A reconstruction method and system for digital assets based on AI video dynamic modification

Abstract

The application discloses a kind of reconstruction method and system of digital assets based on AI video dynamic modification, belong to image communication technical field, including the following steps: first, identify the fuzzy area of target frame, and divide each sub-region, analyze edge information core index, determine the first processing strategy of each sub-region edge information, enhance to determine the second processing strategy of each enhanced sub-region, carry out light flow confidence detection, to determine the third processing strategy of corresponding sub-region, the reconstruction method of digital assets based on AI video dynamic modification provided by the application is combined by fine processing mechanism and data driving strategy, realizes from local problem to global quality accurate control, flexible strategy switching adapts to the scene of different fuzzy degree, edge feature, while improving the reconstruction efficiency of digital assets, maximizes the natural degree and reliability of dynamic modification, provides efficient and high-quality solution for asset reuse under complex video scene.

Description

Technical Field

[0001] This invention relates to the field of image communication technology, and in particular to a method and system for reconstructing digital assets based on AI-driven dynamic video modification. Background Technology

[0002] With the development of artificial intelligence, video reuse algorithms based on machine learning and deep learning have gradually emerged. Existing technologies mostly achieve video reuse through deep learning motion estimation algorithms. For example, DHVC2.0 (Hierarchical Predictive Learning Framework) decomposes video frames into multi-scale features through a hierarchical variational autoencoder, completely abandoning the block matching mode of traditional motion estimation. Its core lies in joint spatial-temporal modeling. The feature residuals of each scale are jointly predicted through temporal references of adjacent frames and spatial references of the same frame, effectively eliminating the computational bottleneck of motion compensation. At the same time, this technology adopts conditional entropy coding, and the probability model parameters are dynamically generated by context features, realizing adaptive adjustment of the bitrate.

[0003] For example, Chinese invention patent CN101969550B discloses an IP video stream multiplexing method and system, which includes: presetting a constant bitrate based on the bandwidth of each channel of the IPQAM modulator and the actual service needs, and multiplexing the original bitstream received from the streaming media server into a bitstream with the preset constant bitrate; analyzing the PAT and PMT in the original bitstream in real time to find the video PID, audio PID and PCR PID of each original bitstream; removing the PCR from the original video stream, and calculating a new PCR value based on the current bitrate and system clock and multiplexing it into the bitstream.

[0004] For example, the data multiplexing method and system disclosed in Chinese invention patent CN103974088A includes: establishing a data multiplexing protocol between a digital signal microprocessor and a multiplexing protocol parsing module; generating multiplexed data based on the data multiplexing protocol, the video data to be transmitted, and the selected channel, and sending the multiplexed data to the multiplexing protocol parsing module; parsing the multiplexed data according to the data multiplexing protocol, and copying the video data to be transmitted to the corresponding output channel of the video device that needs to receive video through a digital-to-analog converter chip.

[0005] The above-mentioned technology has at least the following technical problems:

[0006] Limited by current technology and complex video application scenarios, it is still impossible to completely eliminate single-frame positioning errors caused by factors such as blurred edges of high-speed moving objects, as well as a series of subsequent problems that affect AI's judgment of real edges. Summary of the Invention

[0007] On the one hand, a method for reconstructing digital assets based on AI-driven dynamic video modification, the method comprising:

[0008] The process of reconstructing digital assets by dynamically modifying AI videos involves: designating the reconstructed digital assets as target digital assets; performing fuzziness analysis on the target digital assets; identifying the target frames and obtaining the set of reference frames corresponding to the target frames; identifying the fuzzy regions of the target frames and dividing them into sub-regions, which are designated as sub-regions of the target frames; and thus obtaining the reference sub-regions of each reference frame corresponding to the target frames.

[0009] Obtain edge information of each sub-region, analyze the core indicators of edge information of each reference sub-region of each reference frame, and thus determine the first processing strategy for edge information of each sub-region.

[0010] When the first processing strategy is to perform sub-region enhancement, the corresponding sub-region is recorded as each enhanced sub-region. Enhancement of each enhanced sub-region is performed based on the obtained ambiguity of each enhanced sub-region, and the second processing strategy of each enhanced sub-region is determined after enhancement.

[0011] When mapping is performed using either the first or second processing strategy, the optical flow confidence of each reference sub-region of each reference frame is detected, the confidence score of each reference sub-region of each reference frame is analyzed, the rationality label of direct mapping is determined, and thus the third processing strategy for the corresponding sub-region is determined.

[0012] On the other hand, this application provides a reconstruction system for digital assets based on AI video dynamic modification, including: an execution object determination module, which executes the reconstruction process of digital assets based on AI video dynamic modification, records the reconstructed digital assets as target digital assets, performs fuzziness analysis on the target digital assets, identifies the target frame and obtains the reference frame set corresponding to the target frame, identifies the fuzzy region of the target frame and divides it into sub-regions, which are recorded as sub-regions of the target frame, thereby obtaining each reference sub-region of each reference frame corresponding to the target frame.

[0013] The execution strategy determination module acquires the edge information of each sub-region, analyzes the core indicators of the edge information of each reference sub-region in each reference frame, and thereby determines the first processing strategy for the edge information of each sub-region.

[0014] The strategy execution module, when the first processing strategy is to perform sub-region enhancement, records the corresponding sub-region as each enhanced sub-region, performs enhancement of each enhanced sub-region based on the obtained ambiguity of each enhanced sub-region, and determines the second processing strategy of each enhanced sub-region after enhancement.

[0015] The rationality determination module performs optical flow confidence detection on each reference sub-region of each reference frame when the first or second processing strategy is mapping, analyzes the confidence score of each reference sub-region of each reference frame, determines the rationality label of direct mapping, and thus determines the third processing strategy for the corresponding sub-region.

[0016] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0017] 1. This invention forms a multi-dimensional collaborative processing system. By combining a refined processing mechanism (sub-region division) with a data-driven strategy (edge ​​information, optical flow confidence), it achieves precise control from local problems to global quality. A closed-loop feedback mechanism ensures the traceability and optimization of processing quality. Flexible strategy switching (direct mapping / model prediction) adapts to scenarios with different degrees of ambiguity and edge features. While improving the efficiency of digital asset reconstruction, it maximizes the naturalness and reliability of dynamic modifications, providing an efficient and high-quality solution for asset reuse in complex video scenarios.

