Internet data processing methods, devices, and storage media based on artificial intelligence

Abstract

This invention relates to the field of image data processing, and more particularly to an artificial intelligence-based internet data processing method, apparatus, and storage medium. The invention reads the information to be identified, performs spelling detection, filters confidence image region samples, constructs a structural observation box, determines the distribution pattern characteristics corresponding to the confidence image region samples based on the structural observation box, and then constructs a user-specific user profile library based on the correlation between text and distribution pattern characteristics. During the recognition process, abnormal image regions and potentially related text are identified, and the user profile library is used to construct a segmentation frame to segment and re-recognize the text outlines within the abnormal image regions. This invention utilizes prior data to construct a user-specific user profile library that reflects the handwriting characteristics of the user, subsequently guiding the recognition process of the collected data to be processed, improving the accuracy of image data recognition and processing, and reducing misidentification.

Description

Technical Field

[0001] This invention relates to the field of image data processing, and more particularly to an artificial intelligence-based Internet data processing method, apparatus, and storage medium. Background Technology

[0002] With the integration of cloud computing and artificial intelligence technologies, a large amount of image data containing text is uploaded to the cloud for recognition and information extraction to support various Internet application scenarios. However, due to significant differences in individual writing habits, the character outlines often exhibit unique offsets and tilts in spatial arrangement. Traditional unified models struggle to capture such personalized font spatial distribution characteristics, resulting in poor text recognition efficiency in this type of image data and affecting the backend functions of Internet platforms.

[0003] For example, Chinese Patent Publication No. CN121072476A discloses a multimodal document parsing method, electronic device, and storage medium. The method includes: preprocessing the original document; using an object detection model to identify key elements (at least formulas, tables, text, and images) to obtain an element location annotation table; coarsely identifying the task type based on the annotation table to obtain an identifier; combining the original document metadata to identify the domain and generate domain adaptation parameters; determining the element range according to the annotation table; combining the task identifier to detect obstacles layer by layer and repair them accordingly to obtain an accessible element document; co-encoding the accessible elements; splitting sub-tasks based on the task identifier and encoding results for parallel processing; calling the adaptation parameters to adjust the accuracy; verifying and correcting the processing results; and integrating them into a final structured document report according to the adaptation parameter format. This invention can improve the efficiency and accuracy of multimodal document parsing.

[0004] However, the following problems still exist in the existing technology.

[0005] When processing handwritten text, some existing internet platforms typically perform contour segmentation before text recognition. However, due to differences in individual writing habits, handwritten text may exhibit character offsets, connected strokes, and extended strokes. These phenomena can easily introduce segmentation errors or cause character feature distortion during the contour segmentation stage, thereby reducing the accuracy of text recognition. Summary of the Invention

[0006] To address this, the present invention provides an internet data processing method, apparatus, and storage medium based on artificial intelligence, in order to overcome the influence of differences in users' personal writing habits in the prior art. Handwritten characters are prone to introducing segmentation errors or causing character feature distortion in the contour segmentation stage, thereby reducing the accuracy of text recognition.

[0007] To achieve the above objectives, in one aspect, the present invention provides an internet data processing method based on artificial intelligence, comprising:

[0008] The text to be identified is read, and several confidence image region samples are selected based on the spelling detection results.

[0009] The standard text image is gradient scaled, and the construction size is determined based on the degree of overlap with the font outline in the confidence image region sample during the process, so as to construct the structural observation box;

[0010] Based on the structural observation frame, determine the distribution pattern characteristics of the samples corresponding to the confidence image region;

[0011] Cluster confidence image region samples corresponding to the same text, verify the spatial regularity of the text based on the consistent probability of the corresponding distribution pattern features, construct the association between the text and the distribution pattern features, and build a user profile database.

[0012] In response to new information uploaded by the user, identify abnormal image areas, extract text outlines and match them with standard text images to select potentially related text;

[0013] Based on the user profile database corresponding to potentially related text, a segmentation frame is constructed to segment and then recognize the text outlines in the abnormal image area. After the text spelling detection is qualified, the abnormal text is replaced.

