Video recognition text adaptive adjustment method and system

Abstract

The present application relates to a method and system for adaptively adjusting text in video recognition, which sequentially obtains the total data of current video recognition, obtains the first recognition font data with an intermediate accuracy rate and the second recognition font data with an advanced accuracy rate; obtains the first recognition feature data, obtains the second recognition feature data; obtains the recognition difference feature; performs adaptive incremental learning on the video text recognition network model according to the recognition difference feature, generates an enhanced text recognition network model after adaptive incremental learning, and performs text recognition on the current video to be recognized based on the enhanced text recognition network model, generates the current text recognition result after recognition, and displays the current text recognition result. The present invention makes the current text recognition result generated after recognition accurate and efficient, and at the same time, in order to achieve more utilization of user viewing, the current text recognition result is further displayed.

Description

Technical Field

[0001] The present application relates to the technical field of text recognition, and in particular to a method and system for adaptively adjusting text in video recognition. Background Art

[0002] Character recognition is a technology that uses computers to automatically recognize characters. It is an important field of pattern recognition applications. People have to deal with a lot of characters, reports and texts in production and life. In order to reduce people's labor and improve processing efficiency, the general character recognition method began to be explored in the 1950s, and optical character recognizers were developed. In the 1960s, practical machines using magnetic ink and special fonts appeared. In the late 1960s, a variety of fonts and handwritten character recognition machines appeared, and their recognition accuracy and machine performance basically met the requirements. For example, handwritten digital recognition machines and printed English digital recognition machines are used for letter sorting. In the 1970s, the basic theory of character recognition and the development of high-performance character recognition machines were mainly studied, and the research on Chinese character recognition was focused.

[0003] With the development and progress of society, more and more people need to perform text recognition on videos, and there are various methods for video text recognition. For example, the invention patent with patent number ZL201610245068.5 discloses a method and device for recognizing text in videos, and the method comprises: obtaining multiple frames of video images; extracting multiple texts to be recognized from the multiple frames of video images respectively; calculating the edit distance between the multiple texts to be recognized; obtaining one or more candidate items according to the edit distance; recognizing the current text from the one or more candidate items;

[0004] For another example, the invention patent with publication number CN109583443A discloses a method for judging video content based on text recognition, which includes the following steps: A. taking a screenshot of the video; B. calling a pre-trained text detection model to analyze the text area of ​​the screenshot, finding the text area in the picture and segmenting it, and obtaining one or more text areas; C. after detecting the text area, calling a pre-trained text recognition model, cyclically performing text recognition on each text area, and identifying the text content of each text area; D. performing natural language processing on the identified text content, understanding its semantics, and making corresponding video playback settings.

[0005] Although the technology disclosed in the above patent document can obtain accurate recognition results and avoid dependence on a corpus, or can recognize text information in the video and set up scenes according to the prompts of the text information to realize the recognition of video text in multiple fields, it still has disadvantages. Specifically, it cannot be adjusted adaptively, that is, it cannot realize adaptive self-learning based on text recognition content, which leads to the situation that the recognition mode is prone to lag behind as time goes by, which leads to the problem of backward recognition tools or methods, and the inability to accurately recognize text. Summary of the invention

[0006] Based on this, it is necessary to provide a video recognition text adaptive adjustment method and system that can improve data processing efficiency in response to the above technical problems.

[0007] The technical solution of the present invention is as follows:

[0008] A method for adaptively adjusting text in video recognition, the method comprising:

[0009] Obtain the current video recognition total data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and perform data splitting on the current video recognition total data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; obtain the first basic video corresponding to the first recognition font data, and obtain the first recognition feature data for identifying the first basic video, obtain the second basic video corresponding to the second recognition font data, and obtain the second recognition feature data for identifying the second basic video; compare the first recognition feature data with the second recognition feature data, and obtain recognition difference features; perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

[0010] Specifically, obtaining a first basic video corresponding to the first recognition font data and obtaining first recognition feature data for identifying the first basic video, obtaining a second basic video corresponding to the second recognition font data and obtaining second recognition feature data for identifying the second basic video; specifically, including:

[0011] Obtain time progress data corresponding to the first recognition font data; extract a first basic video matching the time progress data from the basic video to be recognized based on the time progress data; obtain initial recognition data when the first basic video is recognized based on the video text recognition network model; obtain current correction data when a text recognition reviewer corrects the initial recognition data, wherein the current correction data is used to characterize error data in the initial recognition data; generate first recognition feature data based on the current correction data; obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for recognizing the second basic video.

