Video character recognition and erasing method and system, storage medium and electronic device

Abstract

The invention discloses a video character recognition and erasing method and system, a storage medium and an electronic device. The method comprises the following steps: acquiring a video; performing frame extraction processing to obtain a video frame picture; the method comprises the following steps: acquiring all text contents and coordinate data in a picture through OCR (Optical Character Recognition) identification, analyzing and identifying flower characters and subtitles on a video frame picture through a multi-modal large model, screening out a region of the flower characters and the subtitles needing to be erased through coordinate matching, and determining the region as an erased region; and intelligently erasing and repairing the flower and subtitle areas by adopting a video repairing technology, recovering the original state of the video, and generating an erased new video file. According to the method and the device, the full process of automation is realized, subtitles and flower characters in the video do not need to be manually marked, identified and erased, printed characters of commodities / packages are reserved, mistaken erasure is avoided, and the processing efficiency of video reediting is greatly improved.

Description

Technical Field

[0001] This application relates to the field of video content processing technology, and more specifically, to a video text recognition and erasure method, system, storage medium, and electronic device. Background Technology

[0002] Short videos have become a core medium for information dissemination in the digital age, and short video creators often need to modify and republish their already released videos.

[0003] During the redistribution process, the accurate removal of on-screen text and subtitles (such as on-screen text or subtitles) from the original video becomes a key requirement.

[0004] In traditional techniques, erasing text or subtitles from videos usually relies on manually marking the areas to be erased and repairing the image, which is inefficient and time-consuming and cannot meet the needs of large-scale video editing.

[0005] Existing automated technologies (such as single OCR recognition) are unable to cope with multimodal interference in complex scenarios. For example, they cannot accurately distinguish between on-screen text / subtitles and the inherent text of the product / packaging, which can easily lead to accidental erasure and affect the integrity of the video content.

[0006] Therefore, developing intelligent erasure technology that combines high-precision recognition with natural repair capabilities is of great significance. Summary of the Invention

[0007] The present invention aims to solve the problems of low efficiency and easy accidental erasure of product text in the existing technology of video subtitle and overlay text erasure, and provides a video text recognition and erasure method, system, storage medium and electronic device.

[0008] To achieve the above objectives, according to one aspect of this application, a video text recognition and erasure method is provided.

[0009] A video text recognition and erasure method according to this application includes: Get the video; Perform frame extraction processing on the video to obtain video frame images; The OCR is used to identify all the text in the video frame image, obtain all the text content and its coordinate data, output the OCR recognition result, and determine the inherent text of the product that needs to be retained as the retention area; The video frame images are analyzed using a pre-trained multimodal large model. The model identifies on-screen text and subtitles, and outputs the multimodal large model recognition results. The areas of decorative text and subtitles that need to be erased are selected by coordinate matching and determined as the erasure areas; Based on the erased area, video restoration technology is used to intelligently erase and repair the on-screen text and subtitle areas, restoring the original video and generating a new video file after erasure.

[0010] Furthermore, the frame extraction process is adaptively performed according to the frame rate of the video. While extracting frames, both computational efficiency and information integrity are taken into account. The frame extraction interval is dynamically adjusted according to the video frame rate. For example, for high frame rate videos, frame extraction is performed every four frames. For low frame rate videos, encrypted frame extraction or even continuous frame extraction is used.

[0011] Furthermore, before recognizing all text in the video frame image via OCR, the process includes: performing noise reduction, contrast enhancement, and edge sharpening preprocessing sequentially on the acquired video frame image.

[0012] Furthermore, the OCR is a multi-engine fusion OCR, through which OCR recognition results are obtained; the multi-engine fusion OCR processes the input video frame image using multiple engines, and adopts a voting mechanism to fuse the recognition results or compare similarity to obtain all text content and the coordinate data of the text, specifically obtaining the X / Y axis position, width and height information of the text; The OCR also includes a dynamic rule base system for automatically adjusting filtering strategies. The dynamic rule base system includes: rule definition, text extraction, keyword matching, and result output.

[0013] Furthermore, the analysis of the video frame images using a pre-trained multimodal large model includes: performing visual-text joint reasoning analysis on the video frame images using a pre-trained multimodal large model Vision-Language model, recognizing onomatopoeia and subtitles through the model, and outputting the multimodal large model recognition result, wherein the multimodal large model recognition result includes the bounding box coordinates of the confidence level.