[0014] Furthermore, the process of selecting several confidence image region samples based on the spelling detection results includes,

[0015] For text segments that show no spelling errors, break them down into individual letters.

[0016] Several image regions mapped by single letters are selected, and each of these image regions is determined as a confidence image region sample.

[0017] Furthermore, the process of determining the construction size based on the degree of overlap between the font outline and the font contour in the confidence image region sample during gradient scaling includes,

[0018] The standard text image is scaled according to the gradient scaling ratio;

[0019] Determine the degree of overlap between the font outlines in the standard text image and the font outlines in the confidence image region samples under different scaling ratios;

[0020] Determine the standard text image corresponding to the maximum overlap, and extract the width and height of the font outline in the standard text image;

[0021] The extracted width and height are respectively increased by a preset error redundancy amount to form the construction size;

[0022] The standard text images are pre-set and correspond one-to-one with each letter.

[0023] Furthermore, the process of constructing a structural observation frame arranged according to a predetermined pattern includes,

[0024] A rectangular window is constructed based on the aforementioned construction size, with the center of the rectangular window located at the center of the confidence image region sample.

[0025] Horizontal and vertical dividing lines are constructed through the center to divide the rectangular window, and the resulting rectangular sub-windows are used as structural observation frames.

[0026] The structural observation frames should be assigned different serial numbers at different distribution locations.

[0027] Furthermore, the process of determining the distribution pattern characteristics corresponding to the confidence image region samples based on the structural observation frame includes,

[0028] Determine the outline of the font portion outside the structural observation frame in the confidence image region sample;

[0029] Determine the offset direction of each of the aforementioned partial font outlines relative to the corresponding structural observation box, and determine the offset distance of the partial font outlines relative to the structural observation box in the offset direction;

[0030] Record the observation frame of each structure, as well as the corresponding offset direction and offset distance.

[0031] Furthermore, the process of clustering confidence image region samples corresponding to the same text and verifying the spatial regularity of the text based on the consistent probability of corresponding distribution pattern features includes:

[0032] The distribution patterns of each confidence image region sample are compared to verify the proportion of confidence image region samples that meet the consistency conditions, and the proportion is used as the consistency probability.

[0033] If the consistency probability is greater than a predetermined consistency probability threshold, then the text is verified to have spatial regularity.

[0034] Wherein, the text in the clustering process contains only a single letter, and the consistency conditions include that the offset directions corresponding to each structural observation box are the same, and the difference ratio of the offset distance is less than a predetermined difference ratio threshold.

[0035] Furthermore, the process of identifying anomalous text to recognize corresponding anomalous image regions includes,

[0036] Identify text with spelling errors as abnormal text;

[0037] The abnormal text is broken down into individual letters or characters, and the image region mapped to each individual letter or character is determined.

[0038] Each of the aforementioned image regions is enlarged by a predetermined ratio and then identified as an abnormal image region.

[0039] Furthermore, the process of extracting the corresponding text outlines and matching them with standard text images to select several potentially related texts includes,

[0040] Calculate the similarity between the text outline corresponding to the abnormal image region and each of the standard text images, and sort the sequence numbers of each of the standard text images in descending order based on the similarity.

[0041] A predetermined number of standard text images are extracted from the beginning of the sorted sequence, and the text corresponding to each of the standard text images is identified as potential related text.

[0042] Furthermore, the process of constructing a segmentation frame based on a user profile database corresponding to potentially related text to segment and then recognize the text contours within the abnormal image region includes,

[0043] Construct a structural observation box within the abnormal image region;

[0044] A user profile library is established to identify potential related texts for abnormal image regions, in order to extract the offset direction and offset distance corresponding to each structural observation box;

[0045] Each of the structural observation frames is extended in the corresponding offset direction, and the extension distance is determined based on the corresponding offset distance;

[0046] The extended observation boxes are combined to form a segmentation box, and only the text outlines within the segmentation box are extracted for re-recognition.

[0047] Furthermore, a storage medium is provided that stores a computer program, which, when executed by a processor, can be used to perform an artificial intelligence-based Internet data processing method.

