Express sheet image text data set integration method and device, equipment and storage medium

Abstract

The invention relates to the technical field of logistics, in particular to a waybill image text data set integration method and device, equipment and a storage medium, and the method comprises the steps: obtaining an express waybill image set on a sorting line, and carrying out the correction processing of the express waybill image set, so as to obtain a plurality of corrected express waybill images; performing type analysis on each corrected express sheet image to obtain a corresponding image type; labeling the corresponding corrected express sheet image by adopting a corresponding preset labeling condition according to the image type to obtain a labeled image; performing text segmentation on the annotated image to obtain a segmented text image; obtaining a preset bad scene type, and performing bad processing on the segmented text image by adopting a corresponding text processing model according to the bad scene type to obtain a text data set; through the method, an efficient and accurate express sheet image processing system can be constructed, the processing speed and the data quality are improved, and diversified data samples are provided for the express logistics industry.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and in particular to a method, device, equipment and storage medium for integrating a waybill image and text data set. Background Art

[0002] Existing express delivery labels often have poor imaging problems when they are captured, resulting in a large number of deformed, bent, blurred and stained text images in the captured images; or a large number of low-quality text sequences, which makes text recognition very difficult; in addition, the logistics scenario includes many domestic express delivery companies, and each company has many express delivery label formats, each format corresponding to a different text layout, which makes it difficult to achieve consistency in text recognition and lacks diverse training samples.

[0003] It can be seen that the existing technology still needs to be improved and enhanced. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the object of the present invention is to provide a method, device, equipment and storage medium for integrating a delivery bill image text data set, which can generate recognition text of defective images in multiple scenarios based on ordinary express delivery bill images.

[0005] The first aspect of the present invention provides a method for integrating a delivery bill image and text data set, comprising: obtaining a set of express delivery bill images on a sorting line, and performing correction processing on the express delivery bill image set to obtain a plurality of corrected express delivery bill images; performing type analysis on each corrected express delivery bill image to obtain a corresponding image type; annotating the corresponding corrected express delivery bill image according to the image type using a corresponding preset annotation condition to obtain an annotated image; performing text segmentation on the annotated image to obtain a segmented text image; obtaining a preset adverse scene type, and performing adverse processing on the segmented text image using a corresponding text processing model according to the adverse scene type to obtain a text data set.

[0006] Optionally, in a first implementation method of the first aspect of the present invention, the method of acquiring a set of express delivery bill images on the sorting line and performing correction processing on the express delivery bill image set to obtain a plurality of corrected express delivery bill images includes: acquiring corresponding express delivery bill images from the sorting line according to preset frame rate parameters and forming an express delivery bill image set; denoising each image in the express delivery bill image set to obtain a plurality of denoised images; and performing deskew and trimming processing on each denoised image to obtain a plurality of corrected express delivery bill images.

[0007] Optionally, in a second implementation method of the first aspect of the present invention, the type analysis of each corrected express delivery bill image to obtain a corresponding image type includes: extracting edge features of each corrected express delivery bill image, and determining the image curvature and image slope of the corresponding corrected express delivery bill image based on the edge features; analyzing the plane features of the corresponding corrected express delivery bill image based on the image curvature and image slope, and determining the plane type based on the plane features; extracting layout features of each corrected express delivery bill image, and determining the layout type based on the layout features; and determining the image type of the corresponding corrected express delivery bill image based on the plane type and the layout type.

[0008] Optionally, in a third implementation method of the first aspect of the present invention, the corresponding corrected express delivery bill image is annotated according to the image type using corresponding preset annotation conditions to obtain an annotated image, including: the image type includes an inclined text type, a curved text type and a non-standard layout type; when the image type is an inclined text type, the annotated image is obtained by selecting from the corrected express delivery bill image according to the preset inclined annotation conditions; when the image type is a curved text type, the continuous edge trajectory of the text area is obtained from the corrected express delivery bill image according to the preset curved annotation conditions, and the annotated image is obtained according to the continuous edge trajectory; when the image type is a non-standard layout type, the corresponding annotation reference information is obtained from a preset layout database according to the non-standard layout type, and the corrected express delivery bill image is annotated according to the annotation reference information to obtain the annotated image.

[0009] Optionally, in a fourth implementation manner of the first aspect of the present invention, performing text segmentation on the annotated image to obtain a segmented text image includes: constructing a multilingual text segmentation model based on the CRAFT text detection model; inputting the annotated image into the multilingual text segmentation model, and predicting the bounding box of the text line in the annotated image through the multilingual text segmentation model to obtain a predicted text image; and normalizing the predicted text image according to a preset image format to obtain a segmented text image.