[0014] Furthermore, the process of filtering out the areas of decorative text and subtitles to be erased through coordinate matching includes: The coordinate data of the text in the OCR recognition result obtained by the multi-engine fusion OCR will be matched and verified with the bounding box coordinates in the multimodal large model recognition result. If the IoU threshold between the coordinate data and the bounding box coordinates is greater than or equal to 0.9, the coordinate areas of the decorative text and subtitles in the coordinate data will be directly identified as the erasure areas; If the IoU threshold between the coordinate data and the bounding box coordinates is between 0.6 and 0.9, a secondary verification is performed to confirm the erased area, combining text semantics and contextual visual features.

[0015] Furthermore, the specific steps for performing secondary verification to confirm the erased area include: Text semantic analysis compares the text recognized by OCR with the text and subtitles recognized by the multimodal model, and uses natural language processing technology to determine whether they are semantically consistent or related. Contextual visual feature analysis, based on the visual features around the erased area, increases the confidence of the match if the visual features match the text area recognized by OCR and the decorative and subtitle areas recognized by the multimodal model. Based on a comprehensive assessment, combining the analysis results of text semantics and contextual visual features, a decision is made on whether to confirm the erasure area.

[0016] To achieve the above objectives, according to another aspect of this application, a video text recognition and erasure system is provided, the system comprising: The acquisition module is used to acquire videos; The video processing module is used to perform frame extraction processing on the video to obtain video frame images; The OCR recognition module is used to recognize all text in the video frame image and obtain all text content and its coordinate data. The multimodal analysis module includes a model input interface with customizable prompts, which is used by the multimodal large model to identify the on-screen text and subtitles in the video frame images through accurate prompts, and output the multimodal large model recognition results; The coordinate matching module is used to filter out the areas of decorative text and subtitles that need to be erased by coordinate matching, and determine them as the areas to be erased; The coordinate matching module also includes a decision module, which calculates the final score based on the weights of different features. During this process, the IoU and content similarity thresholds are dynamically adjusted according to the video frame quality (resolution, clarity) to improve robustness. The video repair module is used to intelligently erase and repair the text and subtitle areas in the erased area using video repair technology, restore the original video, and generate a new video file after erasure.

[0017] To achieve the above objectives, according to another aspect of this application, a computer-readable storage medium is provided, wherein a computer program is stored therein, and the computer program is configured to execute the method at runtime.

[0018] To achieve the above objectives, according to another aspect of this application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method.

[0019] The method in this application embodiment automates the entire process without requiring manual marking, identification, and erasure of subtitles and overlays in the video, while preserving the printed text on the product / packaging, avoiding accidental erasure, and significantly improving the processing efficiency of video re-editing. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 This is a flowchart illustrating a video text recognition and erasure method according to an embodiment of this application; Figure 2 This is a flowchart illustrating the filtering strategy according to an embodiment of this application; Figure 3 This is a flowchart illustrating the secondary verification of the erased area according to an embodiment of this application; Figure 4 This is a schematic diagram of a video text recognition and erasure system according to an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure used in a video text recognition and erasure method according to an embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0024] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0025] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linking," and "socketing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral structure; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] This application provides a video text recognition and erasure method, system, storage medium, and electronic device to solve the problems of low efficiency and time consumption in erasing text or subtitles in videos, which usually relies on manual marking of erasure areas and image repair, making it difficult to meet the needs of large-scale video editing. It also addresses the issue that existing automated technologies cannot accurately distinguish between text / subtitles and the inherent text of products / packaging, easily leading to accidental erasure and affecting the integrity of video content. The solution achieves fully automated process without manual marking, recognition, and erasure of subtitles and text in videos, while preserving the printed text of products / packaging, avoiding accidental erasure, and significantly improving the processing efficiency of video re-editing.

[0028] like Figure 1 The image shows a video text recognition and erasure method according to an embodiment of the present invention, comprising: Step 101: Obtain the video.

[0029] Step 102: Perform frame extraction processing on the video to obtain video frame images.

[0030] The frame extraction process is adaptively performed based on the frame rate of the video. While extracting frames, both computational efficiency and information integrity are taken into account. The frame extraction interval is dynamically adjusted according to the video frame rate. For example, for high frame rate videos, frame extraction is performed every four frames. For low frame rate videos, the frame extraction is encrypted or even performed continuously to obtain video frame images. The acquired video frame images are preprocessed sequentially with denoising, contrast enhancement, and edge sharpening. Unlike the traditional order of denoising, edge sharpening, and contrast enhancement, contrast is enhanced immediately after denoising. This may amplify the noise (e.g., histogram equalization can highlight noise details), leading to excessive amplification of noise during subsequent edge sharpening, resulting in unclear edges. However, it improves the overall contrast, making details clearer and more visible, which may make edge detection more sensitive and improve the subsequent determination of the boundaries of the erased area.