[0048] Furthermore, an apparatus is also provided, comprising:

[0049] One or more processors;

[0050] Memory;

[0051] and one or more programs,

[0052] The one or more programs are configured to be executed by one or more processors, and the memory includes the storage medium.

[0053] Compared with existing technologies, this invention reads the information to be recognized uploaded by the user to the cloud, performs spelling detection, filters confidence image region samples, constructs a structural observation box, determines the distribution pattern features corresponding to the confidence image region samples based on the structural observation box, clusters confidence image region samples corresponding to the same text, verifies the spatial regularity of the text based on the consistency probability of the corresponding distribution pattern features, and then constructs a user profile library exclusive to the user based on the correlation between the text and the distribution pattern features. Subsequently, when the user uploads new information to be recognized, abnormal image regions and potentially related texts are identified. The corresponding user profile library is used to construct a segmentation box to segment and re-recognize the text outlines in the abnormal image regions. After the obtained text passes the spelling detection, the abnormal text is replaced. This invention uses prior data to construct a user profile library exclusive to the user, reflecting the handwriting performance patterns of the user, and subsequently guides the recognition process of the collected data to be processed, improving the recognition and processing accuracy of image data and reducing misrecognition.

[0054] In particular, this invention constructs a structural observation box for confidence image region samples. These confidence image region samples are image regions selected from text with no spelling anomalies, reflecting the handwritten appearance of individual letters. Subsequently, this invention constructs the structural observation box by performing gradient scaling on a standard text image. The purpose of this scaling is to account for slight differences in handwritten font size. Therefore, during gradient scaling, by determining the overlap of different gradients, the standard text image size that best matches the font outline in the confidence image region sample is typically found when the overlap is at its maximum. This standard text image size is then used as an anchor object to determine the construction size and construct the structural observation box. The structural observation boxes are arranged according to a predetermined pattern, forming a 2×2 grid. In reality, due to differences in personal habits, handwriting styles vary. For example, some letters may have extended strokes or font offsets. Therefore, the structural observation boxes can determine the distribution patterns of confidence image region samples, reflecting the offset of handwriting in different positions and the offset distance. This allows for accurate capture of individual handwriting patterns for different letters, providing data support for building a user profile database. It also guides the recognition process of the collected data to be processed, improving the accuracy of image data recognition and reducing misidentification.

[0055] In particular, this invention clusters confidence image region samples corresponding to the same text to verify the spatial regularity of the text. When clustering, the text corresponds to a single letter. By determining the consistency probability of the distribution regularity features of the text, it reflects whether the offset direction and offset distance of the handwritten characters of a single letter have regularity in various directions. The association between the verified text and the corresponding distribution regularity features is stored in a user profile database dedicated to the user to reflect the performance regularity of the handwritten characters on the user's end. This then guides the recognition process of the collected data to be processed, improves the recognition and processing accuracy of image data, and reduces misidentification.

[0056] In particular, when the user uploads information to be recognized, this invention performs text reading, identifies abnormal text, and then identifies abnormal image regions. In reality, abnormal text is usually caused by misidentification. Due to differences in handwriting, the outlines of some letters may interfere with the outlines of adjacent letters, thus affecting the outline segmentation in the text recognition process, and leading to the problem of misidentifying the outlines of letters as incorrect letters or characters. Therefore, this invention identifies abnormal image regions and finds several potentially related texts that are similar to the outlines of the letters after matching them with standard text images. Subsequently, the distribution pattern features corresponding to the user profile database are used to re-segment the outlines of the letters in the abnormal image regions. Because the distribution pattern features are used, the re-segmented outlines can capture more complete outlines of the letters with a high probability or segment the real outlines of the letters using the characteristics of handwriting, thereby improving the recognition progress of fonts in image data. Attached Figure Description

[0057] Figure 1 This is a schematic diagram illustrating the steps of an artificial intelligence-based internet data processing method according to an embodiment of the invention.