[0018] 2. This invention performs fuzzy analysis on target digital assets, accurately identifies the target frame and its corresponding set of reference frames, and divides the fuzzy area of ​​the target frame into sub-regions, achieving a refined breakdown of the problem area. Its advantage lies in breaking away from the traditional, coarse-grained approach of overall processing. By focusing on the fuzzy localities through sub-region division, it lays the foundation for subsequent targeted processing. This avoids ineffective operations on clear areas and concentrates resources on solving fuzzy problems, significantly improving the accuracy and efficiency of processing.

[0019] 3. This invention establishes a "closed-loop optimization" mechanism by determining a second processing strategy based on the actual effect of the sub-region (whether the ambiguity meets the standard) after enhancement. Its advantage lies in dynamic feedback adjustment: if the ambiguity meets the standard after enhancement, it is marked as complete; if it does not, further processing (such as secondary enhancement) is triggered, avoiding the limitations of a "one-size-fits-all" approach and ensuring that each enhanced sub-region ultimately reaches the quality threshold, fundamentally guaranteeing the basic quality of digital assets.

[0020] 4. This invention evaluates the reliability of motion trajectories by detecting optical flow confidence and determines a third processing strategy by combining edge characteristic correction coefficients. It takes into account both dynamic matching accuracy and edge quality. Direct mapping of high-confidence regions can ensure motion continuity and edge consistency, while model prediction of low-confidence regions can make up for sample defects by fitting motion equations. The flexible switching between the two strategies reduces the risk of incorrect mapping and improves the adaptability in complex scenes, ensuring the naturalness of cross-frame reuse. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a method for reconstructing digital assets based on AI video dynamic modification, provided by an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the structure of a digital asset reconstruction system based on AI video dynamic modification provided in an embodiment of the present invention;

[0024] Figure 3 This is an execution strategy diagram of a digital asset reconstruction method based on AI video dynamic modification provided in an embodiment of the present invention;

[0025] Figure 4 This is a mind map of a digital asset reconstruction system based on AI-driven video dynamic modification, provided by an embodiment of the present invention.

[0026] Figure 5 This is a diagram of the digital asset preprocessing interface of the digital asset reconstruction platform involved in this embodiment of the invention.

[0027] Figure 6 This is a continuation of the diagram of the digital asset preprocessing interface of the digital asset reconstruction platform involved in the embodiments of the present invention.

[0028] Figure 7 This is a diagram of the digital asset reconstruction interface of the digital asset reconstruction platform involved in this embodiment of the invention.

[0029] Figure 8 This is a diagram of the digital asset export interface of the digital asset reconstruction platform involved in this embodiment of the invention. Detailed Implementation

[0030] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0031] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0032] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0033] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0034] Reference Figure 1 The flowchart shown illustrates a digital asset reconstruction method based on AI-driven video dynamic modification. First, blurred regions of the target frame are identified and divided into sub-regions. The core indicators of edge information for each reference sub-region of each reference frame are analyzed to determine a first processing strategy for the edge information of each sub-region. After enhancement, a second processing strategy for each enhanced sub-region is determined. When mapping is performed using either the first or second processing strategy, optical flow confidence is detected for each reference sub-region of each reference frame, thereby determining a third processing strategy for the corresponding sub-region. The processing flow of this method may include the following steps:

[0035] The process of reconstructing digital assets by dynamically modifying AI videos involves: designating the reconstructed digital assets as target digital assets; performing fuzziness analysis on the target digital assets; identifying the target frames and obtaining the set of reference frames corresponding to the target frames; identifying the fuzzy regions of the target frames and dividing them into sub-regions, which are designated as sub-regions of the target frames; and thus obtaining the reference sub-regions of each reference frame corresponding to the target frames.

[0036] It should be noted that the process of reconstructing digital assets for AI-driven video dynamic modification is called digital asset reuse: existing digital assets such as video scenes and characters are migrated and applied to the creation of other video content through technical means, so as to realize the dynamic adjustment and reconstruction of video content and reduce the cost of repeated development.

[0037] like Figure 5 The diagram shown is a digital asset preprocessing interface diagram of the digital asset reconstruction platform involved in this embodiment of the invention. This interface displays a fuzzy analysis area, which displays the fuzzy state of video frames according to the degree of fuzziness, thereby identifying the target frame.

[0038] Figure 6This is a continuation of the digital asset preprocessing interface diagram of the digital asset reconstruction platform involved in the embodiments of the present invention. Figure 5 The target frame shown is displayed in the region processing area, showing the state of the target frame before and after enhancement.

[0039] Furthermore, the reference sub-regions of each reference frame corresponding to the target frame are obtained, and the specific analysis method is as follows:

[0040] Blur detection is performed on the reconstructed video sequence, and frames with blur exceeding the blur threshold are marked as target frames.

[0041] It should be noted that blur detection involves extracting keyframes from the reconstructed video sequence and removing duplicate or redundant frames. An inter-frame registration algorithm is used to eliminate displacement deviations caused by camera shake, ensuring that blur detection focuses on the content itself rather than motion interference. The Farneback algorithm is used to calculate the optical flow field of adjacent frames, and the least squares algorithm is used to solve for the optical flow vectors. The variance of all optical flow vectors in the current frame is calculated, and a real-time motion error matrix is ​​constructed based on the obtained variance. The gain matrix and error matrix are calculated based on the optical flow vectors of the current frame and between the current and next frames. The Kalman gain coefficient is calculated from the gain matrix, error matrix, and real-time motion error matrix and denoted as the blur index. The mean of the global blur index is taken as the blur degree.

[0042] Extract the preset blur threshold from the database. If the blur of a frame exceeds the blur threshold, it indicates that the frame has obvious ghosting or chaotic motion trajectory, reflecting pixel overlap caused by dynamic blur. The dynamic blur problem is prominent, so it needs to be adjusted and marked as the target frame.

[0043] If the blurriness of a frame does not exceed the blurriness threshold, it means that the motion trajectory of that frame is stable, coherent and natural, without abnormal jumps, and the dynamic transition is smooth, so no processing is required.

[0044] The number of reference frames for the target frame is determined based on the difference between the ambiguity and the ambiguity threshold.

[0045] It should be noted that the number of reference frames corresponding to the ambiguity difference interval is extracted from the database, and the number of reference frames corresponding to the ambiguity difference interval is also extracted and mapped, and named as the number of reference frames of the target frame.

[0046] It should be added that the larger the blur difference, the greater the deviation between the blur and the blur threshold, meaning the more obvious the dynamic blur problem of the frame. In order to make the frame clearer, more reference frames need to be extracted.