[0031] First, median filtering is used to effectively suppress noise while preserving the edge structure of the identified font. Then, the Sobel / Prewitt gradient method is used to detect edges, enhance high-frequency components of the image, highlight edges and contours, and make the font boundaries clearer. Finally, histogram equalization is used to redistribute gray levels, improve global contrast, expand the gray level range, and make details clearer and easier to see, so as to help determine the boundaries of the erasing area.

[0032] Step 103: Recognize all text in the video frame image using OCR, obtain all text content and its coordinate data, output the OCR recognition result, and determine the inherent text of the product that needs to be retained as the retention area.

[0033] The OCR method identifies all text in the video frame image. The OCR is a multi-engine fusion OCR, and the OCR recognition result is obtained through the multi-engine fusion OCR. The input video frame image is processed by multiple engines, and a voting mechanism is used to fuse the recognition results or compare similarity. For example, setting the similarity threshold to 0.85 can significantly reduce the false recognition rate of a single engine, making it suitable for recognizing complex scenes in videos or mixed text of multiple languages ​​on products. The aim is to reduce the false recognition rate and improve the overall recognition accuracy. Finally, all text content and its coordinate data in the video frame image are obtained. The coordinate data specifically includes the X / Y axis position, width, and height information of the text.

[0034] Step 104: Analyze the video frame images using a pre-trained multimodal large model, identify onomatopoeic text and subtitles through the model, and output the multimodal large model recognition results.

[0035] The video frame images are analyzed using a pre-trained multimodal large model. Specifically, the pre-trained multimodal large model Vision-Language is used to perform visual-text joint reasoning analysis on the video frame images. The model identifies on-screen text and subtitles, and outputs the multimodal large model recognition results, which include the bounding box coordinates of the confidence level.

[0036] In this embodiment, a multimodal large model is trained using a combination of visual localization and contextual description through example samples. Specific implementation steps are as follows: Visual localization is used to identify all possible text regions from video frames. Example video frames are input into the model, with white subtitles "Hello, welcome to watch" at the bottom and glowing text "Exclusive Reveal" in the top left corner. Input via a customizable model input interface: bottom and top left corner, for localization over a larger area.

[0037] Contextual descriptions are used to guide the distinction between on-screen text and subtitles. After all the text in the video frame image is recognized by OCR, the contextual descriptions are used to distinguish and label the recognized text.

[0038] Featured text usually includes special effects (such as glowing or outlining), and the overall structure is usually large. The area formed by marking the bounding box coordinates of the recognized text area is not standardized. Subtitles are usually presented as dialogue / narration text, usually located at the bottom of the screen. The font is relatively regular, and the bounding box needs to tightly wrap the text content. The area marked by the bounding box coordinates of the recognized text area is neat, with coordinates in pixels, calculated from the top left corner of the image (0,0), and the content corresponds to the audio and video.

[0039] For on-screen text, the input content via the customizable prompt model input interface includes tags, emotion words, and special effects descriptions such as glow and outlines. For subtitles, the input content via the customizable prompt model input interface can be key sentence content, which is compared with the recognized text and audio content, or no input is required, and the recognized text is compared with the audio content.

[0040] Before the conclusion is output, the identified on-screen text area is compared with the subtitle area to determine the size, and the semantics or audio content are compared to determine the on-screen text and subtitles.

[0041] In the model input interface, relevant parameters, such as the weight of the prompt and the confidence threshold, can be set to optimize the model's recognition performance.

[0042] Step 105: Select the areas of decorative text and subtitles that need to be erased by using coordinate matching, and determine them as the areas to be erased; In this embodiment, the coordinate data of the text in the OCR recognition result obtained by the multi-engine fusion OCR is matched and verified with the bounding box coordinates in the multimodal large model recognition result.

[0043] If the IoU threshold between the coordinate data and the bounding box coordinates is greater than or equal to 0.9, the coordinate regions of the decorative text and subtitles in the coordinate data are directly identified as the erasure regions. When the IoU is close to 1, it means that the overlap between the two regions is very high, so the erasure regions are directly identified in this case.

[0044] At the same time, the coordinate data is directly compared with the coordinates of the four vertices of the bounding box. If they are completely equal or the error is within the set threshold, it is determined to be a complete match.

[0045] If the IoU threshold between the coordinate data and the bounding box coordinates is between 0.6 and 0.9, it is considered a partial match. In this case, a secondary verification is needed to confirm the erased area, combining text semantics and contextual visual features.

