semantic analysis to provide reference for automatic annotation in specific fields. In summary, when processing an image containing complex industrial parts, text recognition may extract part numbers and technical specifications. This information will be used to initialize the annotation network in combination with the dynamic self-optimization mechanism to ensure that these technical details are accurately reflected in the final image annotation. Through such an initialization process, the annotation network can perform efficient and accurate automatic annotation for specific image content and preset user needs, greatly improving processing capabilities, efficiency, and application flexibility.

[0022] The geometric figure segmentation unit in the annotation network is called to perform image segmentation of the target image based on the size recognition result, and an associated identifier of the image segmentation result is established.

[0023] Exemplarily, the target image is segmented using the geometric segmentation unit in the annotation network. The purpose of this process is to refine the entire image into manageable and identifiable small parts for more accurate analysis and annotation. First, the resolution of the image processing is adjusted based on the previous size recognition results. Size recognition involves analyzing the physical size and resolution of the image to ensure that the details of the image remain unchanged in subsequent processing. For example, an engineering drawing containing multiple mechanical parts may retain more details at high resolution, making the edges and connection points of each part more clearly visible. Subsequently, the geometric segmentation unit in the annotation network is called. This unit uses advanced image analysis techniques, such as edge detection algorithms or region growing algorithms to identify and segment geometric shapes and structures in the image. For example, in the annotation of engineering drawings, the unit can identify the outline of a single component and segment elements such as screws, gears, and bearings. The algorithm can also be a grouping algorithm based on the topological structure of the graph. Specifically, the topological structure of the graph is analyzed by an intelligent algorithm, and the connected lines and shapes are automatically grouped into a logical object, and corresponding annotations are generated for it, so that complex geometric structures can be more accurately annotated, such as closed areas composed of polylines, connected mechanical parts, etc. After segmentation, each independent image part (such as a specific mechanical part or an annotation on a drawing) will be assigned an associated identifier, which is a key element linked to the original image and subsequent processing steps, allowing the system to track the specific information of each part and ensure data consistency and accuracy during the annotation process. For example, each segmented part will be equipped with a unique serial number, which can be used to reference specific design parameters or assembly instructions. In short, through this sophisticated image segmentation processing, the annotation network is able to conduct in-depth analysis of the image and improve the overall annotation accuracy, which not only improves the efficiency of the automated system, but also reduces errors and inconsistencies introduced by manual annotation. Moreover, when processing complex drawings or technical images, this step can automatically identify and record every small part and detail, greatly improving the continuity and traceability of the workflow from design to production, and ensuring the processing capability of complex images.

[0024] The association identification and image segmentation results are synchronized to the recognition layer of the annotation network to perform automatic recognition and annotation, and the recognition layer is initialized by the recognition bias and annotation strategy.

[0025] Further, once the image segmentation and its associated identifiers are established, the next step is to synchronize these segmentation results and associated identifiers to the recognition layer of the annotation network to perform an automated recognition and annotation process. Specifically, each image segmentation result generated in the previous step and its corresponding associated identifier are synchronized to the recognition layer of the annotation network. As a kind of metadata, the associated identifier provides a unique reference identifier for each segmented part, so that subsequent processing can accurately correspond to a specific part of the original image. The initialization of the recognition layer is based on the previously set recognition bias and annotation strategy. The recognition bias involves adjusting the processing preference of the network to prioritize the recognition and processing of key information in the image, such as a specific type of object or feature. The annotation strategy defines the rules used in the automatic annotation process, such as how to annotate different types of objects, the level of annotation detail, and how to handle complex interactions in the image. After the recognition layer is initialized, the automated recognition and annotation process is performed, which includes using advanced technologies such as machine learning models and image recognition algorithms to analyze the synchronized image segmentation results and identify elements such as objects, shapes, and texts therein. Each element is automatically labeled with an appropriate label according to its nature and context. Through this automated recognition and annotation process, the efficiency and accuracy of image processing are significantly improved. Automated processing reduces the need for manual intervention and speeds up the entire workflow from image capture to annotation. In addition, the flexibility and adaptability of the system are enhanced, and the recognition bias and annotation strategies can be adjusted according to different application requirements to achieve customized image processing solutions.