[0047] Identify the target frame and obtain the set of reference frames corresponding to the target frame. Identify the blurred regions of the target frame and divide them into sub-regions, which are denoted as the sub-regions of the target frame. Thus, the reference sub-regions of each reference frame corresponding to the target frame are obtained.

[0048] It should be noted that the reference frame set corresponding to the target frame is a set of reference frames whose number of reference frames is obtained forward and backward according to the sequence position of the target frame.

[0049] This invention performs fuzzy analysis on target digital assets to accurately identify the target frame and its corresponding set of reference frames, and divides the fuzzy area of ​​the target frame into sub-regions, achieving a refined breakdown of the problem area. Its advantage lies in breaking away from the traditional, coarse-grained approach of overall processing. By focusing on the fuzzy localities through sub-region division, it lays the foundation for subsequent targeted processing. This avoids ineffective operations on clear areas and concentrates resources on solving fuzzy problems, significantly improving the accuracy and efficiency of the processing.

[0050] Obtain edge information of each sub-region, analyze the core indicators of edge information of each reference sub-region of each reference frame, and thus determine the first processing strategy for edge information of each sub-region.

[0051] like Figure 3 The diagram shows the execution strategy of a digital asset reconstruction method based on AI video dynamic modification provided by an embodiment of the present invention. First, the core indicators of the edge information of each reference sub-region of each reference frame are analyzed to determine the first processing strategy of the edge information of each sub-region. Then, the enhancement of each enhanced sub-region is performed based on the obtained ambiguity of each enhanced sub-region. After enhancement, the second processing strategy of each enhanced sub-region is determined. When the first processing strategy or the second processing strategy is to perform mapping, the optical flow confidence of each reference sub-region of each reference frame is detected.

[0052] Furthermore, the core indicators of edge information for each reference sub-region of each reference frame are analyzed. The specific analysis method is as follows:

[0053] Obtain the edge information of each reference sub-region of each reference frame.

[0054] Collect edge information parameters, including edge width, gradient intensity, and percentage of connected pixels.

[0055] It should be noted that the edge width reflects the pixel span of the grayscale transition zone. The larger the width, the more blurred the transition zone becomes, the weaker the sense of boundary between the object and the background, and the lower the sharpness and recognizability of the edge. The smaller the core indicator of edge information is.

[0056] Gradient strength is the gradient magnitude of the edge point in the current frame. True edges are usually accompanied by significant gray-scale jumps. The greater the gradient strength, the higher the spatial dimensionality and the greater the core indicator of edge information.

[0057] The proportion of connected pixels is the percentage of valid edge points in the connected domain to which an edge point belongs. It reflects the integrity of the local structure. The larger the proportion of connected pixels, the more coherent the edge pixels are, the higher the structural consistency, and the greater the core indicator of edge information.

[0058] The greater the gradient strength, the more drastic the grayscale change, the narrower the pixel span of the grayscale transition zone, and the smaller the edge width. The greater the gradient strength, the higher the proportion of connected pixels. Clear high-gradient edges are less likely to break, and the proportion of completely connected pixels increases accordingly. The greater the edge width, the smaller the gradient strength. A wide transition zone means a smooth grayscale change, and the corresponding grayscale change rate will decrease.

[0059] It should be noted that the edge width can be determined by first identifying the edge detected by the Canny algorithm as the "central axis" of the image where grayscale changes most drastically. Using this edge as the center, the search extends outwards to find the complete range of grayscale values ​​transitioning from the "peak on one side of the edge" to the "peak on the other side" (e.g., the area transitioning from a grayscale value of 200 inside the object to a grayscale value of 50 in the background). Within this transition area, the number of pixels whose grayscale value drops from the "90% peak" to the "10% peak" is counted; this number represents the edge width.

[0060] Gradient strength is calculated by determining the image gradient. The gradient components (Gx, Gy) in the x and y directions are calculated using the gradient operator on the original image, and the gradient magnitude is then calculated. This is done through the formula... Where G represents the gradient magnitude, Gx represents the component of the gradient magnitude in the X direction, and Gy represents the component of the gradient magnitude in the Y direction, the gradient intensity (magnitude) of each pixel is obtained. The gradient intensity of edge regions is extracted. For edge pixels obtained from edge detection, the average gradient intensity is taken as the gradient intensity of that edge.

[0061] The proportion of connected pixels is determined by performing connectivity analysis on the binary edge map of a single frame. Using 8-neighborhood (including diagonals) connectivity analysis, connected components in edge pixels are labeled (each connected component is a group of consecutive edge pixels), the proportion is calculated, the total number of pixels at that edge is counted, and the sum of the pixels of all connected components (denoted as C) is also calculated. The proportion of connected pixels is then calculated as follows: , where H represents the percentage of connected pixels, C represents the sum of pixels in all connected components, and T represents the total number of pixels.

[0062] Extract the preset edge width feature assignment multiplier, gradient intensity feature assignment multiplier, and connected pixel ratio feature assignment multiplier from the database.

[0063] It should be added that the values ​​of the edge width feature allocation multiplier, gradient intensity feature allocation multiplier, and connected pixel percentage feature allocation multiplier all range from 0 to 1, and the sum of the edge width feature allocation multiplier, gradient intensity feature allocation multiplier, and connected pixel percentage feature allocation multiplier is 1. When using them, the pre-set values ​​can be directly extracted from the database. For example, the edge width, gradient intensity, and connected pixel percentage are mapped one-to-one with their respective edge width feature allocation multipliers, gradient intensity feature allocation multipliers, and connected pixel percentage feature allocation multipliers. When using them, the real-time obtained edge width, gradient intensity, and connected pixel percentage are input into the corresponding mapping sets, thereby extracting the edge width feature allocation multiplier, gradient intensity feature allocation multiplier, and connected pixel percentage feature allocation multiplier.

[0064] Extract the preset edge width reference value, gradient strength reference value, and connected pixel ratio reference value from the database.

[0065] Analyze the core indicators of edge information based on edge information parameters.

[0066] The core indicators of edge information are the edge width, gradient strength, and connected pixel ratio, which together quantify the edge information. The specific analysis process is as follows: The collected gradient strength and connected pixel ratio are compared with the corresponding reference values, and the reference value of the edge width is compared with the edge width. The results of each comparison are coupled with the corresponding feature allocation multipliers to obtain the core indicators of edge information.