[0010] Optionally, in a fifth implementation manner of the first aspect of the present invention, the annotated image is input into a multilingual text segmentation model, and the bounding box of the text line in the annotated image is predicted by the multilingual text segmentation model to obtain a predicted text image, which also includes: obtaining the resolution of the annotated image, and comparing the resolution with a preset minimum resolution threshold; if the resolution is lower than the minimum resolution threshold, marking the annotated image as a low-quality image; and reconstructing the annotated image based on super-resolution technology to obtain an annotated image with high resolution.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, a preset bad scene type is obtained, and a corresponding text processing model is used to perform bad processing on the segmented text image according to the bad scene type to obtain a text data set, including: the bad scene types include scene blur type, stroke blur type, dynamic blur type and stroke defect type; when the bad scene type is a scene blur type, the segmented text image is blurred based on a Gaussian mixture model to obtain a text data set; when the bad scene type is a stroke blur type, the segmented text image is fused based on a graphic dynamics algorithm to obtain a text data set; when the bad scene type is a dynamic blur type, the segmented text image is dynamically shifted based on a dynamic convolution model to obtain a text data set; when the bad scene type is a stroke defect type, the segmented text image is randomly eroded based on a morphological algorithm to obtain a text data set.

[0012] The second aspect of the present invention provides a device for integrating express delivery bill image and text data set, comprising: a correction module, used to obtain a set of express delivery bill images on a sorting line, and perform correction processing on the express delivery bill image set to obtain multiple corrected express delivery bill images; an analysis module, used to perform type analysis on each corrected express delivery bill image to obtain a corresponding image type; a labeling module, used to label the corresponding corrected express delivery bill image according to the image type using a corresponding preset labeling condition to obtain a labeled image; a segmentation module, used to perform text segmentation on the labeled image to obtain a segmented text image; a processing module, used to obtain a preset adverse scene type, and perform adverse processing on the segmented text image using a corresponding text processing model according to the adverse scene type to obtain a text data set.

[0013] Optionally, in a first implementation method of the second aspect of the present invention, the correction module includes: a frame rate unit, used to obtain corresponding express delivery bill images from the sorting line according to preset frame rate parameters, and form an express delivery bill image set; a denoising unit, used to perform denoising processing on each image in the express delivery bill image set to obtain multiple denoised images; and a correction unit, used to perform deskew and trimming processing on each denoised image to obtain multiple corrected express delivery bill images.

[0014] Optionally, in a second implementation method of the second aspect of the present invention, the analysis module includes: an edge unit, used to extract edge features of each corrected express delivery bill image, and determine the image curvature and image slope of the corresponding corrected express delivery bill image based on the edge features; a plane unit, used to analyze the plane features of the corresponding corrected express delivery bill image based on the image curvature and image slope, and determine the plane type based on the plane features; a layout unit, used to extract layout features of each corrected express delivery bill image, and determine the layout type based on the layout features; a determination unit, used to determine the image type of the corresponding corrected express delivery bill image based on the plane type and the layout type.

[0015] Optionally, in a third implementation method of the second aspect of the present invention, the annotation module includes: a tilting unit, which is used to, when the image type is a tilted text type, select from the corrected express delivery bill image according to a preset tilt annotation condition to obtain an annotated image; a bending unit, which is used to, when the image type is a curved text type, obtain the continuous edge trajectory of the text area from the corrected express delivery bill image according to a preset curved annotation condition, and obtain the annotated image based on the continuous edge trajectory; a non-standard unit, which is used to, when the image type is a non-standard layout type, obtain corresponding annotation reference information from a preset layout database according to the non-standard layout type, and annotate the corrected express delivery bill image according to the annotation reference information to obtain the annotated image.

[0016] Optionally, in a fourth implementation manner of the second aspect of the present invention, the segmentation module includes: a construction unit, used to construct a multilingual text segmentation model based on the CRAFT text detection model; a prediction unit, used to input the annotated image into the multilingual text segmentation model, and predict the bounding box of the text line in the annotated image through the multilingual text segmentation model to obtain a predicted text image; and a standardization unit, used to standardize the predicted text image according to a preset image format to obtain a segmented text image.

[0017] Optionally, in a fifth implementation of the second aspect of the present invention, the segmentation module also includes: a comparison unit, used to obtain the resolution of the annotated image and compare the resolution with a preset minimum resolution threshold; a marking unit, used to mark the annotated image as a low-quality image if the resolution is lower than the minimum resolution threshold; and a reconstruction unit, used to reconstruct the annotated image based on super-resolution technology to obtain an annotated image with high resolution.