[0026] The annotation result is established, and the annotation label of the annotation result is generated. The annotation label includes a main structure label, an auxiliary reference line label, and a feature type label. The annotation label is a floating index label used to adjust the annotation layer of the annotation result.

[0027] The annotation results and annotation labels are annotated and displayed.

[0028] Specifically, after completing the automatic recognition and analysis of the image, the annotation results are established and the corresponding annotation labels are generated. The annotation results cover the identified elements and their attributes in the image, and these results are clearly presented through the annotation labels for further application and analysis. Annotation labels are a key component in the image annotation process, including main structure labels, auxiliary reference line labels, and feature type labels. These labels are classified according to the importance and function of the identified content, thereby ensuring the integrity and accuracy of the annotation information. For example, when processing mechanical drawings, the main structure label may identify the main mechanical components, the auxiliary reference line label is used to indicate the assembly or alignment line, and the feature type label describes the specific characteristics of the component such as material or function. In addition, the annotation label is designed as a floating index label, which means that the label is not statically fixed, but can be freely adjusted between different annotation layers according to needs. This design allows users or automatic systems to select the most suitable annotation display mode according to different application scenarios, enhancing the flexibility and user interactivity of the annotation system. Through the dynamic management of these labels, complex annotation information can be effectively organized and displayed, ensuring that users can obtain the most relevant and accurate data when viewing and analyzing images. Finally, the annotation results and their related annotation labels are displayed on the user interface, so that all the identification and annotation information can be presented to the user intuitively. This process not only improves the readability and practicality of the image data, but also facilitates subsequent editing, review or further analysis, thereby greatly improving the efficiency and quality of image processing work, while also ensuring the efficiency and practicality of image annotation.

[0029] Furthermore, the calling of the geometric figure segmentation unit in the annotation network, performing image segmentation of the target image based on the size recognition result, and establishing an associated identifier of the image segmentation result also includes:

[0030] Calculate the image content complexity of the target image and establish the complexity distribution result.

[0031] The complexity distribution results and size recognition results are used to establish the segmentation constraint interval of the segmentation block.

[0032] The contour extraction channel in the geometric figure segmentation unit is called to establish a preliminary segmentation result through the segmentation constraint interval and the contour extraction result.

[0033] The preliminary segmentation results are subjected to semantic recognition based on the image content, and the image segmentation results and associated identifiers are established using the semantic recognition results.

[0034] In a specific embodiment, the system performs a calculation of the image content complexity on the target image. This step establishes a complexity distribution result by analyzing the detail richness and change frequency of each region in the image. The complexity distribution reflects the difference in the processing difficulty of different regions in the image, and provides key data for the subsequent segmentation step. Next, using the obtained complexity distribution result combined with the size recognition result, the system then defines the segmentation constraint interval of the segmentation block. These constraint intervals take into account the specific characteristics of the image, such as the complexity and size information of the region, to ensure that the image segmentation process can be accurately carried out according to the actual needs of the image content. For example, when processing an image with a multi-level structure, the segmentation constraint interval will allow for more detailed segmentation to capture more details. Subsequently, the contour extraction channel of the geometric figure segmentation unit in the annotation network is called. This channel specializes in processing the contour information of the image, and performs preliminary segmentation of the image according to the previously set segmentation constraint interval and the identified complexity distribution. In this way, while ensuring high efficiency, the accuracy of the segmentation results is also maintained, ensuring that each segmentation block can effectively reflect the key features of the original image. After completing the preliminary segmentation, these segmentation results are further subjected to semantic recognition based on the image content. This recognition process involves deep learning models or other intelligent algorithms to parse the specific content in each segmented block, such as the type of object, function, or any other relevant attributes. The results of semantic recognition are then used to further refine the image segmentation results and form a complete annotation framework together with the corresponding association identifiers. These association identifiers not only accurately mark the function and location information of each part, but also provide the necessary links and references for subsequent processing and application. Through the above steps, the fine segmentation and accurate annotation of the target image are achieved, greatly improving the efficiency and reliability of the automatic image annotation system.