[0067] In this embodiment, the core indicators of edge information are analyzed in the following specific process:

[0068] ,

[0069] Where F represents the core index of edge information, A represents the edge width, A0 represents the reference value of edge width, m represents the edge width feature allocation multiplier, S represents the gradient strength, S0 represents the gradient strength reference value, n represents the gradient strength feature allocation multiplier, D represents the proportion of connected pixels, D0 represents the reference value of the proportion of connected pixels, and u represents the proportion of connected pixels feature allocation multiplier.

[0070] By traversing each reference sub-region of each reference frame, the core indicators of edge information for each reference sub-region of each reference frame are obtained.

[0071] Furthermore, the primary processing strategy for the edge information of each sub-region is determined, and the specific analysis method is as follows:

[0072] Extract the preset threshold values ​​of core indicators for edge information from the database.

[0073] For a sub-region, if the core index of the edge information of the corresponding reference sub-region of a certain reference frame is greater than or equal to the threshold of the core index of the edge information, then the corresponding reference sub-region of the reference frame is recorded as a highly similar sub-region.

[0074] It should be noted that if the core index of the edge information of the corresponding reference sub-region of a certain reference frame is greater than or equal to the threshold of the core index of the edge information, it means that the edge features of the sub-region are in a stable and reliable state. In this case, the edge information of the reference sub-region has high reference value and can be used as a reliable basis for the reconstruction of the corresponding sub-region of the target frame. It can reduce the processing error caused by the instability of edge features and improve the accuracy and naturalness of the reconstructed edges. Therefore, the corresponding reference sub-region of the reference frame is recorded as a highly similar sub-region.

[0075] If the core index of the edge information of the corresponding reference sub-region of a certain reference frame is less than the threshold of the core index of the edge information, then the corresponding reference sub-region of the reference frame is recorded as a low similarity sub-region.

[0076] It should be noted that if the core edge information index of the corresponding reference sub-region of a certain reference frame is less than the threshold of the core edge information index, it indicates that the edge quality of the sub-region has obvious defects and is difficult to use as a reliable benchmark for asset reuse. This may lead to edge matching misalignment, awkward blending, or even affect the realism of the overall image in the new scene. Therefore, the corresponding reference sub-region of the reference frame is recorded as a low similarity sub-region.

[0077] By traversing each reference frame, the number of highly similar sub-regions is obtained.

[0078] If the number of highly similar sub-regions is greater than or equal to the threshold for the number of highly similar sub-regions, then the first processing strategy for the edge information of that sub-region is called detail enhancement.

[0079] It should be noted that if the number of highly similar sub-regions is greater than or equal to the threshold of the number of highly similar sub-regions, it indicates that there are enough sub-regions in the reference frame set that are highly matched with the edge features of the target sub-region. This reflects that the edge features of the target sub-region have high reusability and adaptability among similar assets, which helps to improve the efficiency and quality stability of dynamic modification of digital assets. Therefore, the first processing strategy for the edge information of this sub-region is called detail enhancement.

[0080] If the number of highly similar sub-regions is less than the threshold for the number of highly similar sub-regions, then the first processing strategy for the edge information of each region is denoted as waiting for mapping.

[0081] It should be noted that if the number of highly similar sub-regions is less than the threshold for the number of highly similar sub-regions, it indicates that the edge features of the target sub-region lack sufficient matching samples in the reference frame set, and its features have strong randomness. This means that the edge features of the target sub-region are not universal enough and it is difficult to cross-validate through multiple similar samples. In this case, it is necessary to make up for it by mapping. Therefore, the first processing strategy for the edge information of this sub-region is called waiting for mapping.

[0082] When the first processing strategy is to perform detail enhancement, the corresponding sub-regions are recorded as each enhanced sub-region. Enhancement of each enhanced sub-region is performed based on the obtained ambiguity of each enhanced sub-region, and a second processing strategy for each enhanced sub-region is determined after enhancement.

[0083] Furthermore, enhancement of each enhanced sub-region is performed based on the obtained ambiguity of each enhanced sub-region. The specific analysis method is as follows:

[0084] When the first processing strategy is to perform sub-region enhancement, the corresponding sub-region is recorded as the enhanced sub-region, thus obtaining each enhanced sub-region.

[0085] Obtain the blurriness of each enhanced sub-region.

[0086] The ambiguity deviation value of each enhanced sub-region is obtained by performing difference processing between the ambiguity of each enhanced sub-region and the ambiguity threshold.

[0087] The sharpening intensity of each enhanced sub-region is adjusted based on the fuzzy deviation value of each enhanced sub-region.

[0088] It should be noted that the adjustment sharpening intensity corresponding to each fuzzy deviation value range is extracted from the database, and the adjustment sharpening intensity corresponding to the fuzzy deviation value range is also extracted and mapped, and named as the fuzzy deviation adjustment sharpening intensity of each enhanced sub-region.

[0089] It should be added that the larger the blur deviation value, the more significant the deviation in blur level of the enhanced sub-region, and the more blurred the region. In order to compensate for this mismatch of blur features, the intensity of sharpening adjustment for extraction should be greater.

[0090] Sharpening is performed on each enhanced sub-region by adjusting the sharpening intensity based on the blur deviation of each enhanced sub-region.

[0091] It should be noted that the current sharpening intensity value of each enhanced sub-region is extracted, and the sharpening intensity of each enhanced sub-region is adjusted by the blur deviation and added to the current sharpening intensity value to obtain the sharpening enhancement value of each enhanced sub-region.

[0092] Furthermore, a second processing strategy is determined for each enhanced sub-region after enhancement. The specific analysis method is as follows:

[0093] After enhancement, the ambiguity of each enhanced sub-region is re-detected. If the ambiguity of a certain enhanced sub-region is greater than or equal to the ambiguity threshold, the second processing strategy for that enhanced sub-region is recorded as waiting for mapping.

[0094] It should be noted that if the blur of a certain enhanced sub-region is greater than or equal to the blur threshold, it means that the sharpening enhancement of this region has not achieved the expected effect, and the blur problem has not been effectively improved. The edge clarity and detail recognition of such regions are still unqualified, and secondary optimization is required for this region. Therefore, the second processing strategy for this enhanced sub-region is denoted as waiting for mapping.

[0095] If the ambiguity of a certain enhanced sub-region is less than the threshold, then the second processing strategy for that enhanced sub-region is recorded as generating enhancement completion signals for each enhanced sub-region.