[0018] Optionally, in a sixth implementation of the second aspect of the present invention, the processing module includes: a scene unit, which is used to blur the segmented text image based on a Gaussian mixture model when the bad scene type is a scene blur type to obtain a text data set; a stroke unit, which is used to fuse the segmented text image based on a graphics dynamics algorithm when the bad scene type is a stroke blur type to obtain a text data set; a dynamic unit, which is used to dynamically shift the segmented text image based on a dynamic convolution model when the bad scene type is a dynamic blur type to obtain a text data set; a defect unit, which is used to randomly corrode the segmented text image based on a morphological algorithm when the bad scene type is a stroke defect type to obtain a text data set.

[0019] The third aspect of the present invention provides a face sheet image and text dataset integration device, which comprises: a memory and at least one processor, wherein the memory stores instructions; at least one of the processors calls the instructions in the memory to enable the face sheet image and text dataset integration device to perform each step of the face sheet image and text dataset integration method described above.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, the various steps of any of the above-mentioned methods for integrating a face sheet image text dataset are implemented.

[0021] In the technical solution of the present invention, by acquiring the express delivery bill image set on the sorting line and performing correction processing, it can be ensured that the subsequent processing steps are based on clear and accurate images, reducing erroneous recognition caused by image quality problems; type analysis and annotation of the corrected express delivery bill image can greatly reduce manual intervention, improve processing efficiency, and ensure the consistency of annotation at the same time. Corresponding annotation operations can be performed according to different types of images, greatly improving the accuracy of annotation; through text segmentation, text information can be more accurately extracted from the express delivery bill image, the effective area of ​​the image can be increased, and a clearer text image can be provided for subsequent data processing and analysis; by acquiring preset bad scene types and applying corresponding text processing models, the segmented text image can be processed accordingly according to the bad scene type, so that the text image simulates defective images, such as blur, occlusion, missing strokes, etc., to ensure that a text data set with diversity is finally obtained, and the diversity of further data analysis and application is improved; the above method helps to build an efficient and accurate express delivery bill image processing system, which not only improves the processing speed, but also ensures the data quality, and provides a variety of data samples for the express logistics industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1A first flow chart of a method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0023] Figure 2 A second flow chart of the method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0024] Figure 3 A third flow chart of the method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0025] Figure 4 A fourth flow chart of a method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0026] Figure 5 A fifth flow chart of the method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0027] Figure 6 A sixth flow chart of the method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0028] Figure 7 A seventh flow chart of the method for integrating a face sheet image and text dataset provided by an embodiment of the present invention;

[0029] Figure 8 A schematic diagram of the structure of a device for integrating a face sheet image and text data set provided by an embodiment of the present invention;

[0030] Fig. 9 Another structural schematic diagram of a device for integrating a single image and text data set provided by an embodiment of the present invention;

[0031] Fig.10 A schematic diagram of the structure of a device for integrating a face sheet image and text data set provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention provides a method, device, equipment and storage medium for integrating a delivery bill image and text data set. By acquiring an express delivery bill image set on a sorting line and performing correction processing, the present invention can ensure that subsequent processing steps are performed based on clear and accurate images, thereby reducing erroneous recognition caused by image quality problems. By performing type analysis and annotation on the corrected express delivery bill image, manual intervention can be greatly reduced, processing efficiency can be improved, and at the same time, the consistency of annotation can be ensured. Corresponding annotation operations can be performed according to different types of images, greatly improving the accuracy of annotation. By text segmentation, text information can be more accurately extracted from the express delivery bill image, the effective area of ​​the image can be increased, and a clearer text image can be provided for subsequent data processing and analysis. By acquiring a preset bad scene type and applying a corresponding text processing model, the segmented text image can be processed accordingly according to the bad scene type, so that the text image simulates defective images, such as blur, occlusion, missing strokes, etc., to ensure that a text data set with diversity is finally obtained, thereby improving the diversity of further data analysis and application. The above method is helpful to build an efficient and accurate express delivery bill image processing system, which not only improves the processing speed, but also ensures the data quality, and provides a variety of data samples for the express logistics industry.

[0033] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the method for integrating a single image text dataset in an embodiment of the present invention includes:

[0035] 101. Acquire a set of express delivery bill images on a sorting line, and perform correction processing on the express delivery bill image set to obtain a plurality of corrected express delivery bill images;

[0036] In this embodiment, by acquiring the express delivery bill image set on the sorting line and performing correction processing, it can be ensured that the subsequent processing steps are based on clear and accurate images, thereby reducing erroneous recognition caused by image quality problems.