[0012] Further, the current video recognition total data generated after the text recognition of the basic video to be recognized within a preset time period is obtained based on the preset video text recognition network model, and the current video recognition total data is split and the first recognition font data with an accuracy rate of medium accuracy and the second recognition font data with an accuracy rate of high accuracy are obtained; specifically including:

[0013] Acquire initial recognition data generated after performing text recognition on a basic video to be recognized within a preset time period based on a preset video text recognition network model; perform screening according to the initial recognition data, and remove blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text; perform error screening on the recognition data without text, remove error field data, and correct the error field data to generate second-stage recognition data; obtain real-time browsing feedback of experimental browsing users on the second-stage recognition data, and correct the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data; compare the third-stage recognition data with the first-stage recognition data according to preset length segments, and obtain actual comparison results, wherein the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment and an actual correct value; and screen out first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate according to the actual comparison results.

[0014] Furthermore, the first recognition font data with a correct rate of medium accuracy and the second recognition font data with a correct rate of high accuracy are selected according to the actual comparison result, and then the following is further included:

[0015] According to the actual comparison result, an initial data segment with a correct rate lower than the intermediate accuracy rate is screened out, and set as a problem data segment; based on the actual correct value of the problem data segment, an actual difference value between the intermediate accuracy rate is generated, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate; according to the actual difference value, a text recognition supervisor matching the actual difference value is screened out; problem analysis feedback on the problem data segment from the text recognition supervisor is obtained; and the problem analysis feedback is sent to the text recognition reviewer.

[0016] Furthermore, the method further comprises:

[0017] Obtain the initial recognition feedback result of the current text recognition result from the initial auditor; obtain the same-level feedback results of the current text recognition result from other auditors in the same group as the initial auditor; obtain the supervisory auditor of the initial auditor and the other auditors, and obtain the supervisory review result of the supervisory auditor; generate the current recognition comprehensive feedback result according to the initial recognition feedback result, the same-level feedback result and the supervisory review result; determine whether the current text recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result; if the current text recognition result is determined to be greater than or equal to the standard reasonable value, continue to display the current text recognition result, and send the current text recognition result to a preset target user group; if the current text recognition result is determined to be less than the standard reasonable value, review the current text recognition result according to the current recognition comprehensive feedback result, and generate recognition text data with display after the review is completed.

[0018] Furthermore, a video recognition text adaptive adjustment system is provided, the system comprising:

[0019] A font data acquisition module is used to acquire the total current video recognition data generated after the basic video to be recognized is recognized based on the preset video text recognition network model within a preset time period, and to split the total current video recognition data and acquire the first recognition font data with an accuracy rate of an intermediate accuracy rate and the second recognition font data with an accuracy rate of an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate;

[0020] A feature data generating module, used to obtain a first basic video corresponding to the first recognition font data, and obtain first recognition feature data for identifying the first basic video, obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for identifying the second basic video;

[0021] A difference feature generating module, used for comparing the first identification feature data with the second identification feature data, and obtaining an identification difference feature;

[0022] The model incremental display module is used to perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after the adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

[0023] Furthermore, the feature data generating module is also used for:

[0024] Obtain time progress data corresponding to the first recognition font data; extract a first basic video matching the time progress data from the basic video to be recognized based on the time progress data; obtain initial recognition data when the first basic video is recognized based on the video text recognition network model; obtain current correction data when a text recognition reviewer corrects the initial recognition data, wherein the current correction data is used to characterize error data in the initial recognition data; generate first recognition feature data based on the current correction data; obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for recognizing the second basic video.

[0025] Furthermore, the font data acquisition module is also used for:

[0026] The module obtains initial recognition data generated after text recognition is performed on a basic video to be recognized within a preset time period based on a preset video text recognition network model; screens according to the initial recognition data, and removes blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text; performs error screening on the recognition data without text, removes error field data, and corrects the error field data to generate second-stage recognition data; obtains real-time browsing feedback of experimental browsing users on the second-stage recognition data, and corrects the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data; compares the third-stage recognition data with the first-stage recognition data according to preset length segments, and obtains actual comparison results, wherein the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment, and an actual correct value; and screens out first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate according to the actual comparison results;

[0027] The font data acquisition module is also used to: select an initial data segment whose accuracy is lower than the intermediate accuracy rate according to the actual comparison result, and set it as a problem data segment; generate an actual difference value between the actual correct value of the problem data segment and the intermediate accuracy rate based on the actual correct value, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate; select a text recognition supervisor who matches the actual difference value according to the actual difference value; obtain the problem analysis feedback of the text recognition supervisor on the problem data segment; and send the problem analysis feedback to the text recognition reviewer;

[0028] The model incremental display module is also used to: obtain the initial recognition feedback result of the initial auditor on the current text recognition result; obtain the same-level feedback results of the current text recognition result from other auditors in the same group as the initial auditor; obtain the supervisory auditors of the initial auditor and the other auditors, and obtain the supervisory review results of the supervisory auditor; generate the current recognition comprehensive feedback result according to the initial recognition feedback result, the same-level feedback result and the supervisory review result; judge whether the current text recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result; if the current text recognition result is judged to be greater than or equal to the standard reasonable value, continue to display the current text recognition result, and send the current text recognition result to a preset target user group; if the current text recognition result is judged to be less than the standard reasonable value, review the current text recognition result according to the current recognition comprehensive feedback result, and generate recognition text data with display after the review is completed.

