Method for quickly generating image based on AIGC technology

Abstract

The invention is suitable for the field of intelligent image generation, and provides a method for quickly generating an image based on an AIGC technology, which comprises the following steps: picking up feature points, recording the feature points in an analysis module of a system, and recording the coordinates of each feature point according to a picture setting coordinate system; classifying the feature points by using a deep learning model, and distributing corresponding abnormal thresholds; comparing the displacement in the time period, comparing the displacement with an abnormal threshold value, and marking abnormity; establishing a time axis of each feature point, and marking an abnormal interval; performing coincidence query to obtain a non-abnormal interval or a pre-selected processing interval; extracting a static image, and analyzing feature point distribution to obtain a clear static image or a to-be-processed static image; automatically repairing the static image to be processed by using an AIGC technology, and storing the repaired static image as an intermediate processing image; and finally, carrying out definition enhancement and color correction by using an AIGC technology. According to the scheme, efficient and rapid processing of the image is realized.

Description

Technical Field

[0001] The present invention is applicable to the field of intelligent image generation and provides a method for quickly generating images based on AIGC technology. Background Art

[0002] With the rapid development of artificial intelligence technology, AIGC (Artificial Intelligence Generated Content) technology has become an important branch in the field of image processing. AIGC technology automatically or semi-automatically generates image content through algorithm models, which not only greatly enriches the means of image creation, but also improves the efficiency and quality of image generation. In many fields such as film and television production, advertising design, and game development, images generated by AIGC technology have become an indispensable element. It can not only quickly respond to market demand, but also achieve personalized customization to meet users' diverse aesthetic needs.

[0003] The process of generating images using AIGC technology usually includes four steps: data collection, model training, image generation, and post-processing. First, a large amount of image data is collected as training samples, which cover a wide range of image features and styles. Then, a model is built using algorithms such as deep learning, and the collected data is trained to enable the model to learn the laws of image generation. In the image generation stage, the model automatically generates images that meet the requirements based on the input conditions or instructions.

[0004] In the prior art, AIGC technology still has deficiencies in the efficiency and accuracy of image processing. When processing video clips or live photos, existing methods often generate images that have a certain content gap with the original video or live picture, and the process of intercepting and reprocessing the original video is also difficult to quickly and accurately query the optimal image, resulting in poor image extraction and processing effects. Secondly, the prior art lacks pertinence and accuracy when identifying and processing images, resulting in the image generation process that is not only inefficient but also prone to errors. Therefore, there is an urgent need for a method that can overcome the deficiencies of the prior art and achieve efficient and accurate image generation. Summary of the invention

[0005] In view of the above-mentioned defects, the object of the present invention is to provide a method for quickly generating an image based on AIGC technology, the purpose of which is to solve the problems raised in the background technology, comprising the following steps:

[0006] S1. Generate a static image based on a video clip or a live image;

[0007] The specific steps are:

[0008] S1.1. Picking feature points of live images; recording the picked feature points in a computer analysis module inside the system, which sets a coordinate system according to the frame and records the coordinates of each feature point on the coordinate system;

[0009] S1.2. Use the deep learning model to classify feature points, and assign corresponding abnormal thresholds to different types of feature points;

[0010] S1.3. Compare the displacement of the feature point within the set time period of the live image, compare the displacement change value of the feature point with the abnormal threshold of the classification, and mark the abnormality;

[0011] S1.4. Establish a time axis corresponding to the total video or live time for each feature point, and mark the abnormal interval of the feature point on the time axis of each feature point according to the abnormal displacement;

[0012] S1.5. Perform a coincidence query on the time axis of each feature point. If there is a common non-abnormal interval on the time axis of each feature point, then take this interval as the non-abnormal interval; if there is no common non-abnormal interval on the time axis of each feature point, then take the abnormal interval with the least number of non-overlapping or overlapping feature points as the pre-selected processing interval;

[0013] S1.6. Extract a static image from a non-abnormal interval or a pre-selected processing interval, and analyze the distribution of feature points; output one or more static images with the most reasonable distribution of feature points to obtain a clear static image or a static image to be processed;

[0014] S1.7. If the result is a clear static image, the clear static image is directly stored as an intermediate processed image; if the result is a static image to be processed, the AIGC technology is used to automatically repair the static image to be processed, and the repaired image is stored as an intermediate processed image;

[0015] S2. Use AIGC technology to enhance clarity;

[0016] S3. Color correction through AIGC technology.