[0035] Furthermore, the performing of semantic recognition based on the image content on the preliminary segmentation result and using the semantic recognition result to establish the image segmentation result and the associated identifier also includes:

[0036] A standard shape database is established, and the preliminary segmentation results are matched with the standard shape database to establish standard shape semantic labels.

[0037] Perform texture feature extraction on the preliminary segmentation results, establish texture feature extraction results, and use the texture feature extraction results to establish texture semantic labels.

[0038] Functional objects are identified on the preliminary segmentation results and functional semantic labels are established.

[0039] Perform semantic recognition of spatial relationships on the preliminary segmentation results and establish spatial relationship semantic labels.

[0040] Integrate standard shape semantic labels, texture semantic labels, functional semantic labels, and spatial relationship semantic labels to establish image segmentation results and associated labels.

[0041] Optionally, in-depth semantic recognition of the preliminary segmented image results is a key step in achieving high-precision automatic annotation. By comprehensively applying multiple recognition technologies and databases, each segmented block in the image can be analyzed and annotated in detail, further improving the accuracy and practicality of annotation. First, a standard shape database is established, which not only contains basic geometric shapes such as circles, rectangles, etc., but also includes complex shape combinations, such as industrial components such as bearings composed of multiple circles. These standard shapes are used as templates for association matching with the preliminary segmentation results. Through this matching, the parts of the image that meet the standard shapes can be quickly identified and the corresponding standard shape semantic labels are assigned to these parts. For example, when a graphic matching the bearing shape in the database is identified in a mechanical drawing, the graphic will be automatically labeled as "bearing". Subsequently, texture feature extraction is performed on each preliminary segmentation result, which involves analyzing the surface features of the image, such as the roughness, pattern, and repetitive nature of the texture. The extracted texture features are used to generate texture semantic labels, which help the annotation system to more accurately describe the physical and visual properties in the image, such as indicating that a part has a metallic luster or a wood texture. In addition, the system also performs functional object recognition, which involves identifying functional objects and operating symbols in the image, such as switches, buttons, etc., and assigning functional semantic labels to these objects. This step is particularly important for understanding the operating logic and user interface in the image, enabling the system to provide more accurate operating instructions and component information in an automated environment. In addition, the system also performs semantic recognition of spatial relationships, identifying and annotating the spatial arrangement and relationship between objects in the image. Spatial relationship semantic labels can show the relative position and connection between components, which is extremely important for assembling, maintaining or designing complex equipment. Finally, all the recognized semantic labels, standard shape labels, texture labels, functional labels and spatial relationship labels are integrated to establish detailed association labels for each part of the image segmentation result. These association labels not only include the physical and geometric features of the image, but also integrate its functionality and contextual information, thereby greatly enriching the content and depth of automatic annotation. Through this multi-dimensional recognition and annotation strategy, a comprehensive, accurate and feature-rich image processing solution can be provided, thereby improving the adaptability of automatic annotation.

[0042] Furthermore, the method further comprises:

[0043] Generates the default layer aggregation result of the annotation result based on the global recognition result.

[0044] Read the user account information of the user, and determine whether the user account information has a label preference layer constraint.

[0045] If there is no label preference layer constraint, the default layer aggregation result will be displayed as the label result.

[0046] Specifically, the process of generating annotation results begins with the analysis of global recognition results, based on which default layer aggregation results are generated. This process covers the comprehensive recognition of the image, including the identification of the main elements and structures in the image. This information is then used to build a hierarchical view in which each element is automatically assigned to the appropriate layer. For example, when processing architectural blueprints, structural elements, electrical systems, and plumbing projects may be aggregated into different layers to facilitate user viewing and management. Further processing involves user interaction. The system reads the user's user account information, including their previous operation preferences and settings to determine whether there are specific annotation preference layer constraints. This step is to personalize the user experience and ensure that the output meets the user's specific needs. For example, an engineer who often works with electrical installation drawings may set preferences to prioritize the display of electrical-related layers. If it is detected that there is no specific layer display preference in the user account, it will default to the previously generated layer aggregation result as the display method of the annotation results. This means that the system will automatically select the most appropriate layer display method based on the initial global recognition results and the subsequent layer aggregation logic. This approach not only ensures the accuracy and availability of the annotation results, but also improves processing efficiency because it avoids the need for users to manually search for the required information in multiple layers. Through such a processing flow, not only the efficiency and applicability of image annotation are optimized, but also the user experience is enhanced by considering the user's personalized settings. This flexible and intelligent layer management strategy enables the system to maintain efficient data processing while meeting the specific needs of different users in different application scenarios, making it widely applicable to image processing tasks in various professional fields.