[0029] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned video recognition text adaptive adjustment method when executing the computer program.

[0030] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for adaptively adjusting text in video recognition.

[0031] The technical effects achieved by the present invention are as follows:

[0032] The above-mentioned video recognition text adaptive adjustment method and system sequentially obtains the current video recognition total data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and performs data splitting on the current video recognition total data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; obtains the first basic video corresponding to the first recognition font data, and obtains the first recognition feature data for identifying the first basic video, obtains the second basic video corresponding to the second recognition font data, and obtains the second recognition feature data for identifying the second basic video; compares the first recognition feature data with the second recognition feature data, and obtains recognition difference features; performs adaptive incremental learning on the video text recognition network model according to the recognition difference features, generates an enhanced text recognition network model after adaptive incremental learning, and performs text recognition on the current video to be recognized based on the enhanced text recognition network model, and generates the current text recognition network model after recognition. The present invention sets the intermediate accuracy rate to be lower than the advanced accuracy rate. With such a high accuracy rate and a low accuracy rate design, when the video text recognition network model is incrementally learned, the defects of the data with low accuracy rate can be used for incremental learning. After learning the defects, the text recognition can be performed the next time, thereby further improving the accuracy and reliability of subsequent recognition. In order to accurately obtain the original video corresponding to the recognition data with two different accuracy rates, the original video is traced to trace the deviation in the text recognition process, and the deviation is converted into difference features, specifically the recognition difference features, so as to obtain the specific features that cause the accuracy deviation, thereby ensuring the incremental learning of the model based on accurate features. Based on the model after incremental learning, the text recognition of the current video to be recognized is realized based on the enhanced text recognition network model, so that the current text recognition result generated after recognition is accurate and efficient. At the same time, in order to achieve better utilization of user viewing, the current text recognition result is displayed. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of a flow chart of a method for adaptively adjusting text in video recognition according to an embodiment;

[0034] Figure 2 A structural block diagram of a video recognition text adaptive adjustment system in one embodiment;

[0035] Figure 3 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0037] In one embodiment, an application scenario of the video recognition text adaptive adjustment method is provided, and the application scenario includes an intelligent device terminal, the intelligent device terminal is used to obtain the current video recognition total data generated after the basic video to be recognized is recognized based on the preset video text recognition network model within a preset time period, and split the current video recognition total data and obtain the first recognition font data with an intermediate accuracy rate and the second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; obtain the first basic video corresponding to the first recognition font data, and obtain the first recognition feature data for recognizing the first basic video, obtain the second basic video corresponding to the second recognition font data, and obtain the second recognition feature data for recognizing the second basic video; compare the first recognition feature data with the second recognition feature data, and obtain recognition difference features; perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after the adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

[0038] Furthermore, the smart device terminal may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices.

[0039] In one embodiment, Figure 1 As shown, a method for adaptively adjusting text in video recognition is provided, the method comprising:

[0040] Step S100: obtaining the total current video recognition data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and performing data splitting on the total current video recognition data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate;

[0041] Furthermore, in this embodiment, the video text recognition network model is a neural network model selected by a person skilled in the art for performing text recognition on the basic video to be recognized, specifically including but not limited to the neural network model with application number CN202010082008.2, or others, that is, as long as basic text recognition can be achieved. This application is intended to protect the method and steps for processing the data generated after the video text recognition network model is recognized.

[0042] Furthermore, in order to achieve subsequent incremental learning, so that the video text recognition network model can be adaptively adjusted to achieve subsequent more accurate and more in line with the current text recognition environment, the intermediate accuracy is set lower than the advanced accuracy. Such a high accuracy and a low accuracy design enables incremental learning to be performed by utilizing the defects of low-accuracy data when incremental learning is performed on the video text recognition network model. After learning the defects, the next time text recognition is performed, recognition can be performed, thereby further improving the accuracy and reliability of subsequent recognition.

[0043] That is, the total data of the current video recognition is split and first recognition font data with a correct rate of a medium accuracy rate and second recognition font data with a correct rate of a high accuracy rate are obtained.