[0096] It should be noted that if the blur of a certain enhanced sub-region is less than the threshold, it indicates that the sharpening enhancement of that region has achieved the expected effect, the blur deviation value and the sharpening adjustment coefficient are accurately matched, the edge blur problem is effectively repaired, and the edge clarity and detail recognition of the region are in a qualified state. Therefore, the second processing strategy for this enhanced sub-region is recorded as generating the enhancement completion signal for each enhanced sub-region.

[0097] This invention establishes a "closed-loop optimization" mechanism by determining a second processing strategy based on the actual effect of the sub-region (whether the ambiguity meets the standard) after enhancement. Its advantage lies in dynamic feedback adjustment: if the ambiguity meets the standard after enhancement, it is marked as complete; if it does not, further processing (such as secondary enhancement) is triggered, avoiding the limitations of a "one-size-fits-all" approach. This ensures that each enhanced sub-region ultimately reaches the quality threshold, fundamentally guaranteeing the basic quality of digital assets.

[0098] like Figure 7 The diagram shown is a digital asset reconstruction interface of the digital asset reconstruction platform according to an embodiment of the present invention. It displays the before and after mapping effects of video frames waiting to be mapped and provides an interactive window for operators to configure mapping settings.

[0099] When the first or second processing strategy is to wait for mapping, the optical flow confidence of each reference sub-region of each reference frame is detected, the confidence score of each reference sub-region of each reference frame is analyzed, the rationality label of direct mapping is determined, and thus the third processing strategy of the corresponding sub-region is determined.

[0100] Furthermore, the rationality labels for direct mapping are determined, and the specific analysis method is as follows:

[0101] Optical flow confidence is detected for each reference sub-region of each reference frame to obtain the reliability score of the optical flow vector for each reference sub-region of each reference frame.

[0102] It should be noted that optical flow confidence detection is performed for each reference sub-region of each reference frame. A dense optical flow algorithm (such as the Farneback algorithm) is used to calculate the optical flow field of adjacent reference frames, obtaining the motion vectors (including direction and magnitude information) of all pixels within each sub-region. The optical flow vectors are then preliminarily filtered to remove outliers exceeding the physical motion range. The difference between the predicted optical flow value and the actual pixel grayscale value is calculated; the smaller the difference, the higher the optical flow confidence. The formula is: Difference = |I(x+t,y+s)-I(x,y)| (where I is the pixel grayscale value, and t and s are the optical flow vector components), where x refers to the actual pixel grayscale value in the x-direction, and y refers to the actual pixel grayscale value in the y-direction.

[0103] Edge characteristic correction coefficients for each reference sub-region are extracted based on the core indicators of edge information for each reference sub-region in each reference frame.

[0104] It should be noted that the edge characteristic correction coefficients corresponding to the core indicator intervals of each edge information are extracted from the database. At the same time, the edge characteristic correction coefficients corresponding to the intervals where the core indicators of the edge information are located are mapped and extracted, and named as the edge characteristic correction coefficients of each reference sub-region.

[0105] It should be added that the larger the core index of edge information, the better the edge quality of the reference sub-region, the smaller the correction force required, and the smaller the correction coefficient for the extracted edge characteristics.

[0106] The confidence score of each reference sub-region of each reference frame is obtained based on the reliability score of the optical flow vector of each reference sub-region of each reference frame and the edge characteristic correction coefficient of each reference sub-region.

[0107] It should be noted that the confidence score of each reference sub-region of each reference frame is obtained by multiplying the reliability score of the optical flow vector of each reference sub-region of each reference frame with the edge characteristic correction coefficient of each reference sub-region.

[0108] It should be explained that the reliability of the optical flow vector only reflects the credibility of dynamic motion features, while the edge characteristic correction coefficient only reflects the static edge quality. Relying solely on one indicator may lead to misjudgments such as "dynamically reliable but with blurred edges" or "clear edges but chaotic motion trajectories." Combining both indicators can take into account both dynamic and static features, ensuring that the sub-region meets the reuse requirements in terms of both motion matching and edge fusion.

[0109] For a sub-region, if the confidence score of a certain reference frame is greater than or equal to the confidence score threshold, then the corresponding sub-region of the reference frame is recorded as a high-confidence region.

[0110] It should be noted that if the confidence score of a certain reference frame is greater than or equal to the confidence score threshold, it means that the overall quality of the sub-region in terms of both dynamic motion and static edge meets the benchmark requirements for asset reuse. It satisfies both natural motion matching and seamless edge blending. Therefore, the corresponding sub-region of the reference frame is recorded as a high-confidence region.

[0111] If the confidence score of a reference frame is less than the confidence score threshold, the corresponding sub-region of the reference frame is recorded as a low-confidence region.

[0112] It should be noted that if the confidence score of a certain reference frame is less than the confidence score threshold, it means that the sub-region may have insufficient reliability of optical flow vector, chaotic motion trajectory, or blurred edges. In this case, the region cannot guarantee dynamic coherence and it is difficult to meet edge fusion. Therefore, the corresponding sub-region of the reference frame is recorded as a low confidence region.

[0113] By iterating through each reference frame, the number of high-confidence regions is obtained.

[0114] If the number of high-confidence regions is greater than or equal to the threshold for the number of high-confidence regions, then the direct mapping rationality label will be recorded as direct mapping rationality.

[0115] It should be noted that if the number of high-confidence regions is greater than or equal to the threshold for the number of high-confidence regions, it indicates that there are enough high-confidence regions in the reference frame set to serve as a reliable benchmark. These regions also possess stable dynamic motion characteristics, and meeting the quantity threshold means that the core requirements of the target reuse scenario can be covered. In this case, directly mapping the features of these regions to the target frame can ensure the coherence and clarity of dynamic matching without the need for additional large-scale optimization, thus guaranteeing both reuse efficiency and quality stability.

[0116] If the number of high-confidence regions is less than the threshold for the number of high-confidence regions, then the direct mapping rationality label will be marked as direct mapping unreasonable.

[0117] It should be noted that if the number of high-confidence regions is less than the threshold for the number of high-confidence regions, it means that there are not enough reliable samples in the reference frame set to support direct reuse. A small number or low-confidence regions mean that most sub-regions have unreliable dynamic motion or problems with blurred or broken edges. Direct mapping may lead to batch quality problems such as motion misalignment and edge fragmentation in the new video. Therefore, the reasonableness label of direct mapping is recorded as unreasonable direct mapping.

[0118] Furthermore, the third processing strategy for the corresponding sub-region is determined, and the specific analysis method is as follows:

[0119] If the direct mapping rationality label is "direct mapping is reasonable", then the third processing strategy for the corresponding sub-region is direct mapping, which directly maps the sub-region corresponding to the reference frame of the highest confidence region to the blurred part of the target frame.