[0047] Furthermore, the method further comprises:

[0048] Receive the annotation label selection result of the user, re-cluster the layer of the annotation result according to the selection result, convert the annotation label layer according to the layer clustering result, and display the conversion result in a floating manner.

[0049] If a confirmation command from the user is received in the floating display state, the conversion result is adjusted to a fixed display.

[0050] Furthermore, the system also provides a highly interactive image annotation function, which enables users to dynamically adjust the layers of annotation results according to their personal preferences and needs. Specifically, the system first receives the user's annotation label selection results. This function is similar to "one-click adjustment", allowing users to select or modify the annotation labels they want to emphasize, such as selecting labels for specific elements such as all electrical lines or structural supports. Based on the user's selection, the system automatically re-clustered the layers of the annotation results. This process involves reorganizing the existing annotations to ensure that the content that the user is concerned about is appropriately clustered and highlighted. Subsequently, the layer position of the annotation labels is adjusted according to this new layer clustering result to reflect the user's preferences. For example, if the user chooses to emphasize visual elements such as safety signs, these labels will be placed in the upper layer of the view, making this information more visible. In addition, the results of this label hierarchy adjustment will be displayed in the form of a floating display, providing users with a flexible view, in which the annotation content can be adjusted or confirmed according to further instructions from the user. If the user provides a confirmation command in the floating display state, these adjustments will be fixed, and the conversion result will change from floating display to fixed display. The completion of this step marks the user's final confirmation of the annotation level and details, ensuring that the displayed content accurately reflects the user's needs and preferences. Through such a mechanism, not only the availability and personalized experience of the annotated data are improved, but also the interaction efficiency between the user and the system is greatly enhanced, allowing users to obtain and operate the required information in the most intuitive way when processing complex image data. This dynamic and user-driven annotation process has shown great flexibility and practical value in a variety of application scenarios, especially in professional fields that require precise layer management and highly personalized settings.

[0051] Furthermore, the labeling network is initialized according to the text recognition results, image identification, and global recognition results. The initialization content includes labeling strategy, recognition bias, and also includes:

[0052] A drawing type analysis is performed based on the text recognition results, image identification, and global recognition results, and a first annotation strategy is generated based on the drawing type analysis results.

[0053] Execution standard matching is performed according to the drawing type analysis results, and a second annotation strategy is established based on the execution standard matching results.

[0054] A labeling strategy is established according to the first labeling strategy and the second labeling strategy, and the labeling network is initialized.

[0055] Exemplarily, the initialization of the annotation network is a precise and complex process that involves extracting key information from the text recognition results, image identification, and global recognition results, and using this information to develop a detailed annotation strategy. First, a detailed analysis of the drawing type is performed by analyzing the results of text recognition and the identification of the image (such as information such as source and type), as well as the global recognition results (i.e., the main elements and structures in the image). This analysis helps the system understand the specific nature of the image content, such as determining whether the image is an engineering drawing or other type of document. Based on the analysis results of the drawing type, a first annotation strategy is then generated, which is specifically targeted at the specific type of drawing to ensure that the annotation activity can focus on the key elements and details in the drawing. For example, for mechanical engineering drawings, the first annotation strategy may emphasize the identification of mechanical parts and the accurate annotation of dimension annotations. Subsequently, execution standard matching is performed based on the drawing type analysis results, which involves matching execution standards within a specific industry or field, such as the requirements of ISO standards for engineering drawings, and then establishing a second annotation strategy. The second annotation strategy may involve annotation priority, annotation detail, and areas that require special attention. Finally, the two strategies are combined to form a comprehensive annotation strategy that takes into account multiple factors such as the type of drawing, the importance of the content, and industry standards. After the establishment of this annotation strategy, the annotation network is initialized to set up the network to adapt to specific annotation requirements and optimize the annotation effect. Through this initialization process, the annotation network is ready to handle various complex annotation tasks, ensuring that the final annotation results are both accurate and meet the specific needs of users and industries. This approach not only improves the accuracy and relevance of annotation, but also optimizes the entire image processing process, making each step from image recognition to information extraction to specific annotation customized and efficient.