[0044] Step S200: obtaining a first basic video corresponding to the first recognition font data, and obtaining first recognition feature data for identifying the first basic video, obtaining a second basic video corresponding to the second recognition font data, and obtaining second recognition feature data for identifying the second basic video;

[0045] Step S300: performing data comparison on the first identification feature data and the second identification feature data, and obtaining identification difference features;

[0046] Step S400: Adaptively incrementally learn the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after the adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate a current text recognition result after recognition, and display the current text recognition result.

[0047] Furthermore, in order to accurately obtain the original videos corresponding to two recognition data with different accuracy rates, the present invention then traces the source based on the original videos to trace the deviations that occur in the text recognition process, and converts the deviations into difference features, specifically the recognition difference features, so as to obtain the specific features that cause the accuracy deviations, and then ensure that incremental learning of the model is achieved based on accurate features, that is, by first obtaining the first basic video corresponding to the first recognition font data, and obtaining the first recognition feature data for identifying the first basic video, obtaining the second basic video corresponding to the second recognition font data, and obtaining the second recognition feature data for identifying the second basic video, then comparing the first recognition feature data with the second recognition feature data, and obtaining recognition difference features, and then performing adaptive incremental learning on the video text recognition network model according to the recognition difference features, and generating an enhanced text recognition network model after adaptive incremental learning.

[0048] Furthermore, based on the model after incremental learning, text recognition is implemented on the current video to be recognized based on the enhanced text recognition network model, so that the current text recognition result generated after recognition is accurate and efficient. At the same time, in order to achieve better user viewing, the current text recognition result is displayed.

[0049] Furthermore, incremental learning means that a learning system can continuously learn new knowledge from new samples and can preserve most of the previously learned knowledge, that is, the model can perform well on both old and new tasks. Through the incremental learning, the model can continuously learn new knowledge from new tasks and new data, and it is trainable when new tasks appear at different times. In this embodiment, the learning system is the video text recognition network model, the new tasks and new data are the recognition difference features, and the model after incremental learning is the enhanced text recognition network model. The enhanced text recognition network model can continuously learn new knowledge from new tasks and new data, and when new tasks appear at different times, after training, it can achieve subsequent more accurate text adaptive recognition and adjustment.

[0050] In one embodiment, step S200: obtaining a first basic video corresponding to the first recognition font data, and obtaining first recognition feature data for identifying the first basic video, obtaining a second basic video corresponding to the second recognition font data, and obtaining second recognition feature data for identifying the second basic video; specifically includes:

[0051] Step S210: obtaining time progress data corresponding to the first recognized font data;

[0052] Step S220: extracting a first basic video matching the time progress data from the basic videos to be identified based on the time progress data;

[0053] Step S230: Acquire initial recognition data when the first basic video is recognized based on the video text recognition network model;

[0054] Step S240: obtaining current correction data when the text recognition auditor corrects the initial recognition data, wherein the current correction data is used to characterize the erroneous data in the initial recognition data;

[0055] Step S250: generating first identification feature data according to the current correction data;

[0056] Step S260: Obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for identifying the second basic video.

[0057] Furthermore, in the present embodiment, in order to realize the accurate acquisition of the corresponding characteristic data, the time progress data corresponding to the first recognition font data is acquired; then, based on the time progress data, a first basic video matching the time progress data is extracted from the basic video to be recognized; then, initial recognition data when the first basic video is recognized based on the video text recognition network model is acquired; then, current correction data when the text recognition reviewer corrects the initial recognition data is acquired, wherein the current correction data is used to characterize the error data in the initial recognition data; then, the first recognition characteristic data is generated according to the current correction data; finally, the second basic video corresponding to the second recognition font data is acquired. video, and obtain second recognition feature data for identifying the second basic video, wherein the time progress data is the time period occupied by the position of the first recognition font data in the basic video to be recognized, so the first basic video matching the time progress data can be extracted from the basic video to be recognized through the time progress data, and the initial recognition data is the initial data generated based on the model. In order to perform text calibration more accurately, the initial recognition data is corrected by a text recognition reviewer, and current correction data is generated at the same time. The current correction data is used to characterize the error data in the initial recognition data, and then the second recognition feature data is obtained by the same recognition method as that for obtaining the first recognition feature data.

[0058] In one embodiment, step S100: obtaining the total current video recognition data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and performing data splitting on the total current video recognition data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate; specifically comprising:

[0059] Step S110: obtaining initial recognition data generated by performing text recognition on a basic video to be recognized within a preset time period based on a preset video text recognition network model;

[0060] Step S120: screening according to the initial recognition data and removing blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text;

[0061] In this embodiment, in order to filter out other useless data and thus obtain accurate data, initial recognition data is generated after text recognition is performed on the basic video to be recognized within a preset time period based on a preset video text recognition network model, and then screening is performed based on the initial recognition data, and blank text data is eliminated from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text.