[0017] Furthermore, step S1.1 includes the following steps:

[0018] S1.1.1. Set the upper limit of the number of feature points in the plane space range. When the number of picked feature points exceeds the upper limit, the feature points are excluded according to the rules;

[0019] S1.1.2. Set the picking range of feature points and restrict the area of ​​feature points;

[0020] S1.1.3. Calculate the stability index of each feature point and select the feature point with higher stability;

[0021] S1.1.4. Based on the machine learning method, a certain number of highly homogeneous feature points are excluded.

[0022] Further, step S1.3 includes the following steps:

[0023] S1.3.1. Generate feature point trajectory; Use optical flow method to track the trajectory of feature points in the entire video or live photo sequence, and generate the feature point trajectory by calculating the displacement vector of each feature point from the initial frame to the subsequent frame;

[0024] S1.3.2. Traverse each feature point trajectory and calculate its total displacement within the set time period, and at the same time count the displacement distribution and speed change information of the feature points;

[0025] S1.3.3. Detect abnormal displacement of feature points by statistical methods.

[0026] Further, step S1.6 includes the following steps:

[0027] S1.6.1 Extract static images from non-abnormal intervals or pre-selected processing intervals, integrate the feature points in all static images according to their classification labels, divide them into a number of feature point elements, and form a feature point element database; for each image, record the number, type and position of its feature point elements in the image;

[0028] S1.6.2 Analyze the composition of each image using a fully trained AIGC technology module to obtain an evaluation result with a score or ranking as the evaluation criteria;

[0029] S1.6.3 Based on the score or ranking of the composition analysis, select the one with the highest composition evaluation from all the still images as the clear still image or the still image to be processed.

[0030] Furthermore, the processing method of the static image to be processed in step S1.7 is specifically as follows:

[0031] Extract the fuzzy feature points in the static image to be processed, and extract the range of the fuzzy feature points to obtain a fuzzy image block; use the original image block of the non-abnormal interval on the time axis corresponding to the feature points in the fuzzy feature block, and generate a replacement image block after adaptation processing to replace the fuzzy feature block.

[0032] Furthermore, the processing of the static image to be processed includes the following steps:

[0033] S1.7.1. The system extracts the time axis data of the feature point in the fuzzy feature block area and queries the non-abnormal interval on the time axis; if there is a non-abnormal interval, the image of the feature point in the non-abnormal interval is extracted as the original image block;

[0034] S1.7.2. When the system queries the time axis data of the feature point and finds that there is no non-abnormal interval on the time axis or the system chooses not to use the original image block obtained from the non-abnormal interval, the AIGC technology is used to automatically repair the blurred feature block;

[0035] S1.7.3. Use AIGC technology to optimize the size, shape and color of the fuzzy feature blocks when matching the original image blocks;

[0036] S1.7.4. Store the repaired static image to be processed as an intermediate processing image.

[0037] Furthermore, when the original image block is obtained using the non-abnormal interval, the time interval between the original image block and the interval where the blurred point is located is evaluated, and a computer tool is used to calculate the similarity measure between the area around the blurred point and the original image block in the non-abnormal interval. If the similarity measure result or the time interval is greater than a threshold, the original image block is reselected or the original image block obtained using the non-abnormal interval is not used.

[0038] Beneficial effects:

[0039] 1. Through scientific module division and database design, efficient and fast image processing is achieved. It can quickly and accurately query the best image in the video clip or live photo, extract it and process the image quickly and intelligently. Compared with the existing technology, this method has significantly improved the processing speed and accuracy.

[0040] 2. Through feature point detection, classification, displacement comparison and other steps, the key areas in the image can be accurately identified and processed, effectively reducing the workload of image processing using AIGC technology and greatly improving the pertinence and accuracy of processing.