[0037] 102. Perform type analysis on each corrected express delivery bill image to obtain a corresponding image type;

[0038] 103. Annotate the corresponding corrected express delivery bill image using corresponding preset annotation conditions according to the image type to obtain an annotated image;

[0039] 104. Perform text segmentation on the annotated image to obtain a segmented text image;

[0040] In this embodiment, type analysis and annotation of the corrected express delivery bill image can significantly reduce manual intervention, improve processing efficiency, and ensure the consistency of annotation. Corresponding annotation operations can be performed according to different types of images, greatly improving the accuracy of annotation. Text segmentation can be used to more accurately extract text information from the express delivery bill image, increase the effective area of ​​the image, and provide a clearer text image for subsequent data processing and analysis.

[0041] 105. Obtain a preset bad scene type, and use a corresponding text processing model to perform bad processing on the segmented text image according to the bad scene type to obtain a text data set.

[0042] In this embodiment, by obtaining preset adverse scene types (such as blurred fonts, missing fonts, etc.) and applying corresponding text processing models, the segmented text image can be processed accordingly according to the adverse scene type, so that the text image can be simulated to form defective images and texts, such as blurring, occlusion, missing strokes, etc., to ensure that a diverse text data set is finally obtained, thereby improving the diversity of further data analysis and applications.

[0043] In the embodiment of the present invention, by acquiring the express delivery bill image set on the sorting line and performing correction processing, it can be ensured that the subsequent processing steps are based on clear and accurate images, reducing erroneous recognition caused by image quality problems; type analysis and annotation of the corrected express delivery bill image can greatly reduce manual intervention, improve processing efficiency, and ensure the consistency of annotation at the same time. Corresponding annotation operations can be performed according to different types of images, greatly improving the accuracy of annotation; through text segmentation, text information can be more accurately extracted from the express delivery bill image, the effective area of ​​the image is increased, and a clearer text image is provided for subsequent data processing and analysis; by acquiring preset bad scene types and applying corresponding text processing models, the segmented text image can be processed accordingly according to the bad scene type, so that the text image simulates defective images, such as blur, occlusion, missing strokes, etc., to ensure that a text data set with diversity is finally obtained, and the diversity of further data analysis and application is improved; the above method helps to build an efficient and accurate express delivery bill image processing system, which not only improves the processing speed, but also ensures the data quality, and provides a variety of data samples for the express logistics industry.

[0044] See also Figure 2 , a second embodiment of the method for integrating a single image text dataset in an embodiment of the present invention includes:

[0045] 201. Acquire corresponding express delivery bill images from the sorting line according to preset frame rate parameters, and form an express delivery bill image set;

[0046] 202. Perform denoising on each image in the express delivery bill image set to obtain a plurality of denoised images;

[0047] 203. Perform deskew and trimming processing on each denoised image to obtain multiple corrected express delivery bill images.

[0048] In this embodiment, the corresponding express delivery bill image is obtained from the sorting line according to the preset frame rate parameters, and the express delivery bill image set is formed, which helps to ensure the continuity and stability of image acquisition. Denoising each image in the express delivery bill image set can improve the image quality, reduce blur and interference, so that the text and information on the bill are clearer and more legible. Correcting and trimming each denoised image can further optimize the image, ensure that each part of the express delivery bill is in a suitable position and angle, facilitate subsequent information extraction and processing, and improve sorting efficiency and accuracy. For denoising, algorithms such as median filtering and Gaussian filtering can be used to remove salt and pepper noise, Gaussian noise, etc. in the image. For correction, algorithms such as Hough transform can be used to detect straight lines in the image, calculate the tilt angle of the image, and then perform rotation correction. The cropping operation is based on preset rules, such as retaining the bill area and removing the background part, and is implemented through coordinate positioning and image cropping functions.

[0049] See also Figure 3 The third embodiment of the method for integrating a single image text dataset in the embodiment of the present invention includes:

[0050] 301. Extract edge features of each corrected express delivery bill image, and determine image curvature and image slope of the corresponding corrected express delivery bill image according to the edge features;

[0051] 302. Analyze the plane features of the corresponding corrected express delivery order image according to the image curvature and the image slope, and determine the plane type according to the plane features;

[0052] 303. Extracting layout features of each corrected express delivery bill image, and determining the layout type according to the layout features;

[0053] 304. Determine the image type of the corresponding corrected express delivery label image according to the plane type and the layout type.