[0056] Furthermore, the method further comprises:

[0057] When the learning interface is activated, it receives the user's input annotation image, learns the input annotation image, and establishes a temporary memory library.

[0058] The temporary memory library is used to perform annotation recognition within a preset time window to establish a learning annotation result.

[0059] Specifically, in order to enhance the adaptability and learning ability of the automatic annotation system, a dynamic learning interface is introduced, which, when activated, will initiate a series of operations to enhance the system's ability to handle new image annotation tasks. First, the input annotation images provided by the user are received. These images serve as new learning samples. In this way, the user can directly demonstrate to the system a specific annotation method or expected annotation results. This link is the key to achieving personalized learning and rapid adaptation to new tasks. Next, these input annotation images are learned. By analyzing the annotation style, the labels used, and the accuracy of the annotations in the images, the system is able to extract useful information and patterns from these specific examples. This information is then stored in a temporary memory bank, which can store all relevant data learned in a short period of time, allowing the system to quickly review and reference this new knowledge. Then, using the data in the temporary memory bank, the system performs the annotation recognition task within a preset time window. This means that the system will use the information in the memory bank to optimize the annotation process within a defined time period, improving the accuracy and efficiency of the annotation. For example, if images containing a specific type of mechanical parts are repeatedly received in the past few days, and the annotations in these images show a new and more effective annotation method, the system will adopt this new method in similar annotation tasks in the future. This learning mechanism based on temporary memory not only enables the system to better adapt to the specific needs of users, but also continuously improves processing results and reflects the latest developments in the market and technology. Through such a learning and adaptation mechanism, the flexibility and user satisfaction of the annotation system are significantly improved, so that the system can not only handle routine annotation tasks, but also quickly adapt to new annotation challenges, providing users with highly personalized and continuously optimized services.

[0060] Through the technical solutions of the above-mentioned embodiments, the present application provides a method for automatic extraction and annotation of image data, which solves the technical problems existing in image annotation technology, such as insufficient processing accuracy of complex images, lack of necessary user interaction and personalized adjustment, lack of flexibility and adaptability to cope with long-term annotation performance, resulting in reduced annotation efficiency and reliability, and achieves the technical effect of improving the processing and response capability, accuracy, efficiency, system adaptability and flexibility of image annotation, thereby meeting the personalized needs of different users.

[0061] Embodiment 2 is based on the same inventive concept as the method for automatically extracting and annotating image data in the above embodiment. Figure 2 As shown, the present application provides an image data automatic extraction and annotation platform, the platform comprising:

[0062] The image recognition module 11 is used to read the image identifier of the target image, and perform size recognition on the target image content, configure the initial resolution based on the size recognition result of the image content, perform global acquisition of the target image using the initial resolution, and establish a global recognition result.

[0063] The initialization module 12 is used to perform text recognition on the target image and initialize the annotation network according to the text recognition result, image identification and global recognition result. The initialization content includes the annotation strategy and recognition bias.

[0064] The image segmentation module 13 is used to call the geometric figure segmentation unit in the annotation network, perform image segmentation of the target image based on the size recognition result, and establish an associated identifier of the image segmentation result.

[0065] The labeling module 14 is used to synchronize the association identification and image segmentation results to the recognition layer of the labeling network, perform automatic recognition and labeling, and the recognition layer is initialized by the recognition bias and labeling strategy.

[0066] The annotation adjustment module 15 is used to establish the annotation result and generate the annotation label of the annotation result. The annotation label includes a main structure label, an auxiliary reference line label, and a feature type label. The annotation label is a floating index label and is used to adjust the annotation layer of the annotation result.

[0067] The marking display module 16 is used to mark and display the marking results and marking labels.