[0041] 3. By using feature point analysis and processing and deep learning technology, the composition of image elements can be analyzed, and the best image can be selected for output through evaluation, which can greatly improve the image extraction effect and ensure high-quality image selection.

[0042] 4. This method uses a variety of advanced algorithms and technologies (such as SIFT, SURF, ORB feature point detection algorithms, deep learning models, AIGC technology, etc.), and is highly robust to image rotation, scale changes, illumination changes, etc.

[0043] 5. This method is not only suitable for the processing of video clips and live photos, but can also be extended to other image processing fields, such as image restoration, image synthesis, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a step diagram of the method;

[0045] Figure 2 A diagram showing the steps for automatically repairing a static image to be processed. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] See also Figure 1 , 2 The object of the present invention is to provide a method for quickly generating an image based on AIGC technology, comprising the following steps:

[0049] S1. Generate a static image based on a video clip (a video clip extending from a specified time point to both sides of the time) or a live image. Depending on the processing process, the static image includes a clear static image or a static image to be processed.

[0050] This step uses feature point detection algorithms in computer vision technology (such as SIFT, SURF, ORB, etc.) to pick up key feature points in the live image. SIFT (Scale-Invariant Feature Transform) feature, that is, scale-invariant feature transform, is a feature extraction algorithm in computer vision, used to detect and describe local features in images. SURF (Speeded Up Robust Features) is a robust image recognition and description algorithm. It is an efficient variant of SIFT and also extracts scale-invariant features. The algorithm steps are roughly the same as the SIFT algorithm, but the method used is different and it is more efficient than the SIFT algorithm. ORB (Oriented FAST and Rotated BRIEF) This feature detection algorithm is proposed based on the famous FAST feature detection and BRIEF feature descriptor. Its running time is much better than SIFT and SURF, and it can be applied to real-time feature detection.

[0051] The feature points obtained by computer tools in this step are obvious and stable points in the image, such as corner points, edge points, etc. Feature points are searched and marked in each frame of the image, and the feature points are recorded in the computer analysis module. The computer analysis module detects the displacement of the feature points and records the abnormalities, and then extracts them according to the non-abnormal segments or abnormal segments with low overlap on the corresponding time axis, and obtains clear static images or static images to be processed respectively. The existing technology often directly processes the image as a whole, which lacks specificity. Through feature point analysis, this method can more accurately identify and process key areas in the image.

[0052] The specific steps of this step are as follows:

[0053] S1.1. Picking up feature points of live images. The picked feature points are recorded in the computer analysis module inside the system. The computer analysis module sets up a coordinate system according to the frame and records the coordinates of each feature point on the coordinate system.

[0054] The establishment of the coordinate system facilitates the displacement analysis of feature points in subsequent steps. The actual displacement analysis only requires a few feature points. When the image to be processed is complex and the number of feature points picked is too large, it is necessary to exclude too many homogeneous feature points to avoid complex calculation processing and too many feature points.

[0055] Therefore, step S1.1 includes the optimization and elimination of feature points, specifically:

[0056] S1.1.1. Set an upper limit on the number of feature points in the plane space range according to actual needs. When the number of picked feature points exceeds the upper limit, select feature points for exclusion according to the rules (such as stability sorting from high to low, and the surrounding area with the highest density). This step can effectively reduce homogeneity and redundant feature points, reduce computational complexity, and improve processing efficiency.

[0057] S1.1.2. Perform regional restrictions on feature points according to actual needs; set the picking range of feature points and limit the picking range of feature points. The system only picks up feature points in key areas.

[0058] S1.1.3. Calculate the stability index of each feature point (such as displacement standard deviation, velocity change rate, etc.) and select the feature points with higher stability.

[0059] Feature points with high stability refer to feature points that have small position changes in consecutive frames, are not easily affected by noise or changes in illumination, and can continue to exist stably in the image. This step is to eliminate feature points that flicker or exist for less than a set period of time. The remaining feature points have strong robustness and reliability and can provide accurate and consistent information in subsequent image processing or analysis.