[0054] In this embodiment, by extracting edge features, the curvature and slope of the image can be more accurately identified and corrected, thereby reducing information reading errors caused by image distortion and improving the accuracy of express delivery bill image processing; automated analysis of plane features and layout features can quickly determine the type of express delivery bill and improve overall processing efficiency; by determining the image type, key information on the express delivery bill, such as recipient information, address, etc., can be more effectively extracted, reducing manual intervention, lowering the error rate, and improving the reliability of data extraction; accurately classified express delivery bill images can be more efficiently stored and retrieved, facilitating subsequent management and query work.

[0055] See also Figure 4 The fourth embodiment of the method for integrating a single image text dataset in the embodiment of the present invention includes:

[0056] The image types include oblique text types, curved text types, and non-standard layout types;

[0057] 401. When the image type is an inclined text type, a labeled image is selected from the corrected express delivery bill image according to a preset inclined labeling condition;

[0058] 402. When the image type is a curved text type, a continuous edge trajectory of the text area is obtained from the corrected express delivery bill image according to a preset curved annotation condition, and an annotation image is obtained according to the continuous edge trajectory;

[0059] 403. When the image type is a non-standard layout type, corresponding annotation reference information is obtained from a preset layout database according to the non-standard layout type, and the corrected express delivery label image is annotated according to the annotation reference information to obtain an annotated image.

[0060] In this embodiment, for images of inclined text types, the annotated image is obtained by selecting the inclined annotation condition, and the four top corners of the text area are captured to form a rectangular frame area, so as to obtain the annotated image. In this way, the readability of the text can be improved, which is convenient for subsequent image processing and information extraction; for images of curved text types, the annotated image is obtained by obtaining the continuous edge trajectory of the text area, which helps to correct the bending deformation of the text and ensure the accuracy of the text information; for images of non-standard layout types, the non-standard layout can be effectively annotated by obtaining the corresponding annotation reference information from the preset layout database, so that the image processing system can adapt to more diverse express delivery bill formats and improve the processing flexibility and accuracy.

[0061] See also Figure 5 The fifth embodiment of the method for integrating a single image text dataset in the embodiment of the present invention includes:

[0062] 501. Build a multilingual text segmentation model based on the CRAFT text detection model;

[0063] 502. Input the annotated image into a multilingual text segmentation model, and predict the bounding box of the text line in the annotated image by the multilingual text segmentation model to obtain a predicted text image;

[0064] 503. Perform standardization processing on the predicted text image according to a preset image format to obtain a segmented text image.

[0065] In this embodiment, by using the CRAFT model, the text area in the image can be detected more accurately, thereby improving the quality of text segmentation and the accuracy of text detection. The model can process text in multiple languages, so that text segmentation technology can be applied to a wider range of scenarios and language environments; by automatically predicting the bounding box of the text line, a clear text image can be quickly extracted from a complex image background, simplifying the subsequent image processing steps; the predicted text image is standardized to ensure that the output text image conforms to the preset image format, which is convenient for further analysis, storage or display.

[0066] See also Figure 6 The sixth embodiment of the method for integrating a single image text dataset in the embodiment of the present invention includes:

[0067] 601. Obtain the resolution of the annotated image, and compare the resolution with a preset minimum resolution threshold;

[0068] 602. If the resolution is lower than the minimum resolution threshold, mark the annotated image as a low-quality image;

[0069] 603. Reconstruct the annotated image based on super-resolution technology to obtain an annotated image with high resolution.

[0070] In this embodiment, by obtaining the resolution of the annotated image and comparing it with the preset minimum resolution threshold, low-quality images can be effectively identified. When the image resolution is lower than the set minimum threshold, it is marked as a low-quality image, which helps maintain the overall quality of the data set. Subsequently, these low-quality images are reconstructed using super-resolution technology, which can significantly improve the resolution of the image, thereby obtaining a clearer and more detailed annotated image. This process not only improves the visual effect of the image, but also may enhance the accuracy of subsequent image processing tasks, such as image recognition, classification, and analysis.

[0071] See also Figure 7 The seventh embodiment of the method for integrating a single image text dataset in the embodiment of the present invention includes:

[0072] The bad scene types include scene blur type, stroke blur type, dynamic blur type and stroke defect type;

[0073] 701. When the undesirable scene type is a scene blur type, blurring the segmented text image based on a Gaussian mixture model to obtain a text data set;

[0074] 702. When the undesirable scene type is a stroke blur type, the segmented text images are fused based on a graphic dynamics algorithm to obtain a text data set;

[0075] 703. When the adverse scene type is a dynamic blur type, dynamically shifting the segmented text image based on a dynamic convolution model to obtain a text data set;

[0076] 704. When the bad scene type is a stroke defect type, the segmented text image is randomly eroded based on a morphological algorithm to obtain a text data set.