[0062] Step S130: Screening the recognition data without text for errors, removing erroneous field data, and correcting the erroneous field data to generate second-stage recognition data;

[0063] Step S140: obtaining real-time browsing feedback of the experimental browsing user on the second-stage recognition data, and correcting the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data;

[0064] Step S150: comparing the third-stage recognition data with the first-stage recognition data according to a preset length segment, and obtaining an actual comparison result, wherein the third-stage recognition data includes a plurality of later data segments after being split according to a preset degree, and the first-stage recognition data includes a plurality of initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment, and an actual correct value;

[0065] Step S160: Filter out the first recognized font data with a correct rate of a medium level and the second recognized font data with a correct rate of a high level according to the actual comparison result.

[0066] Further, in the present embodiment, in order to accurately extract data and filter out the first recognition font data and the second recognition font data, correction is first performed, that is, by performing error screening on the recognition data without text, eliminating the error field data, and correcting the error field data to generate the second-stage recognition data, so that the accuracy of the generated second-stage recognition data is high. Then, in order to further achieve accurate data acquisition, manual correction is added, specifically including obtaining real-time browsing feedback of the experimental browsing user on the second-stage recognition data, and correcting the second-stage recognition data according to the real-time browsing feedback to generate the third-stage recognition data. In order to achieve accurate and detailed comparison during comparison, a preset length segment is set. Based on the preset length segment, segmented data comparison can be achieved. Specifically, it is pre-set that the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree. Then, segmented comparison can be achieved during comparison. In the actual comparison results obtained in this way, each actual comparison result includes a later data segment, an initial data segment and an actual correct value. Finally, the first recognition font data with a medium accuracy rate and the second recognition font data with a high accuracy rate are screened out according to the actual comparison result, so as to accurately and reliably obtain the required first recognition font data and second recognition font data.

[0067] In one embodiment, step S160: selecting first recognized font data with a medium accuracy rate and second recognized font data with a high accuracy rate according to the actual comparison result, and then further comprising:

[0068] Step S171: selecting the initial data segment whose accuracy is lower than the intermediate accuracy according to the actual comparison result, and setting it as the problem data segment;

[0069] Step S172: generating an actual difference value between the actual correct value of the problem data segment and the intermediate accuracy rate, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate;

[0070] Step S173: selecting a text recognition supervisor matching the actual difference value according to the actual difference value;

[0071] Step S174: obtaining the problem analysis feedback of the text recognition supervisor on the problem data segment;

[0072] Step S175: Send the problem analysis feedback to the text recognition reviewer.

[0073] Furthermore, in the present embodiment, in order to further evaluate and monitor the text data, when the accuracy is too low, supervisory personnel are required to supervise, specifically, according to the actual comparison result, an initial data segment with an accuracy lower than the intermediate accuracy is screened out, and set as a problem data segment; then, based on the actual correct value of the problem data segment, an actual difference value between the intermediate accuracy is generated, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy, and different actual difference values ​​and different levels of auditors corresponding to the actual difference values ​​are pre-set at this time, specifically, according to the actual difference value, a text recognition supervisor matching the actual difference value is screened out; then, the problem analysis feedback of the text recognition supervisor on the problem data segment is obtained; finally, the problem analysis feedback is sent to the text recognition auditor, so that by sending it to the text recognition auditor, the text recognition auditor can perform subsequent text audit work according to the relevant information, thereby improving the efficiency and accuracy of subsequent text audits.

[0074] In one embodiment, the method further comprises:

[0075] Step S510: obtaining an initial recognition feedback result of the initial auditor on the current text recognition result;

[0076] Step S520: obtaining feedback results of the same level on the current text recognition result from other auditors in the same group as the initial auditor;

[0077] Step S530: obtaining the supervisory auditors of the initial auditor and the other auditors, and obtaining the supervisory audit results of the supervisory auditors;

[0078] Step S540: generating a current identification comprehensive feedback result according to the initial identification feedback result, the peer feedback result and the supervisor review result;

[0079] Step S550: judging whether the current character recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result;

[0080] Step S560: If it is determined that the current text recognition result is greater than or equal to the standard reasonable value, the current text recognition result is continuously displayed, and the current text recognition result is sent to a preset target user group;

[0081] Step S570: If it is determined that the current text recognition result is less than the standard reasonable value, the current text recognition result is reviewed according to the current recognition comprehensive feedback result, and after the review is completed, the recognized text data with display is generated.