[0046] Step 106: Based on the erased area, video restoration technology is used to intelligently erase and repair the on-screen text and subtitle areas, restoring the original video and generating a new video file after erasure.

[0047] The video restoration techniques used are existing technologies and will not be described in detail. In short, for static subtitles, inter-frame padding is used to merge corresponding pixels from adjacent frames, eliminating residual traces. For dynamic text areas, spatiotemporally coherent texture filling is performed based on generative adversarial networks (such as DeepFill v2 and CoModGAN) to ensure the restored area remains consistent with the background in its motion trajectory. A Poisson fusion algorithm is used to optimize the restored edges, making them smooth, and a color migration algorithm ensures that the restored area has a unified color tone with surrounding pixels. For the newly generated erased video file, the restored frames are reassembled in time, and frame interpolation is used to fill the gaps between dropped frames, ensuring video smoothness.

[0048] like Figure 2 The image shows a video text recognition and erasure method according to an embodiment of the present invention, including: the OCR further includes a filtering strategy. To achieve text content recognition in the video, the inherent text of the product, such as the product name, brand name, and ingredient list on the packaging, is determined as the retention area. In this embodiment, a dynamic rule base system is designed in conjunction with the video scene to automatically adjust the filtering strategy. The filtering strategy includes the following steps: Rule definition 201 defines different filtering rules to identify the inherent text of a product; Text extraction 202: Extracting inherent text related to products from video images using the multi-engine fusion OCR; Keyword matching 203 filters out inherent text related to products in videos and images through semantic analysis such as keyword matching and syntactic structure parsing. The output is 204, which represents the matched product text.

[0049] like Figure 3 The image shows a video text recognition and erasure method according to an embodiment of the present invention. A matching algorithm is used for filtering and matching. If the coordinate data and the bounding box coordinate IoU threshold are between 0.6 and 0.9, it is determined to be a partial match. At this point, it is necessary to combine text semantics and contextual visual features for secondary verification to confirm the erasure area. Specific steps include: Step 301: Text semantic analysis. Compare the text (non-preserved area) recognized by OCR with the subtitles and captions recognized by the multimodal model to determine whether they are semantically consistent or related. This can be achieved through natural language processing (NLP) techniques, such as word embedding and text similarity calculation.

[0050] Step 302 involves contextual visual feature analysis, considering visual features surrounding the erased area, such as color, font style, and layout. If these features match both the text region (non-preserved region) identified by OCR and the decorative text and subtitle regions identified by the multimodal model, the confidence level of the match is increased.

[0051] Step 303 involves a comprehensive judgment, combining the analysis results of text semantics and contextual visual features to determine whether to confirm the erasure area. This requires designing a decision model to calculate the final score based on the weights of different features. During this process, the IoU and content similarity thresholds are dynamically adjusted based on the video frame quality (resolution, clarity) to improve robustness.

[0052] By employing a two-stage verification and comprehensive matching strategy, we can reduce the risk of misjudgment while maintaining high efficiency, thereby improving the accuracy and reliability of erased region identification and reducing missed matches. This method has demonstrated excellent performance in practical applications, particularly when handling complex scenarios and diverse text layouts.

[0053] like Figure 4 As shown, an embodiment of the present invention provides a video text recognition and erasure system, comprising: Module 401 is used to acquire video; The video processing module 402 is used to perform frame extraction processing on the video to obtain video frame images; OCR recognition module 403 is used to recognize all text in the video frame image and obtain all text content and its coordinate data. The multimodal analysis module 404 includes a model input interface with customizable prompts, which is used by the multimodal large model to identify the on-screen text and subtitles in the video frame images through accurate prompts, and output the multimodal large model recognition results; The coordinate matching module 405 is used to filter out the areas of decorative text and subtitles that need to be erased by coordinate matching and determine them as the erasure areas; The video repair module 406 is used to intelligently erase and repair the text and subtitle areas of the erased area using video repair technology, restore the original video, and generate a new video file after erasure.

[0054] like Figure 5 As shown, the video text recognition and erasure method embodiment provided in this application can be executed on a mobile terminal, computer terminal, or similar terminal device. Taking its operation on a mobile terminal as an example, Figure 5 This is a hardware structure block diagram of a video text recognition and erasure method according to an embodiment of this application. (See diagram for reference.) Figure 1 As shown, a mobile terminal may include one or more ( Figure 5 Only one is shown in the diagram. A processor 501 (processor 501 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 502 for storing data are also shown. Optionally, the mobile terminal may further include a transmission device 503 for communication functions and an input / output device 504. Those skilled in the art will understand that... Figure 5 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 5 The more or fewer components shown, or having the same Figure 5 The different configurations shown.