[0120] It should be noted that the highest confidence region is the sample with the best features in the reference frame set. Its feature accuracy has been quantitatively verified and can be used as a standard template to directly adapt to the target scene. It is mapped to the blurred part of the target frame. Essentially, it replaces the defective region with verified high-quality features, which can quickly make up for the quality shortcomings of the target frame and ensure the continuity of the features after replacement.

[0121] If the direct mapping rationality label is that the direct mapping is unreasonable, then the third processing strategy for the corresponding sub-region is model prediction. The motion trajectory reference length is determined based on the core indicators of edge information, and the second-order motion equation is fitted based on the motion trajectory reference length.

[0122] It should be noted that in the absence of sufficiently high-confidence regions to support direct reuse, reliable features are generated through algorithmic prediction to compensate for the lack of samples.

[0123] Model prediction determines the motion trajectory reference length based on the core indicators of edge information, extracts the motion trajectory within the edge reference length, fits the second-order motion equation, transforms it into a matrix equation, solves it using the least squares method, and outputs the predicted edge position of the target frame.

[0124] It should be explained that if the edge core index is high, it indicates that the edge features of the sub-region are more stable and the historical motion trajectory has higher reference value. A shorter reference length can be used to avoid introducing irrelevant interference from too far frames. If the edge core index is low, it indicates poor feature stability. A longer historical trajectory is needed to capture the overall motion trend in order to offset the impact of local fluctuations.

[0125] Based on a defined reference length, motion trajectory data of sub-regions in historical frames are extracted and fitted using second-order motion equations (constant acceleration model). Let the time series be t1, t2, t3, ..., tn, where n is the reference length and the corresponding pixel coordinates be (x1, y1), (x2, y2), (x3, y3), ..., (x...). n ,y n The second-order equation is in the form of: , Where X represents the coordinates of the predicted target frame pixel in the X direction, Y represents the coordinates of the predicted target frame pixel in the Y direction, t is the time series, and the coefficients a are solved using the least squares method. x ,b x ,c x ,a y ,b y,c y The fitted equation is used to predict the target frame time point t. n+1 The pixel coordinates are used to obtain the motion trajectory of the sub-region in the target frame.

[0126] This invention evaluates the reliability of motion trajectories by detecting optical flow confidence and determines a third processing strategy by combining edge characteristic correction coefficients. It takes into account both dynamic matching accuracy and edge quality. Direct mapping of high-confidence regions can ensure motion continuity and edge consistency, while model prediction is used for low-confidence regions to make up for sample defects by fitting motion equations. The flexible switching between the two strategies reduces the risk of incorrect mapping and improves the adaptability in complex scenes, ensuring the naturalness of cross-frame reuse.

[0127] Furthermore, it also includes constraining the rate of change of sub-regions based on the rate of curvature change and gradient direction deviation, with the specific analysis method as follows:

[0128] Traverse each sub-region. If a completion signal exists in each sub-region, obtain the overall trajectory of the target frame and perform verification analysis on its curvature and gradient direction.

[0129] If the rate of change of curvature is greater than or equal to the rate of change of curvature threshold and the gradient direction deviation is greater than or equal to the gradient direction deviation threshold, it is determined to be a structural abrupt change anomaly point, and an early warning message is generated.

[0130] It should be noted that if the rate of change of curvature is greater than or equal to the rate of change of curvature threshold and the gradient direction deviation is greater than or equal to the gradient direction deviation threshold, it indicates that there is a significant abnormality in the edge features of the region. The curvature of the edge curve is severe and unstable, and the gray-scale change direction of the edge is chaotic. The edge of the region is not a naturally formed continuous structure. There may be edge breakage and reconstruction errors caused by noise interference, motion blur or occlusion. In this case, it is judged as a structural abrupt change anomaly point, and a warning message is generated.

[0131] In this embodiment, the warning message could be: "Warning! A structural mutation anomaly has occurred."

[0132] If the rate of change of curvature is greater than or equal to the rate of change of curvature threshold but the gradient direction deviation is less than or equal to the gradient direction deviation threshold, it is considered as natural edge curvature and abnormal residuals are not calculated.

[0133] It should be noted that if the rate of change of curvature is greater than or equal to the rate of change of curvature threshold but the gradient direction deviation is less than or equal to the gradient direction deviation threshold, it means that the region has a complex shape but a stable structural direction. It belongs to a naturally existing complex edge, rather than an anomaly caused by noise or motion blur. Although its shape is irregular, its structural direction is stable. It belongs to a regular complex edge, so the abnormal residual is not calculated.

[0134] If the rate of change of curvature is less than the rate of change of curvature threshold but the gradient direction deviation is greater than or equal to the threshold, perform abnormal residual analysis.

[0135] It should be noted that if the rate of change of curvature is less than the rate of change of curvature threshold but the gradient direction deviation is greater than or equal to the threshold, it indicates that the overall edge shape is gentle and conforms to the continuous trend of natural edges. However, if the gradient direction deviation exceeds the threshold, it means that there is a significant abrupt change in the grayscale change direction of local pixels, so abnormal residual analysis is required.

[0136] Perform anomaly residual analysis, specifically: obtain the residual between the mapped edge gradient direction and the actual edge gradient direction, and record it as anomaly residual.

[0137] It should be noted that the gradient direction residual is obtained by subtracting the actual edge gradient direction from the mapped edge gradient direction, and is denoted as the abnormal residual.

[0138] If the abnormal residual is less than the abnormal residual threshold, the edge position is retained and an optimization completion signal is generated.

[0139] It should be noted that if the abnormal residual is less than the abnormal residual threshold, it indicates that the gradient direction deviation problem in this region is small, the edge features are relatively flat, and the gray-scale change direction of the edge is relatively consistent. There is no need to continue to invest resources for iterative repair. The edge position is retained, and the optimization completion signal is generated.

[0140] If the abnormal residual is greater than or equal to the abnormal residual threshold, an early warning message is generated.

[0141] It should be noted that if the abnormal residual is greater than or equal to the abnormal residual threshold, it means that although the edge of the region has a smooth overall shape (normal rate of curvature change), there are still significant contradictions in the local structure. The disorder of the gradient direction breaks the characteristic logic of the natural edge, which is an edge with a qualified shape but a failed structure, so a warning message is generated.

[0142] In this embodiment, the warning message could be: "Attention! Abnormal residual exceeds the threshold."