[0077] In this embodiment, the segmented text image is blurred by the Gaussian mixture model, which can simulate the blurring of scenes in the real world, thereby generating a more realistic blurred text data set. This helps to train and test the robustness and accuracy of image processing algorithms in the face of blurred scenes; the segmented text image is fused using a graphic dynamics algorithm to create a text data set containing blurred strokes. This data set is particularly useful for developing and optimizing algorithms that can recognize and process text with blurred strokes; the segmented text image is dynamically shifted based on a dynamic convolution model to generate a dynamically blurred text data set. Such a data set is very valuable for training and evaluating algorithms that can process motion blurred images; the segmented text image is randomly eroded by a morphological algorithm to create a text data set containing stroke defects. This data set is crucial to improving the recognition accuracy and robustness of text recognition algorithms in the face of stroke defects.

[0078] The above describes the method for integrating a single image text dataset in an embodiment of the present invention. The following describes the device for integrating a single image text dataset in an embodiment of the present invention. Figure 8 In one embodiment of the present invention, a device for integrating a single image and text data set includes:

[0079] The correction module 801 is used to obtain the express delivery bill image set on the sorting line, and perform correction processing on the express delivery bill image set to obtain multiple corrected express delivery bill images;

[0080] An analysis module 802 is used to perform type analysis on each corrected express delivery bill image to obtain a corresponding image type;

[0081] The marking module 803 is used to mark the corresponding corrected express delivery bill image using corresponding preset marking conditions according to the image type to obtain a marked image;

[0082] The segmentation module 804 is used to perform text segmentation on the annotated image to obtain a segmented text image;

[0083] The processing module 805 is used to obtain a preset bad scene type, and perform bad processing on the segmented text image using a corresponding text processing model according to the bad scene type to obtain a text data set.

[0084] In this embodiment, the correction module 801 can ensure that subsequent processing steps are based on clear and accurate images by acquiring the express delivery bill image set on the sorting line and performing correction processing, thereby reducing erroneous recognition caused by image quality problems; the analysis module 802 and the annotation module 803 perform type analysis and annotation on the corrected express delivery bill image, which can greatly reduce manual intervention and improve processing efficiency, while ensuring the consistency of annotation, and can perform corresponding annotation operations according to different types of images, greatly improving the accuracy of annotation; the segmentation module 8034 can more accurately extract text information from the express delivery bill image through text segmentation, increase the effective area of ​​the image, and provide a clearer text image for subsequent data processing and analysis; the processing module 805 can perform corresponding processing on the segmented text image according to the type of bad scene by acquiring the preset bad scene type and applying the corresponding text processing model, so that the text image simulates defective images, such as blur, occlusion, missing strokes, etc., to ensure that a diverse text data set is finally obtained, thereby improving the diversity of further data analysis and application.

[0085] See also Fig. 9 Another embodiment of the apparatus for integrating a single image and text data set in the embodiment of the present invention includes:

[0086] The correction module 801 is used to obtain the express delivery bill image set on the sorting line, and perform correction processing on the express delivery bill image set to obtain multiple corrected express delivery bill images;

[0087] An analysis module 802 is used to perform type analysis on each corrected express delivery bill image to obtain a corresponding image type;

[0088] The marking module 803 is used to mark the corresponding corrected express delivery bill image using corresponding preset marking conditions according to the image type to obtain a marked image;

[0089] The segmentation module 804 is used to perform text segmentation on the annotated image to obtain a segmented text image;

[0090] The processing module 805 is used to obtain a preset bad scene type, and perform bad processing on the segmented text image using a corresponding text processing model according to the bad scene type to obtain a text data set.

[0091] In this embodiment, in the first implementation method of the second aspect of the present invention, the correction module 801 includes: a frame rate unit 8011, which is used to obtain the corresponding express delivery bill image from the sorting line according to the preset frame rate parameters, and form an express delivery bill image set; a denoising unit 8012, which is used to perform denoising processing on each image in the express delivery bill image set to obtain a plurality of denoised images; a correction unit 8013, which is used to perform deflection correction and trimming processing on each denoised image to obtain a plurality of corrected express delivery bill images.