[0058] Figure 2 A logic block diagram for filtering several confidence image region samples according to an embodiment of the invention;

[0059] Figure 3 This is a schematic diagram of the rectangular observation frame according to an embodiment of the invention;

[0060] Figure 4 This is a logic block diagram for verifying the spatial regularity of text in an embodiment of the invention. Detailed Implementation

[0061] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0062] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] Please see Figure 1 The diagram illustrates the steps of an artificial intelligence-based internet data processing method according to an embodiment of the invention. The artificial intelligence-based internet data processing method of this embodiment includes:

[0064] Step S1: Read the text of the information to be recognized uploaded to the cloud by the user terminal, and filter the image region mapped by the text based on the spelling detection results of the text to select a number of confidence image region samples.

[0065] Step S2: Gradient scaling is performed on the standard text image, and the construction size is determined based on the degree of overlap between the standard text image and the font outline in the confidence image region sample during the gradient scaling process, so as to construct a structural observation box arranged according to a predetermined rule. The standard text image is selected based on the text mapped by the confidence image region sample.

[0066] Step S3: Determine the distribution pattern characteristics of the confidence image region samples based on the structural observation boxes, including the offset direction and offset distance of the font outlines from each of the structural observation boxes;

[0067] Step S4: Cluster the confidence image region samples corresponding to the same text, verify the spatial regularity of the text based on the consistency probability of the corresponding distribution pattern features, construct the association relationship between each verified text and the distribution pattern features, and store it in the user's exclusive user profile library.

[0068] Step S5: In response to the new information to be identified uploaded by the user, the text is read, abnormal text is identified, the corresponding abnormal image area is identified, the corresponding text outline is extracted and matched with the standard text image, and several potentially related texts are selected.

[0069] Step S6: Construct a segmentation frame based on the user profile library corresponding to the potentially related text to segment and re-identify the text outlines in the abnormal image area. Replace the abnormal text after the obtained text spelling detection is qualified.

[0070] Specifically, there are no specific limitations on the method of obtaining the information to be identified uploaded by the user. For example, the information to be identified can be received through an online learning platform. The information to be identified is an image containing English handwriting, such as a photo of someone doing a problem, a photo of someone doing dictation, a photo of someone writing from memory, or a photo of a handwritten note.

[0071] Specifically, there are no restrictions on the specific method of text reading. Any existing technology that can extract text outlines from images and recognize corresponding text can be used. This is existing technology and will not be elaborated further.

[0072] There are no restrictions on the specific method of spell detection; any existing spell detection software can be used, as long as it can identify the spelling errors in the English and Chinese text to be identified and output the corresponding spell detection results. For example, a general English spell check library can be integrated or an online spell check interface can be called; these are existing technologies and will not be elaborated further.

[0073] Specifically, please refer to Figure 2The diagram shown is a logic block diagram of filtering several confidence image region samples according to an embodiment of the invention. The process of filtering several confidence image region samples based on spell detection results includes:

[0074] For text segments that show no spelling errors, break them down into individual letters.

[0075] Several image regions mapped by single letters are selected, and each of these image regions is determined as a confidence image region sample.

[0076] It is understandable that when reading text, the text content is determined based on the text outline. Therefore, there is a mapping relationship between the text and the corresponding text outline. Thus, the image region mapped to the text can be located based on the mapping relationship.

[0077] Understandably, the absence of spelling anomalies reflects that each letter in the word has been correctly detected, and therefore, it can be used as a sample for further observation of the performance patterns of handwritten characters.

[0078] This invention constructs a structural observation box for confidence image region samples. The confidence image region samples are image regions selected from text with no spelling anomalies, reflecting the handwritten appearance of individual letters. In the subsequent construction of the structural observation box, a standard text image is used for gradient scaling. The purpose is to account for slight differences in handwritten font size. Therefore, during gradient scaling, the overlap of different gradients is determined. Typically, when the overlap is maximum, the standard text image size that best matches the font outline in the confidence image region sample can be found. This standard text image is then used as an anchor object to determine the construction size and construct the structural observation box. The structural observation boxes are arranged according to a predetermined pattern, forming a 2×2 grid. In reality, due to differences in personal habits, handwriting styles vary. For example, some letters may have extended strokes or font offsets. Therefore, the structural observation boxes can determine the distribution pattern characteristics of confidence image region samples, reflecting the offset of handwriting in different positions and the offset distance. This allows for accurate capture of individual handwriting patterns for different letters, providing data support for building a user profile database. It also guides the recognition process of the collected data to be processed, improving the accuracy of image data recognition and reducing misidentification.