[0055] The memory 502 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to a video text recognition and erasure method in this embodiment. The processor 501 executes various functional applications and data processing by running the computer program stored in the memory 502, thus implementing the aforementioned method. The memory 502 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 502 may further include memory remotely located relative to the processor 501, and these remote memories can be connected to the vehicle terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0056] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A video text recognition and erasure method, characterized in that, The method includes: Get the video; Perform frame extraction processing on the video to obtain video frame images; The OCR is used to identify all the text in the video frame image, obtain all the text content and its coordinate data, output the OCR recognition result, and determine the inherent text of the product that needs to be retained as the retention area; The video frame images are analyzed using a pre-trained multimodal large model to identify on-screen text and subtitles, and the multimodal large model recognition results are output. The areas of decorative text and subtitles that need to be erased are selected by coordinate matching and determined as the erasure areas; Based on the erased area, video restoration technology is used to intelligently erase and repair the on-screen text and subtitle areas, restoring the original video and generating a new video file after erasure.

2. The method according to claim 1, characterized in that, The frame extraction process is performed according to the frame rate of the video. For high frame rate videos, frame extraction is performed every four frames. For low frame rate videos, the frame extraction is encrypted or even performed continuously.

3. The method according to claim 1, characterized in that, Before recognizing all text in the video frame image via OCR, the process includes: performing noise reduction, contrast enhancement, and edge sharpening preprocessing on the acquired video frame image in sequence.

4. The method according to claim 1, characterized in that, The OCR is a multi-engine fusion OCR. Through the multi-engine fusion OCR, the OCR recognition result is obtained. The multi-engine fusion OCR processes the input video frame image using multiple engines and uses a voting mechanism to fuse the recognition results or compare similarity to obtain all text content and the coordinate data of the text. Specifically, the coordinate data is the X / Y axis position, width and height information of the text. The OCR also includes a dynamic rule base system for automatically adjusting filtering strategies. The dynamic rule base system includes: rule definition, text extraction, keyword matching, and result output.

5. The method according to claim 1, characterized in that, The analysis of the video frame images using a pre-trained multimodal large model includes: performing visual-text joint reasoning analysis on the video frame images using a pre-trained multimodal large model Vision-Language model, recognizing onomatopoeia and subtitles through the model, and outputting the multimodal large model recognition result, wherein the multimodal large model recognition result includes the bounding box coordinates of the confidence level.

6. The method according to claim 4, characterized in that, The method of filtering out the areas of decorative text and subtitles that need to be erased through coordinate matching includes: The coordinate data of the text in the OCR recognition result obtained by the multi-engine fusion OCR will be matched and verified with the bounding box coordinates in the multimodal large model recognition result. If the IoU threshold between the coordinate data and the bounding box coordinates is greater than or equal to 0.9, the coordinate areas of the decorative text and subtitles in the coordinate data will be directly identified as the erasure areas; If the IoU threshold between the coordinate data and the bounding box coordinates is between 0.6 and 0.9, a secondary verification is performed to confirm the erased area, combining text semantics and contextual visual features.

7. The method according to claim 6, characterized in that, The specific steps for performing secondary verification to confirm the erased area include: Text semantic analysis compares the text recognized by OCR with the text and subtitles recognized by the multimodal model, and uses natural language processing technology to determine whether they are semantically consistent or related. Contextual visual feature analysis, based on the visual features around the erased area, increases the confidence of the match if the visual features match the text area recognized by OCR and the decorative and subtitle areas recognized by the multimodal model. Based on a comprehensive assessment, combining the analysis results of text semantics and contextual visual features, a decision is made on whether to confirm the erasure area.

8. A video text recognition and erasure system, characterized in that, include: The acquisition module is used to acquire videos; The video processing module is used to perform frame extraction processing on the video to obtain video frame images; The OCR recognition module is used to recognize all text in the video frame image and obtain all text content and its coordinate data. The multimodal analysis module includes a model input interface with customizable prompts, which is used by the multimodal large model to identify the on-screen text and subtitles in the video frame images through accurate prompts, and output the multimodal large model recognition results; The coordinate matching module is used to filter out the areas of decorative text and subtitles that need to be erased by coordinate matching, and determine them as the areas to be erased; The video repair module is used to intelligently erase and repair the text and subtitle areas in the erased area using video repair technology, restore the original video, and generate a new video file after erasure.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method according to any one of claims 1 to 7 when executed.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 7.

Images