[0068] Furthermore, the image segmentation module 13 is further configured to perform the following steps:

[0069] Calculate the image content complexity of the target image and establish the complexity distribution result.

[0070] The complexity distribution results and size recognition results are used to establish the segmentation constraint interval of the segmentation block.

[0071] The contour extraction channel in the geometric figure segmentation unit is called to establish a preliminary segmentation result through the segmentation constraint interval and the contour extraction result.

[0072] The preliminary segmentation results are subjected to semantic recognition based on the image content, and the image segmentation results and associated identifiers are established using the semantic recognition results.

[0073] Furthermore, the image segmentation module 13 is further configured to perform the following steps:

[0074] A standard shape database is established, and the preliminary segmentation results are matched with the standard shape database to establish standard shape semantic labels.

[0075] Perform texture feature extraction on the preliminary segmentation results, establish texture feature extraction results, and use the texture feature extraction results to establish texture semantic labels.

[0076] Functional objects are identified on the preliminary segmentation results and functional semantic labels are established.

[0077] Perform semantic recognition of spatial relationships on the preliminary segmentation results and establish spatial relationship semantic labels.

[0078] Integrate standard shape semantic labels, texture semantic labels, functional semantic labels, and spatial relationship semantic labels to establish image segmentation results and associated labels.

[0079] Furthermore, the annotation display module 16 is further configured to perform the following steps:

[0080] Generates the default layer aggregation result of the annotation result based on the global recognition result.

[0081] Read the user account information of the user, and determine whether the user account information has a label preference layer constraint.

[0082] If there is no label preference layer constraint, the default layer aggregation result will be displayed as the label result.

[0083] Furthermore, the annotation display module 16 is further configured to perform the following steps:

[0084] Receive the annotation label selection result of the user, re-cluster the layer of the annotation result according to the selection result, convert the annotation label layer according to the layer clustering result, and display the conversion result in a floating manner.

[0085] If a confirmation command from the user is received in the floating display state, the conversion result is adjusted to a fixed display.

[0086] Furthermore, the initialization module 12 is further configured to perform the following steps:

[0087] A drawing type analysis is performed based on the text recognition results, image identification, and global recognition results, and a first annotation strategy is generated based on the drawing type analysis results.

[0088] Execution standard matching is performed according to the drawing type analysis results, and a second annotation strategy is established based on the execution standard matching results.

[0089] A labeling strategy is established according to the first labeling strategy and the second labeling strategy, and the labeling network is initialized.

[0090] Furthermore, the platform also includes:

[0091] The annotation learning module is used to receive the user's input annotation image when the learning interface is activated, and learn the input annotation image to establish a temporary memory library.

[0092] The learning recognition module is used to use the temporary memory library to perform annotation recognition within a preset time window and establish a learning annotation result.

[0093] Through the above-mentioned detailed description of a method for automatic extraction and annotation of image data in this specification, those skilled in the art can clearly understand an automatic extraction and annotation platform for image data in this embodiment. For the platform disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0094] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatically extracting and annotating image data, characterized in that: The method comprises: Read the image identifier of the target image, and perform size recognition on the target image content, configure the initial resolution based on the size recognition result of the image content, perform global acquisition of the target image using the initial resolution, and establish a global recognition result; Perform text recognition on the target image, and initialize the annotation network according to the text recognition result, image identification, and global recognition result. The initialization content includes the annotation strategy and recognition bias, wherein the recognition bias is a preference setting for the annotation network; Calling the geometric figure segmentation unit in the annotation network, performing image segmentation of the target image based on the size recognition result, and establishing an associated identifier of the image segmentation result; Synchronizing the association identifier and image segmentation result to the recognition layer of the annotation network, performing automatic recognition and annotation, wherein the recognition layer is initialized by the recognition bias and the annotation strategy; Establishing a labeling result and generating a labeling label of the labeling result, wherein the labeling label includes a main structure label, an auxiliary reference line label, and a feature type label. The labeling label is a floating index label and is used to adjust the labeling layer of the labeling result; Annotate and display the annotation results and annotation labels; The labeling network is initialized according to the text recognition results, image identification, and global recognition results. The initialization content includes labeling strategy and recognition bias, specifically including: Perform drawing type analysis based on text recognition results, image identification, and global recognition results, and generate a first annotation strategy based on the drawing type analysis results; Perform execution standard matching according to the drawing type analysis results, and establish a second marking strategy based on the execution standard matching results; A labeling strategy is established according to the first labeling strategy and the second labeling strategy, and the labeling network is initialized.