[0079] Specifically, the process of determining the construction size based on the degree of overlap between the font outline and the font contour in the confidence image region sample during gradient scaling includes:

[0080] The standard text image is scaled according to the gradient scaling ratio;

[0081] Determine the degree of overlap between the font outlines in the standard text image and the font outlines in the confidence image region samples under different scaling ratios;

[0082] Determine the standard text image corresponding to the maximum overlap, and extract the width and height of the font outline in the standard text image;

[0083] The extracted width and height are respectively increased by a preset error redundancy amount to form the construction size;

[0084] The standard text images are pre-set and correspond one-to-one with each letter.

[0085] Specifically, in order to make the standard text image fit the text outline in the actual confidence image region sample as closely as possible, at each scaling ratio, the standard text image can be translated, rotated, and transformed to search for the maximum overlap at the corresponding scaling ratio, which is then used as the overlap of the font outline at the corresponding scaling ratio.

[0086] The scaling ratio has an upper limit, which is the scaling ratio corresponding to the area of ​​the standard text image being equal to twice the area of ​​the confidence image region sample. This is to avoid the scaling ratio being too large, which would cause the local text outline of the standard text image to cover the entire text outline of the confidence image region sample.

[0087] The degree of overlap can be calculated based on the ratio of the overlapping area to the total area.

[0088] The purpose of setting the error redundancy is to take into account the inherent individual differences in handwriting. Therefore, by appropriately increasing the construction size, a certain degree of error redundancy is provided. In practice, the error redundancy is selected within 5%-10% of the original width or height, preferably 5%. This is to slightly expand the width and height, ensuring complete capture of the target character outline while avoiding excessive enlargement of the frame and the introduction of interference from adjacent characters or the background. This ensures the accuracy of the subsequent statistical analysis of the offset direction and offset distance of each structure observation frame.

[0089] Specifically, the standard text image is pre-set. Therefore, the standard text image is a reference image containing the outline of the corresponding letter. Since there are 26 letters in total, a corresponding standard text image is set for each letter.

[0090] Specifically, please refer to Figure 3 As shown, it is a structural schematic diagram of a rectangular observation frame according to an embodiment of the invention. The process of constructing the structural observation frame arranged according to a predetermined rule includes,

[0091] A rectangular window is constructed based on the aforementioned construction size, with the center of the rectangular window located at the center of the confidence image region sample.

[0092] Horizontal and vertical dividing lines are constructed through the center to divide the rectangular window, and the resulting rectangular sub-windows are used as structural observation frames.

[0093] The structural observation frames should be assigned different serial numbers at different distribution locations.

[0094] Understandably, the dimensions include the height and width after adding error redundancy. A rectangular window can be constructed based on this height and width. On this basis, the constructed rectangular sub-windows are distributed in a 2×2 grid, which can reflect the offset of most fonts in the main direction.

[0095] Specifically, the process of determining the distribution pattern characteristics of confidence image region samples based on the structural observation frame includes:

[0096] Determine the outline of the font portion outside the structural observation frame in the confidence image region sample;

[0097] Determine the offset direction of each of the aforementioned partial font outlines relative to the corresponding structural observation box, and determine the offset distance of the partial font outlines relative to the structural observation box in the offset direction;

[0098] Record the observation frame of each structure, as well as the corresponding offset direction and offset distance.

[0099] It is understandable that the offset distance is the farthest distance between the edge of the structure view box and the partial outline of the font in the offset direction.

[0100] Specifically, in the implementation, each structural observation box was assigned a number, namely 1, 2, 3, and 4. For structural observation box number 1, its offset direction is the top and left side; for structural observation box number 2, its offset direction is the top and right side; for structural observation box number 3, its offset direction is the left and bottom side; and for structural observation box number 4, its offset direction is the bottom and right side.