[0143] If the rate of change of curvature is less than the rate of change of curvature threshold and the gradient direction deviation is less than the gradient direction deviation threshold, it is determined to be a structural optimization point, and a completion signal is generated.

[0144] It should be noted that if the rate of change of curvature is less than the threshold for the rate of change of curvature and the gradient direction deviation is less than the threshold for the gradient direction deviation, it means that the curvature of the curve changes gently in this region, without any sharp turns or abrupt changes. The curve shape is relatively regular and smooth, the edge direction is stable, without any chaotic directional changes, and the continuity and regularity of the edge are strong. In this case, it is determined to be a structural optimization point, and a completion signal is generated.

[0145] In this embodiment, the completion signal can be: "Optimization complete".

[0146] This invention forms a multi-dimensional collaborative processing system. By combining a refined processing mechanism (sub-region division) with a data-driven strategy (edge ​​information, optical flow confidence), it achieves precise control from local problems to global quality. A closed-loop feedback mechanism ensures the traceability and optimizability of processing quality. Flexible strategy switching (direct mapping / model prediction) adapts to scenarios with different degrees of ambiguity and edge features. While improving the efficiency of digital asset reconstruction, it maximizes the naturalness and reliability of dynamic modifications, providing an efficient and high-quality solution for asset reuse in complex video scenarios.

[0147] The embodiments of the present invention provide, as follows Figure 2 The diagram shows a structural schematic of a digital asset reconstruction system based on AI video dynamic modification. The processing flow of the system may include the following steps: an execution object determination module, used to execute the reconstruction process of digital assets based on AI video dynamic modification, denoting the reconstructed digital asset as the target digital asset, performing fuzziness analysis on the target digital asset, identifying the target frame and obtaining the reference frame set corresponding to the target frame, identifying the fuzzy region of the target frame and dividing it into sub-regions, denoted as the sub-regions of the target frame, thereby obtaining the reference sub-regions of each reference frame corresponding to the target frame.

[0148] The execution strategy determination module is used to acquire edge information of each sub-region, analyze the core indicators of edge information of each reference sub-region of each reference frame, and thereby determine the first processing strategy for edge information of each sub-region.

[0149] The strategy execution module is used to record the corresponding sub-region as each enhanced sub-region when the first processing strategy is to perform sub-region enhancement, perform enhancement of each enhanced sub-region based on the obtained ambiguity of each enhanced sub-region, and determine the second processing strategy of each enhanced sub-region after enhancement.

[0150] The rationality determination module is used to perform optical flow confidence detection of each reference sub-region of each reference frame when the first processing strategy or the second processing strategy is to perform mapping, analyze the confidence score of each reference sub-region of each reference frame, determine the rationality label of direct mapping, and thereby determine the third processing strategy of the corresponding sub-region.

[0151] like Figure 4 The image shown is a mind map of a digital asset reconstruction system based on AI video dynamic modification, provided by an embodiment of the present invention.

[0152] First, identify the blurred regions of the target frame and divide them into sub-regions. Analyze the core indicators of the edge information of each reference sub-region of each reference frame to determine the first processing strategy for the edge information of each sub-region. After enhancement, determine the second processing strategy for each enhanced sub-region. When mapping is performed using the first or second processing strategy, perform optical flow confidence detection to determine the third processing strategy for the corresponding sub-region.

[0153] like Figure 8 The diagram shown is a digital asset export interface of the digital asset reconstruction platform involved in this embodiment of the invention. Operators can set export parameters and output settings for the reconstructed video.

[0154] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product, which includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0155] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0156] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0157] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0158] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0159] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0160] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0161] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0162] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0163] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0164] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A reconstruction method of a digital asset based on AI video dynamic modification, characterized in that, The method comprises: Performing a reconstruction process of the AI video dynamic modification digital asset, marking the reconstructed digital asset as a target digital asset, performing blur analysis on the target digital asset, identifying a target frame and obtaining a set of reference frames corresponding to the target frame, identifying a blur area of the target frame and dividing it into sub-areas, marking the sub-areas of the target frame, thereby obtaining each reference sub-area of each reference frame corresponding to the target frame; Obtaining edge information of each sub-area, analyzing edge information core indicators of each reference sub-area of each reference frame, thereby determining a first processing strategy of the edge information of each sub-area, and the specific analysis method is as follows: Extracting preset edge information core indicator thresholds in a database; For a sub-area, if the edge information core indicator of the corresponding reference sub-area of a reference frame is greater than or equal to the edge information core indicator threshold, the corresponding reference sub-area of the reference frame is marked as a high-similarity sub-area; If the edge information core indicator of the corresponding reference sub-area of a reference frame is less than the edge information core indicator threshold, the corresponding reference sub-area of the reference frame is marked as a low-similarity sub-area; Iterating through each reference frame to obtain the number of high-similarity sub-areas; If the number of high-similarity sub-areas is greater than or equal to a high-similarity sub-area number threshold, the first processing strategy of the edge information of the sub-area is marked as detail enhancement; If the number of high-similarity sub-areas is less than the high-similarity sub-area number threshold, the first processing strategy of the edge information of the sub-area is marked as waiting for mapping; When the first processing strategy is to perform sub-area enhancement, the corresponding sub-area is marked as each enhanced sub-area, based on the obtained blur of each enhanced sub-area, the enhancement of each enhanced sub-area is performed, and the second processing strategy of each enhanced sub-area is determined after the enhancement; When the first processing strategy or the second processing strategy is to perform mapping, light flow confidence detection of each reference sub-area of each reference frame is performed, the confidence score of each reference sub-area of each reference frame is analyzed, a direct mapping rationality label is determined, and the third processing strategy of the corresponding sub-area is determined.

2. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The specific analysis method for obtaining each reference sub-area of each reference frame corresponding to the target frame is as follows: Performing blur detection on the reconstructed video sequence, and marking a frame with a blur degree exceeding a blur degree threshold as a target frame; Determining the number of reference frames of the target frame based on the difference between the blur degree and the blur degree threshold; Identifying a target frame and obtaining a set of reference frames corresponding to the target frame, identifying a blur area of the target frame and dividing it into sub-areas, marking the sub-areas of the target frame, thereby obtaining each reference sub-area of each reference frame corresponding to the target frame.

3. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The specific analysis method for analyzing the edge information core indicators of each reference sub-area of each reference frame is as follows: Obtaining edge information of each reference sub-area; Collecting edge information parameters, including edge width, gradient intensity and connected pixel ratio; Analyzing edge information core indicators based on edge information parameters; The edge information core index is a quantitative index of edge information of edge width, gradient intensity and connected pixel proportion of edge information, and the specific analysis process is as follows: the collected gradient intensity and connected pixel proportion are compared with the corresponding reference values respectively, the reference value of the edge width is compared with the edge width, and the comparison results are coupled by combining the corresponding feature allocation multipliers to obtain the edge information core index; Each reference sub-region of each reference frame is traversed to obtain the edge information core index of each reference sub-region of each reference frame.

4. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The enhancement of each enhancement sub-region is performed based on the obtained blur degree of each enhancement sub-region, and the specific analysis method is as follows: When the first processing strategy is to perform sub-region enhancement, the corresponding sub-region is recorded as an enhancement sub-region, and thus each enhancement sub-region is obtained; The blur degree of each enhancement sub-region is obtained; The blur degree of each enhancement sub-region is processed by difference to obtain the blur deviation value of each enhancement sub-region; The blur deviation adjustment sharpening intensity of each enhancement sub-region is determined based on the blur deviation value of each enhancement sub-region; Based on the blur deviation adjustment sharpening intensity of each enhancement sub-region, the sharpening enhancement of each enhancement sub-region is performed.

5. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The second processing strategy of each enhancement sub-region is determined after enhancement, and the specific analysis method is as follows: After enhancement, the blur degree of each enhancement sub-region is detected again, if the blur degree of an enhancement sub-region is greater than or equal to the blur degree threshold, the second processing strategy of the enhancement sub-region is recorded as waiting for mapping; If the blur degree of an enhancement sub-region is less than the blur degree threshold, the second processing strategy of the enhancement sub-region is recorded as generating an enhancement completion signal of each enhancement sub-region.

6. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The direct mapping rationality label is determined, and the specific analysis method is as follows: The optical flow confidence degree of each reference sub-region of each reference frame is detected to obtain the reliability score of the optical flow vector of each reference sub-region of each reference frame; The edge characteristic correction coefficient of each reference sub-region is extracted based on the edge information core index of each reference sub-region of each reference frame; The confidence score of each reference sub-region of each reference frame is obtained based on the reliability score of the optical flow vector of each reference sub-region of each reference frame and the edge characteristic correction coefficient of each reference sub-region; For a sub-region, if the confidence score of a reference frame is greater than or equal to the confidence score threshold, the corresponding sub-region of the reference frame is recorded as a high-confidence region; If the confidence score of a reference frame is less than the confidence score threshold, the corresponding sub-region of the reference frame is recorded as a low-confidence region; Each reference frame is traversed to obtain the number of high-confidence regions; If the number of high-confidence regions is greater than or equal to the number of high-confidence region thresholds, the direct mapping rationality label is recorded as direct mapping is reasonable; If the number of high-confidence regions is less than the number of high-confidence region thresholds, the direct mapping rationality label is recorded as direct mapping is unreasonable.

7. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, The third processing strategy of the corresponding sub-region is determined, and the specific analysis method is as follows: If the direct mapping rationality label is direct mapping is reasonable, the third processing strategy of the corresponding sub-region is direct mapping, and the sub-region corresponding to the reference frame of the highest confidence region is directly mapped to the blurred part of the target frame; If the direct mapping rationality label is direct mapping irrationality, the third processing strategy of the corresponding sub-region is model prediction, the motion trajectory reference length is determined according to the edge information core index, and the second-order motion equation is fitted based on the motion trajectory reference length.

8. The reconstruction method of a digital asset based on AI video dynamic modification according to claim 1, characterized in that, Further comprising the constraint sub-region change rate according to the curvature change rate and the gradient direction deviation, and the specific analysis method is as follows: Traverse each sub-region, if each sub-region exists a completion signal, obtain the overall trajectory of the target frame, and verify and analyze the curvature and gradient direction thereof; If the curvature change rate is greater than or equal to the curvature change rate threshold and the gradient direction deviation is greater than or equal to the gradient direction deviation threshold, it is determined as a structure mutation abnormal point, and a warning information is generated; If the curvature change rate is greater than or equal to the curvature change rate threshold but the gradient direction deviation is less than the gradient direction deviation threshold, it is considered as a natural bending of the edge, and no abnormal residual is calculated; If the curvature change rate is less than the curvature change rate threshold but the gradient direction deviation is greater than or equal to the gradient direction deviation threshold, abnormal residual analysis is performed; If the curvature change rate is less than the curvature change rate threshold and the gradient direction deviation is less than the gradient direction deviation threshold, it is determined as a structure optimization point, and a completion signal is generated; Abnormal residual analysis is performed, specifically: the residual between the mapped edge position and the actual edge position is obtained and recorded as abnormal residual; If the abnormal residual is less than the abnormal residual threshold, the edge position is retained, and an optimization completion signal is generated; If the abnormal residual is greater than or equal to the abnormal residual threshold, a warning information is generated.

9. A reconstruction system of digital assets based on AI video dynamic modification, the reconstruction system of digital assets based on AI video dynamic modification is used to realize the reconstruction method of digital assets based on AI video dynamic modification as claimed in any one of claims 1-8, characterized in that, The system comprises an execution object determination module, an execution strategy determination module, a strategy execution module, and a rationality determination module; The execution object determination module is used to execute the reconstruction process of the digital asset of the AI video dynamic modification, record the digital asset reconstructed as complete as the target digital asset, perform blur analysis on the target digital asset, identify the target frame and obtain the reference frame set corresponding to the target frame, identify the blur area of the target frame and divide it into each sub-region, i.e. each sub-region of the target frame, and thus obtain each reference sub-region of each reference frame corresponding to the target frame; The execution strategy determination module is used to obtain edge information of each sub-region, analyze the edge information core index of each reference sub-region of each reference frame, and thus determine the first processing strategy of the edge information of each sub-region; The strategy execution module is used to record the corresponding sub-region as each enhanced sub-region when the first processing strategy is to execute sub-region enhancement, perform enhancement of each enhanced sub-region based on the obtained blur degree of each enhanced sub-region, and determine the second processing strategy of each enhanced sub-region after enhancement; The rationality determination module is used to perform optical flow confidence detection of each reference sub-region of each reference frame when the first processing strategy or the second processing strategy is to perform mapping, analyze the confidence score of each reference sub-region of each reference frame, determine the direct mapping rationality label, and thus determine the third processing strategy of the corresponding sub-region.

Images