2. The method for automatically extracting and annotating image data according to claim 1, characterized in that: The calling of the geometric figure segmentation unit in the annotation network, performing image segmentation of the target image based on the size recognition result, and establishing an associated identifier of the image segmentation result specifically includes: Calculate the image content complexity of the target image and establish the complexity distribution result; Using the complexity distribution results and size recognition results, the segmentation constraint interval of the segmentation block is established; Calling the contour extraction channel in the geometric figure segmentation unit to establish a preliminary segmentation result through the segmentation constraint interval and the contour extraction result; The preliminary segmentation results are subjected to semantic recognition based on the image content, and the image segmentation results and associated identifiers are established using the semantic recognition results.

3. The method for automatically extracting and annotating image data according to claim 2, characterized in that: The performing of semantic recognition based on the image content on the preliminary segmentation result and using the semantic recognition result to establish the image segmentation result and the associated identifier specifically includes: Establish a standard shape database, perform association matching on the preliminary segmentation results with the standard shape database, and establish standard shape semantic labels; Extract texture features from the preliminary segmentation results, establish texture feature extraction results, and use the texture feature extraction results to establish texture semantic labels; Perform functional object recognition on the preliminary segmentation results and establish functional semantic labels; Perform semantic recognition of spatial relations on the preliminary segmentation results and establish spatial relationship semantic labels; Integrate standard shape semantic labels, texture semantic labels, functional semantic labels, and spatial relationship semantic labels to establish image segmentation results and associated labels.

4. The method for automatically extracting and annotating image data according to claim 1, characterized in that: The method further comprises: Generate the default layer aggregation result of the annotation result based on the global recognition result; Reading user account information of the user, and determining whether the user account information has a labeling preference layer constraint; If there is no label preference layer constraint, the default layer aggregation result will be displayed as the label result.

5. The method for automatically extracting and annotating image data according to claim 4, characterized in that: The method further comprises: Receive the annotation label selection result of the user, re-cluster the layers of the annotation result according to the selection result, convert the annotation label layer according to the layer clustering result, and display the conversion result in a floating manner; If a confirmation command from the user is received in the floating display state, the conversion result is adjusted to a fixed display.

6. The method for automatically extracting and annotating image data according to claim 1, characterized in that: The method further comprises: When the learning interface is activated, it receives the input annotation image from the user, learns the input annotation image, and establishes a temporary memory bank; The temporary memory library is used to perform annotation recognition within a preset time window to establish a learning annotation result.

7. An image data automatic extraction and annotation platform, characterized in that: Used to implement the method for automatically extracting and annotating image data according to any one of claims 1 to 6, the platform comprises: An image recognition module is used to read the image identifier of the target image, and perform size recognition on the target image content, configure the initial resolution based on the size recognition result of the image content, perform global acquisition of the target image using the initial resolution, and establish a global recognition result; An initialization module is used to perform text recognition on the target image and initialize the annotation network according to the text recognition result, image identification, and global recognition result. The initialization content includes the annotation strategy and recognition bias, wherein the recognition bias is a preference setting for the annotation network; An image segmentation module is used to call the geometric figure segmentation unit in the annotation network, perform image segmentation of the target image based on the size recognition result, and establish an associated identifier of the image segmentation result; A labeling module, used to synchronize the association identifier and image segmentation results to the recognition layer of the labeling network, perform automatic recognition and labeling, and the recognition layer is initialized by the recognition bias and labeling strategy; A label adjustment module, used to establish a labeling result and generate a labeling label of the labeling result, wherein the labeling label includes a main structure label, an auxiliary reference line label, and a feature type label. The labeling label is a floating index label and is used to adjust the labeling layer of the labeling result; The annotation display module is used to annotate and display the annotation results and annotation labels.

Images