[0101] In practice, when recording data, a three-dimensional data set can be constructed, which includes the sequence number, offset direction, and offset distance, to facilitate recording.

[0102] Specifically, please refer to Figure 4 As shown, this is a logical block diagram for verifying the spatial regularity of text according to an embodiment of the invention. The process of clustering confidence image region samples corresponding to the same text and verifying the spatial regularity of text based on the consistency probability of corresponding distribution regularity features includes:

[0103] The distribution patterns of each confidence image region sample are compared to verify the proportion of confidence image region samples that meet the consistency conditions, and the proportion is used as the consistency probability.

[0104] If the consistency probability is greater than a predetermined consistency probability threshold, then the text is verified to have spatial regularity.

[0105] Wherein, the text in the clustering process contains only a single letter, and the consistency conditions include that the offset directions corresponding to each structural observation box are the same, and the difference ratio of the offset distance is less than a predetermined difference ratio threshold.

[0106] The purpose of clustering based on the same text is to observe the patterns in how users write individual letters. Setting a consistency probability threshold reflects that the distribution pattern of the text is regular rather than random. In practice, the consistency probability threshold can be selected between 50% and 70%, preferably 60%. When the probability is higher than 60%, it indicates that the writing offset of individual letters forms a stable and inductive spatial pattern rather than random fluctuations. At the same time, it avoids that if the consistency probability threshold is set too high, a large number of real patterns may be excluded due to small fluctuations in the user's normal writing, resulting in a sparse user profile database and an inability to effectively capture the user's writing habits.

[0107] The difference ratio is the ratio of the difference in offset distances to the mean of the offset distances, reflecting the level of the offset amount relative to the offset distance. The purpose of setting a difference ratio threshold is that this invention takes into account that handwritten characters inherently have a certain degree of slight error. Therefore, it allows for a certain degree of error between the distribution patterns of the samples corresponding to the confidence image regions. In practice, the difference ratio threshold can be selected between [0.25, 0.35], preferably 0.3, to avoid setting the difference ratio threshold too high. For example, if it is greater than 0.35, the tolerance for the fluctuation of the offset distance is too large, which may include random samples with unstable writing performance and no real pattern in the consistency judgment range.

[0108] This invention clusters confidence image region samples corresponding to the same text to verify the spatial regularity of the text. The text corresponds to a single letter during clustering. By determining the consistency probability of the distribution pattern features of the text, it reflects whether the offset direction and offset distance of the handwritten characters of a single letter have regularity in various directions. The association between the verified text and the corresponding distribution pattern features is stored in a user-specific user profile database to reflect the performance pattern of the user's handwritten characters. This then guides the recognition process of the collected data to be processed, improves the recognition and processing accuracy of image data, and reduces misidentification.

[0109] Specifically, the process of identifying anomalous text in order to recognize corresponding anomalous image regions includes,

[0110] Identify text with spelling errors as abnormal text;

[0111] The abnormal text is broken down into individual letters or characters, and the image region mapped to each individual letter or character is determined.

[0112] Each of the aforementioned image regions is enlarged by a predetermined ratio and then identified as an abnormal image region.

[0113] It is understandable that text recognition is based on generating corresponding text based on text outlines. Therefore, there is a mapping relationship between text and corresponding text outlines, which can be mapped to specific image regions. This will not be elaborated further.

[0114] It is understandable that the purpose of expanding the image area by a predetermined ratio is to appropriately widen the boundary of the image area to include the adjacent or interfering parts of the text outline, so that it can be re-divided later to more accurately capture the complete outline of the handwritten letter, thereby improving the recognition accuracy.

[0115] The predetermined ratio can be set to 20%. The interference area is usually the part adjacent to the text outline, so the predetermined ratio can be set to a small value. At the same time, setting it to 20% is also to avoid the predetermined ratio being too large, which would introduce too much interference after setting the image area too wide.

[0116] Specifically, the process of extracting the corresponding text outlines and matching them with standard text images, and selecting several potentially related texts, includes...

[0117] Calculate the similarity between the text outline corresponding to the abnormal image region and each of the standard text images, and sort the sequence numbers of each of the standard text images in descending order based on the similarity.