[0092] In this embodiment, in a second implementation method of the second aspect of the present invention, the analysis module 802 includes: an edge unit 8021, which is used to extract edge features of each corrected express delivery bill image, and determine the image curvature and image slope of the corresponding corrected express delivery bill image based on the edge features; a plane unit 8022, which is used to analyze the plane features of the corresponding corrected express delivery bill image based on the image curvature and image slope, and determine the plane type based on the plane features; a layout unit 8023, which is used to extract layout features of each corrected express delivery bill image, and determine the layout type based on the layout features; a determination unit 8024, which is used to determine the image type of the corresponding corrected express delivery bill image based on the plane type and the layout type.

[0093] In this embodiment, in a third implementation method of the second aspect of the present invention, the annotation module 803 includes: a tilting unit 8031, which is used to, when the image type is a tilted text type, select from the corrected express delivery bill image according to a preset tilt annotation condition to obtain an annotated image; a bending unit 8032, which is used to, when the image type is a curved text type, obtain the continuous edge trajectory of the text area from the corrected express delivery bill image according to the preset curved annotation condition, and obtain the annotated image according to the continuous edge trajectory; a non-standard unit 8033, which is used to, when the image type is a non-standard layout type, obtain corresponding annotation reference information from a preset layout database according to the non-standard layout type, and annotate the corrected express delivery bill image according to the annotation reference information to obtain the annotated image.

[0094] In this embodiment, in a fourth implementation of the second aspect of the present invention, the segmentation module 804 includes: a construction unit 8041, which is used to construct a multilingual text segmentation model based on the CRAFT text detection model; a prediction unit 8042, which is used to input the annotated image into the multilingual text segmentation model, and predict the bounding box of the text line in the annotated image through the multilingual text segmentation model to obtain a predicted text image; a standardization unit 8043, which is used to standardize the predicted text image according to a preset image format to obtain a segmented text image.

[0095] In this embodiment, in a fifth implementation of the second aspect of the present invention, the segmentation module 804 also includes: a comparison unit 8044, used to obtain the resolution of the annotated image, and compare the resolution with a preset minimum resolution threshold; a marking unit 8045, used to mark the annotated image as a low-quality image if the resolution is lower than the minimum resolution threshold; a reconstruction unit 8046, used to reconstruct the annotated image based on super-resolution technology to obtain an annotated image with high resolution.

[0096] In this embodiment, in the sixth implementation method of the second aspect of the present invention, the processing module 805 includes: a scene unit 8051, which is used to blur the segmented text image based on a Gaussian mixture model when the bad scene type is a scene blur type to obtain a text data set; a stroke unit 8052, which is used to fuse the segmented text image based on a graphic dynamics algorithm when the bad scene type is a stroke blur type to obtain a text data set; a dynamic unit 8053, which is used to dynamically shift the segmented text image based on a dynamic convolution model when the bad scene type is a dynamic blur type to obtain a text data set; a defect unit 8054, which is used to randomly erode the segmented text image based on a morphological algorithm when the bad scene type is a stroke defect type to obtain a text data set.

[0097] above Figure 8 and Fig. 9 The face sheet image text data set integration device in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the face sheet image text data set integration device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0098] Fig.101 is a schematic diagram of the structure of a face-single image text data set integration device provided by an embodiment of the present invention. The face-single image text data set integration device 900 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. Among them, the memory 920 and the storage medium 930 can be short-term storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the face-single image text data set integration device 900. Furthermore, the processor 910 can be configured to communicate with the storage medium 930, and execute a series of instruction operations in the storage medium 930 on the face-single image text data set integration device 900 to implement the steps of the face-single image text data set integration method provided in the above-mentioned method embodiments.

[0099] The single image text data set integration device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Fig.10 The structure of the single-image-text data set integration device shown does not constitute a limitation on the single-image-text data set integration device, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0100] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for integrating a face sheet image text dataset.

[0101] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0102] 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 storage medium. Based on this understanding, the technical solution of the present invention is essentially 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 storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0103] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for integrating a face sheet image and text dataset, characterized in that: include: Acquire a set of express delivery bill images on a sorting line, and perform correction processing on the express delivery bill image set to obtain a plurality of corrected express delivery bill images; Perform type analysis on each corrected express delivery bill image to obtain the corresponding image type; Annotate the corresponding corrected express delivery bill image using corresponding preset annotation conditions according to the image type to obtain an annotated image; Perform text segmentation on the annotated image to obtain a segmented text image; A preset bad scene type is obtained, and a corresponding text processing model is used to perform bad processing on the segmented text image according to the bad scene type to obtain a text data set.