[0082] Furthermore, in this embodiment, in order to ensure the reliability and accuracy of the audit process without deviation, the initial recognition feedback result of the current text recognition result by the initial auditor is first obtained; then, the same-level feedback results of the current text recognition result by other auditors in the same group as the initial auditor are obtained, so that other auditors in the same group can make different evaluations based on the same level, thereby improving the compatibility of the audit, and then, the supervisory auditors of the initial auditor and the other auditors are obtained, and the supervisory audit results of the supervisory auditor are obtained; in this way, by setting up the supervisory auditor, higher quality audit information is obtained based on the supervisory auditor, and finally, according to the initial recognition feedback The current recognition comprehensive feedback result is generated based on the feedback result of the current recognition comprehensive feedback result, the feedback result of the same level and the review result of the supervisor; therefore, whether the current text recognition result is greater than or equal to a preset standard reasonable value is judged according to the current recognition comprehensive feedback result; the standard reasonable value is preset, so when it is judged that the current text recognition result is greater than or equal to the standard reasonable value, it means that the result at this time is reasonable and accurate, so the current text recognition result is continuously displayed, and the current text recognition result is sent to the preset target user group, then if it is judged that the current text recognition result is less than the standard reasonable value, the current text recognition result is reviewed according to the current recognition comprehensive feedback result, and the recognition text data with display is generated after the review is completed.

[0083] In summary, the video recognition text adaptive adjustment method and system of the present invention sequentially obtains the current video recognition total data generated after text recognition is performed on the basic video to be recognized within a preset time period based on a preset video text recognition network model, and performs data splitting on the current video recognition total data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; obtains a first basic video corresponding to the first recognition font data, and obtains first recognition feature data for identifying the first basic video, obtains a second basic video corresponding to the second recognition font data, and obtains second recognition feature data for identifying the second basic video; compares the first recognition feature data with the second recognition feature data, and obtains recognition difference features; performs adaptive incremental learning on the video text recognition network model according to the recognition difference features, generates an enhanced text recognition network model after the adaptive incremental learning, and performs text recognition on the current video to be recognized based on the enhanced text recognition network model, and generates a text recognition feature after recognition. The current text recognition result is obtained and the current text recognition result is displayed. That is, the present invention sets the intermediate accuracy rate to be lower than the advanced accuracy rate. With such a design of high accuracy rate and low accuracy rate, when the video text recognition network model is incrementally learned, the defects of the data with low accuracy rate can be used for incremental learning. After learning the defects, the text recognition can be performed the next time, thereby further improving the accuracy and reliability of subsequent recognition. In order to accurately obtain the original video corresponding to the recognition data with two different accuracy rates, the original video is traced to trace the deviation in the text recognition process, and the deviation is converted into difference features, specifically the recognition difference features, so as to obtain the specific features that cause the accuracy deviation, thereby ensuring that the incremental learning of the model is realized based on accurate features. Based on the model after incremental learning, the text recognition of the current video to be recognized is realized based on the enhanced text recognition network model, so that the current text recognition result generated after recognition is accurate and efficient. At the same time, in order to achieve better utilization of user viewing, the current text recognition result is displayed.

[0084] In one embodiment, the present invention further provides a video recognition text adaptive adjustment system, the system comprising:

[0085] A font data acquisition module is used to acquire the total current video recognition data generated after the basic video to be recognized is recognized based on the preset video text recognition network model within a preset time period, and to split the total current video recognition data and acquire the first recognition font data with an accuracy rate of an intermediate accuracy rate and the second recognition font data with an accuracy rate of an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate;

[0086] A feature data generating module, used to obtain a first basic video corresponding to the first recognition font data, and obtain first recognition feature data for identifying the first basic video, obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for identifying the second basic video;

[0087] A difference feature generating module, used for comparing the first identification feature data with the second identification feature data, and obtaining an identification difference feature;

[0088] The model incremental display module is used to perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after the adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

[0089] In one embodiment, the feature data generating module is further used to:

[0090] Obtain time progress data corresponding to the first recognition font data; extract a first basic video matching the time progress data from the basic video to be recognized based on the time progress data; obtain initial recognition data when the first basic video is recognized based on the video text recognition network model; obtain current correction data when a text recognition reviewer corrects the initial recognition data, wherein the current correction data is used to characterize error data in the initial recognition data; generate first recognition feature data based on the current correction data; obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for recognizing the second basic video.