[0118] A predetermined number of standard text images are extracted from the beginning of the sorted sequence, and the text corresponding to each of the standard text images is identified as potential related text.

[0119] Specifically, there are no restrictions on the method of calculating similarity. The cosine similarity between the text outlines can be used as the similarity. Of course, other methods for determining image similarity can also be used, which will not be elaborated here.

[0120] In practice, the predetermined number is set to 3. Only the standard text images corresponding to the first 3 numbers of the sorted sequence are extracted. The subsequent standard text images have lost their comparative significance as the similarity decreases.

[0121] Specifically, the process of constructing a segmentation frame based on a user profile database corresponding to potentially related text, and then segmenting and re-identifying the text contours within the abnormal image region includes:

[0122] Construct a structural observation box within the abnormal image region;

[0123] A user profile library is established to identify potential related texts for abnormal image regions, in order to extract the offset direction and offset distance corresponding to each structural observation box;

[0124] Each of the structural observation frames is extended in the corresponding offset direction, and the extension distance is determined based on the corresponding offset distance;

[0125] The extended observation boxes are combined to form a segmentation box, and only the text outlines within the segmentation box are extracted for re-recognition.

[0126] In practice, the extension distance can be the same as the offset distance. Multiple structural observation boxes are combined to form a new dividing frame, which cuts out the text outline in the abnormal image area, making it easier to capture the complete handwritten outline.

[0127] In practice, since there are multiple potentially related texts corresponding to a single abnormal image region, a corresponding number of re-identified texts will be generated during the segmentation and re-identification process. To improve efficiency, spelling detection can be performed on the re-identified texts of potentially related texts with high similarity. Once the spelling detection is successful, the text is replaced, and spelling detection is no longer performed on the remaining texts. If the spelling detection of a text fails, the next text is spelled based on similarity until the spelling detection of all texts is completed. If all texts fail, the corresponding abnormal image region can be marked for manual identification, which will not be elaborated further.

[0128] Understandably, after text recognition is completed, it can support many functions. For example, for online learning platforms, it can be used to correct questions, recommend knowledge points, and directly archive texts, etc., which will not be elaborated here.

[0129] This invention, when users subsequently upload information to be recognized, performs text reading, identifies abnormal text, and then determines abnormal image regions. In reality, abnormal text is usually caused by misidentification. Due to differences in handwriting, the outlines of some letters may interfere with the outlines of adjacent letters, thus affecting the outline segmentation in the text recognition process, leading to the problem of misidentifying text outlines as incorrect letters or characters. Therefore, this invention identifies abnormal image regions and finds several potentially related texts with similar outlines by matching them with standard text images. Subsequently, it uses the distribution pattern features corresponding to the user profile database to re-segment the text outlines within the abnormal image regions. Because it utilizes the distribution pattern features, the re-segmented text outlines can capture more complete text outlines with a high probability or segment the real text outlines using the characteristics of handwriting, thereby improving the progress of font recognition in image data.

[0130] In practice, a storage medium is also provided, which stores a computer program that, when executed by a processor, can be used to perform an artificial intelligence-based Internet data processing method.

[0131] In practice, an apparatus is also provided, comprising:

[0132] One or more processors;

[0133] Memory;

[0134] and one or more programs,

[0135] The one or more programs are configured to be executed by one or more processors, and the memory includes the storage medium.

[0136] If the artificial intelligence-based Internet data processing method of the present invention is implemented as 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, 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 and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An internet data processing method based on artificial intelligence, characterized in that, include: The text to be identified is read, and several confidence image region samples are selected based on the spelling detection results. The standard text image is gradient scaled, and the construction size is determined based on the degree of overlap with the font outline in the confidence image region sample during the process, so as to construct the structural observation box; Based on the structural observation frame, determine the distribution pattern characteristics of the samples corresponding to the confidence image region; Cluster confidence image region samples corresponding to the same text, verify the spatial regularity of the text based on the consistent probability of the corresponding distribution pattern features, construct the association between the text and the distribution pattern features, and build a user profile database. In response to new information uploaded by the user, identify abnormal image areas, extract text outlines and match them with standard text images to select potentially related text; Based on the user profile database corresponding to potentially related text, a segmentation frame is constructed to segment and then recognize the text outlines in the abnormal image area. After the text spelling detection is qualified, the abnormal text is replaced.

2. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of selecting several confidence image region samples based on spell detection results includes, For text segments that show no spelling errors, break them down into individual letters. Several image regions mapped by single letters are selected, and each of these image regions is determined as a confidence image region sample.

3. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of determining the construction size based on the degree of overlap between the font outline and the font contour in the confidence image region sample during gradient scaling includes the following steps: The standard text image is scaled according to the gradient scaling ratio; Determine the degree of overlap between the font outlines in the standard text image and the font outlines in the confidence image region samples under different scaling ratios; Determine the standard text image corresponding to the maximum overlap, and extract the width and height of the font outline in the standard text image; The extracted width and height are respectively increased by a preset error redundancy amount to form the construction size; The standard text images are pre-set and correspond one-to-one with each letter.

4. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of constructing a structural observation frame arranged according to a predetermined pattern includes, A rectangular window is constructed based on the aforementioned construction size, with the center of the rectangular window located at the center of the confidence image region sample. Horizontal and vertical dividing lines are constructed through the center to divide the rectangular window, and the resulting rectangular sub-windows are used as structural observation frames. The structural observation frames should be assigned different serial numbers at different distribution locations.

5. The Internet data processing method based on artificial intelligence according to claim 4, characterized in that, The process of determining the distribution pattern characteristics of confidence image region samples based on the structural observation frame includes, Determine the outline of the font portion outside the structural observation frame in the confidence image region sample; Determine the offset direction of each of the aforementioned partial font outlines relative to the corresponding structural observation box, and determine the offset distance of the partial font outlines relative to the structural observation box in the offset direction; Record the observation frame of each structure, as well as the corresponding offset direction and offset distance.

6. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of clustering confidence image region samples corresponding to the same text and verifying the spatial regularity of the text based on the consistent probability of the corresponding distribution pattern features includes: The distribution patterns of each confidence image region sample are compared to verify the proportion of confidence image region samples that meet the consistency conditions, and the proportion is used as the consistency probability. If the consistency probability is greater than a predetermined consistency probability threshold, then the text is verified to have spatial regularity. Wherein, the text in the clustering process contains only a single letter, and the consistency conditions include that the offset directions corresponding to each structural observation box are the same, and the difference ratio of the offset distance is less than a predetermined difference ratio threshold.

7. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of identifying anomalous text, recognizing corresponding anomalous image regions, extracting corresponding text outlines and matching them with standard text images, and selecting several potentially related texts includes... Identify text with spelling errors as abnormal text; The abnormal text is broken down into individual letters or characters, and the image region mapped to each individual letter or character is determined. Each of the aforementioned image regions is enlarged by a predetermined ratio and then identified as an abnormal image region; Calculate the similarity between the text outline corresponding to the abnormal image region and each of the standard text images, and sort the sequence numbers of each of the standard text images in descending order based on the similarity. A predetermined number of standard text images are extracted from the beginning of the sorted sequence, and the text corresponding to each of the standard text images is identified as potential related text.

8. The Internet data processing method based on artificial intelligence according to claim 1, characterized in that, The process of constructing a segmentation frame based on a user profile database corresponding to potentially related text, and then segmenting and re-identifying the text contours within the abnormal image region includes: Construct a structural observation box within the abnormal image region; A user profile library is established to identify potential related texts for abnormal image regions, in order to extract the offset direction and offset distance corresponding to each structural observation box; Each of the structural observation frames is extended in the corresponding offset direction, and the extension distance is determined based on the corresponding offset distance; The extended observation boxes are combined to form a segmentation box, and only the text outlines within the segmentation box are extracted for re-recognition.

9. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it can be used to perform the artificial intelligence-based Internet data processing method according to any one of claims 1-8.

10. An apparatus, characterized in that, include: One or more processors; Memory; and one or more programs, The one or more programs are configured to be executed by one or more processors, and the memory includes the storage medium as described in claim 9.

Images