2. The method for integrating a single image and text dataset according to claim 1, characterized in that: The step of acquiring a set of express delivery bill images on a sorting line and performing correction processing on the express delivery bill image set to obtain a plurality of corrected express delivery bill images includes: Acquire corresponding express delivery bill images from the sorting line according to preset frame rate parameters, and form an express delivery bill image set; De-noising each image in the express delivery bill image set to obtain multiple de-noised images; Each denoised image is subjected to deskew and trimming processing to obtain multiple corrected express delivery bill images.

3. The method for integrating a single image and text dataset according to claim 1, characterized in that: The type analysis of each corrected express delivery bill image to obtain the corresponding image type includes: Extract edge features of each corrected express delivery bill image, and determine the image curvature and image slope of the corresponding corrected express delivery bill image according to the edge features; Analyze the plane features of the corresponding modified express delivery order image according to the image curvature and image slope, and determine the plane type according to the plane features; Extracting layout features of each corrected express delivery bill image, and determining the layout type according to the layout features; The image type of the corresponding corrected express delivery label image is determined according to the plane type and layout type.

4. The method for integrating a single image and text dataset according to claim 1, characterized in that: The step of labeling the corresponding corrected express delivery bill image using corresponding preset labeling conditions according to the image type to obtain a labeled image includes: The image types include oblique text types, curved text types, and non-standard layout types; When the image type is an inclined text type, a labeled image is selected from the corrected express delivery bill image according to the preset inclined labeling condition; When the image type is a curved text type, a continuous edge track of the text area is obtained from the corrected express delivery bill image according to a preset curved annotation condition, and an annotated image is obtained according to the continuous edge track; When the image type is a non-standard layout type, corresponding annotation reference information is obtained from a preset layout database according to the non-standard layout type, and the corrected express delivery bill image is annotated according to the annotation reference information to obtain an annotated image.

5. The method for integrating a single image and text dataset according to claim 1, characterized in that: The performing text segmentation on the annotated image to obtain a segmented text image includes: Build a multilingual text segmentation model based on the CRAFT text detection model; Input the annotated image into the multilingual text segmentation model, and use the multilingual text segmentation model to predict the bounding box of the text line in the annotated image to obtain a predicted text image; The predicted text image is standardized according to a preset image format to obtain a segmented text image.

6. The method for integrating a face sheet image and text dataset according to claim 5, characterized in that: The method of inputting the annotated image into a multilingual text segmentation model and predicting the bounding box of the text line in the annotated image by the multilingual text segmentation model to obtain a predicted text image also includes: Obtaining the resolution of the annotated image and comparing the resolution with a preset minimum resolution threshold; If the resolution is lower than the minimum resolution threshold, the annotated image is marked as a low-quality image; The annotated image is reconstructed based on super-resolution technology to obtain an annotated image with high resolution.

7. The method for integrating a single image and text dataset according to claim 1, characterized in that: The step of obtaining a preset bad scene type and performing bad processing on the segmented text image using a corresponding text processing model according to the bad scene type to obtain a text data set includes: The bad scene types include scene blur type, stroke blur type, dynamic blur type and stroke defect type; When the undesirable scene type is a scene blur type, the segmented text image is blurred based on a Gaussian mixture model to obtain a text dataset; When the adverse scene type is the stroke blur type, the segmented text images are fused based on the graphic dynamics algorithm to obtain a text data set; When the adverse scene type is dynamic blur type, the segmented text image is dynamically shifted based on the dynamic convolution model to obtain a text dataset; When the bad scene type is a stroke defect type, the segmented text image is randomly eroded based on the morphological algorithm to obtain a text dataset.

8. A device for integrating a face sheet image and text data set, characterized in that: include: A correction module is used to obtain a set of express delivery bill images on the sorting line, and perform correction processing on the express delivery bill image set to obtain multiple corrected express delivery bill images; An analysis module is used to perform type analysis on each corrected express delivery bill image to obtain a corresponding image type; A labeling module, used to label the corresponding corrected express delivery bill image using corresponding preset labeling conditions according to the image type to obtain a labeled image; A segmentation module, used for performing text segmentation on the annotated image to obtain a segmented text image; The processing module is used to obtain a preset bad scene type, and use a corresponding text processing model to perform bad processing on the segmented text image according to the bad scene type to obtain a text data set.

9. A device for integrating a single image and text data set, characterized in that: The face sheet image text data set integration device comprises: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors calls the instructions in the memory so that the single image text dataset integration device performs each step of the single image text dataset integration method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for integrating a single image text dataset as described in any one of claims 1 to 7 are implemented.

Images