[0091] In one embodiment, the font data acquisition module is further used for:

[0092] The module obtains initial recognition data generated after text recognition is performed on a basic video to be recognized within a preset time period based on a preset video text recognition network model; screens according to the initial recognition data, and removes blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text; performs error screening on the recognition data without text, removes error field data, and corrects the error field data to generate second-stage recognition data; obtains real-time browsing feedback of experimental browsing users on the second-stage recognition data, and corrects the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data; compares the third-stage recognition data with the first-stage recognition data according to preset length segments, and obtains actual comparison results, wherein the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment, and an actual correct value; and screens out first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate according to the actual comparison results;

[0093] The font data acquisition module is also used to: select an initial data segment whose accuracy is lower than the intermediate accuracy rate according to the actual comparison result, and set it as a problem data segment; generate an actual difference value between the actual correct value of the problem data segment and the intermediate accuracy rate based on the actual correct value, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate; select a text recognition supervisor who matches the actual difference value according to the actual difference value; obtain the problem analysis feedback of the text recognition supervisor on the problem data segment; and send the problem analysis feedback to the text recognition reviewer;

[0094] The model incremental display module is also used to: obtain the initial recognition feedback result of the initial auditor on the current text recognition result; obtain the same-level feedback results of the current text recognition result from other auditors in the same group as the initial auditor; obtain the supervisory auditors of the initial auditor and the other auditors, and obtain the supervisory review results of the supervisory auditor; generate the current recognition comprehensive feedback result according to the initial recognition feedback result, the same-level feedback result and the supervisory review result; judge whether the current text recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result; if the current text recognition result is judged to be greater than or equal to the standard reasonable value, continue to display the current text recognition result, and send the current text recognition result to a preset target user group; if the current text recognition result is judged to be less than the standard reasonable value, review the current text recognition result according to the current recognition comprehensive feedback result, and generate recognition text data with display after the review is completed.

[0095] In one embodiment, Figure 3 As shown, a computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps described in the above-mentioned video recognition text adaptive adjustment method when executing the computer program.

[0096] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for adaptively adjusting text in video recognition.

[0097] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0098] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A video recognition text adaptive adjustment method, characterized in that: The method comprises: Obtain the current video recognition total data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and perform data splitting on the current video recognition total data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; obtain the first basic video corresponding to the first recognition font data, and obtain the first recognition feature data for identifying the first basic video, obtain the second basic video corresponding to the second recognition font data, and obtain the second recognition feature data for identifying the second basic video; compare the first recognition feature data with the second recognition feature data, and obtain recognition difference features; perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

2. The video recognition text adaptive adjustment method according to claim 1, characterized in that: Acquiring a first basic video corresponding to the first recognition font data, and acquiring first recognition feature data for identifying the first basic video, acquiring a second basic video corresponding to the second recognition font data, and acquiring second recognition feature data for identifying the second basic video; specifically comprising: Obtain time progress data corresponding to the first recognition font data; extract a first basic video matching the time progress data from the basic video to be recognized based on the time progress data; obtain initial recognition data when the first basic video is recognized based on the video text recognition network model; obtain current correction data when a text recognition reviewer corrects the initial recognition data, wherein the current correction data is used to characterize error data in the initial recognition data; generate first recognition feature data based on the current correction data; obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for recognizing the second basic video.

3. The video recognition text adaptive adjustment method according to claim 2, characterized in that: Obtaining the current video recognition total data generated after performing text recognition on the basic video to be recognized within a preset time period based on the preset video text recognition network model, and performing data splitting on the current video recognition total data to obtain first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate; specifically including: Acquire initial recognition data generated after performing text recognition on a basic video to be recognized within a preset time period based on a preset video text recognition network model; perform screening according to the initial recognition data, and remove blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text; perform error screening on the recognition data without text, remove error field data, and correct the error field data to generate second-stage recognition data; obtain real-time browsing feedback of experimental browsing users on the second-stage recognition data, and correct the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data; compare the third-stage recognition data with the first-stage recognition data according to preset length segments, and obtain actual comparison results, wherein the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment and an actual correct value; and screen out first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate according to the actual comparison results.

4. The video recognition text adaptive adjustment method according to claim 3, characterized in that: According to the actual comparison result, the first recognition font data with a correct rate of medium accuracy and the second recognition font data with a correct rate of high accuracy are screened out, and then the method further includes: According to the actual comparison result, an initial data segment with a correct rate lower than the intermediate accuracy rate is screened out, and set as a problem data segment; based on the actual correct value of the problem data segment, an actual difference value between the intermediate accuracy rate is generated, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate; according to the actual difference value, a text recognition supervisor matching the actual difference value is screened out; problem analysis feedback on the problem data segment from the text recognition supervisor is obtained; and the problem analysis feedback is sent to the text recognition reviewer.

5. The video recognition text adaptive adjustment method according to claim 4, characterized in that: The method further comprises: Obtain the initial recognition feedback result of the current text recognition result from the initial auditor; obtain the same-level feedback results of the current text recognition result from other auditors in the same group as the initial auditor; obtain the supervisory auditor of the initial auditor and the other auditors, and obtain the supervisory review result of the supervisory auditor; generate the current recognition comprehensive feedback result according to the initial recognition feedback result, the same-level feedback result and the supervisory review result; determine whether the current text recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result; if the current text recognition result is determined to be greater than or equal to the standard reasonable value, continue to display the current text recognition result, and send the current text recognition result to a preset target user group; if the current text recognition result is determined to be less than the standard reasonable value, review the current text recognition result according to the current recognition comprehensive feedback result, and generate recognition text data with display after the review is completed.

6. A video recognition text adaptive adjustment system, characterized in that: The system comprises: A font data acquisition module is used to acquire the total current video recognition data generated after the basic video to be recognized is recognized based on the preset video text recognition network model within a preset time period, and to split the total current video recognition data and acquire the first recognition font data with an accuracy rate of an intermediate accuracy rate and the second recognition font data with an accuracy rate of an advanced accuracy rate, wherein the intermediate accuracy rate is lower than the advanced accuracy rate; A feature data generating module, used to obtain a first basic video corresponding to the first recognition font data, and obtain first recognition feature data for identifying the first basic video, obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for identifying the second basic video; A difference feature generating module, used for comparing the first identification feature data with the second identification feature data, and obtaining an identification difference feature; The model incremental display module is used to perform adaptive incremental learning on the video text recognition network model according to the recognition difference features, generate an enhanced text recognition network model after the adaptive incremental learning, and perform text recognition on the current video to be recognized based on the enhanced text recognition network model, generate the current text recognition result after recognition, and display the current text recognition result.

7. The video recognition text adaptive adjustment system according to claim 6, characterized in that: The feature data generating module is also used for: Acquire time progress data corresponding to the first recognition font data; extract a first basic video matching the time progress data from the basic video to be recognized based on the time progress data; acquire initial recognition data when the first basic video is recognized based on the video text recognition network model; Obtain current correction data when a text recognition reviewer corrects the initial recognition data, wherein the current correction data is used to characterize erroneous data in the initial recognition data; generate first recognition feature data based on the current correction data; obtain a second basic video corresponding to the second recognition font data, and obtain second recognition feature data for identifying the second basic video.

8. The video recognition text adaptive adjustment system according to claim 7, characterized in that: The font data acquisition module is also used for: Acquire initial recognition data generated after performing text recognition on a basic video to be recognized within a preset time period based on a preset video text recognition network model; perform screening according to the initial recognition data, and remove blank text data from the initial recognition data to generate first-stage recognition data, wherein the blank text data is recognition data without text; perform error screening on the recognition data without text, remove error field data, and correct the error field data to generate second-stage recognition data; obtain real-time browsing feedback of experimental browsing users on the second-stage recognition data, and correct the second-stage recognition data according to the real-time browsing feedback to generate third-stage recognition data; compare the third-stage recognition data with the first-stage recognition data according to preset length segments, and obtain actual comparison results, wherein the third-stage recognition data includes multiple later data segments after being split according to a preset degree, and the first-stage recognition data includes multiple initial data segments after being split according to a preset degree, and each actual comparison result includes a later data segment, an initial data segment, and an actual correct value; screen out first recognition font data with an intermediate accuracy rate and second recognition font data with an advanced accuracy rate according to the actual comparison results; The font data acquisition module is also used to: select an initial data segment whose accuracy is lower than the intermediate accuracy rate according to the actual comparison result, and set it as a problem data segment; generate an actual difference value between the actual correct value of the problem data segment and the intermediate accuracy rate based on the actual correct value, wherein the actual difference value is the difference between the actual correct value and the intermediate accuracy rate; select a text recognition supervisor who matches the actual difference value according to the actual difference value; obtain the problem analysis feedback of the text recognition supervisor on the problem data segment; and send the problem analysis feedback to the text recognition reviewer; The model incremental display module is also used to: obtain the initial recognition feedback result of the initial auditor on the current text recognition result; obtain the same-level feedback results of the current text recognition result from other auditors in the same group as the initial auditor; obtain the supervisory auditors of the initial auditor and the other auditors, and obtain the supervisory review results of the supervisory auditor; generate the current recognition comprehensive feedback result according to the initial recognition feedback result, the same-level feedback result and the supervisory review result; judge whether the current text recognition result is greater than or equal to a preset standard reasonable value according to the current recognition comprehensive feedback result; if the current text recognition result is judged to be greater than or equal to the standard reasonable value, continue to display the current text recognition result, and send the current text recognition result to a preset target user group; if the current text recognition result is judged to be less than the standard reasonable value, review the current text recognition result according to the current recognition comprehensive feedback result, and generate recognition text data with display after the review is completed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images