[0060] S1.1.4. Based on machine learning methods, use classifiers (such as support vector machines, decision trees, etc.) to classify feature points, distinguish useful feature points from useless feature points, exclude a certain number of highly homogeneous feature points, and only retain a few "typical feature points".

[0061] The above steps for optimizing and eliminating feature points and the training direction of machine learning must follow the following principles:

[0062] 1. Feature points should remain stable in consecutive frames and should not disappear and reappear frequently;

[0063] 2. Feature points should have unique appearance or characteristics so that they can be easily identified and matched in subsequent processing;

[0064] 3. Feature points should be evenly distributed throughout the image or target area to avoid certain areas being too dense or sparse.

[0065] S1.2. Use a deep learning model (such as a convolutional neural network) to classify feature points. By training the deep learning model, it is possible to identify different objects and parts (such as hands, feet, body, objects, etc.) at the feature points in the image and classify each feature point into the corresponding category. For example, if the feature point is located on a person's hand, the feature point of the hand is classified as a hand. Feature points of different categories have different abnormal thresholds when comparing the displacement in the subsequent step.

[0066] The displacement abnormality threshold of hand feature points is higher than that of body feature points, because hands usually have larger displacement in videos or live photos. The classification methods in the existing technology often rely on artificial features and rules, and the classification accuracy is low. By using feature points as an intermediate medium to identify and extract the position of each feature point, the deep learning model has higher classification accuracy and generalization ability than traditional classification methods.

[0067] S1.3. Compare the displacement of the feature point within the set time period of the live image, compare the displacement change (coordinate change) value of the feature point with the abnormal threshold of the classification, and mark the abnormality. The computer analysis module will record the initial coordinates of each feature point and the coordinates in the subsequent frames, and calculate the displacement between them; then compare these displacements with the corresponding abnormal threshold. If the displacement exceeds the threshold, the abnormal time period will be marked as abnormal displacement.

[0068] Among them, the specific steps of this step are:

[0069] S1.3.1. Generate feature point trajectories: Use optical flow methods (such as Lucas-Kanade method or Farneback method) to track the trajectory of feature points in the entire video or live photo sequence. The feature point trajectory is generated by calculating the displacement vector of each feature point from the initial frame to the subsequent frame. The feature point trajectory is used to analyze the motion pattern and displacement of the feature point.

[0070] S1.3.2. Displacement calculation and statistics: traverse the trajectory of each feature point and calculate its total displacement within the set time period. At the same time, statistical information such as the displacement distribution and speed change of the feature point is calculated.

[0071] S1.3.3. Detect abnormal displacement of feature points through statistical methods (such as threshold detection).

[0072] S1.4. For each feature point, a time axis corresponding to the total video or live time is established, and the abnormal interval of the feature point is marked on the time axis of each feature point according to the abnormal displacement (abnormal displacement). That is, the computer analysis module creates a time axis for each feature point and marks the abnormal interval on the time axis. The abnormal interval means that the displacement of the feature point exceeds the set threshold within this time period.

[0073] The computer analysis module is internally provided with a database corresponding to each feature point to store the timeline and abnormal information of each feature point. The database contains data such as feature point ID, timeline start point, timeline end point, abnormal time list, etc. The timeline and abnormal information of each feature point are stored in the database for easy subsequent query and analysis. Among them, the time unit on the timeline is set according to the time resolution and duration of the video or live photo. For example, if the video resolution is 24 frames per second, the time unit can be set to seconds or frames.

[0074] Establishing a timeline for feature points can intuitively record the change process of feature points and improve the accuracy of subsequent query and analysis.

[0075] S1.5. Use the overlapping recognition method to identify non-abnormal intervals or pre-selected processing intervals. Perform an overlap query on the time axis of each feature point. If there is a common non-abnormal interval on the time axis of each feature point, then take this interval as the non-abnormal interval; if there is no common non-abnormal interval on the time axis of each feature point, then take the abnormal interval with the least number of non-overlapping or overlapping feature points as the pre-selected processing interval.

[0076] Since the resolution of the time axis (i.e. the size of the minimum time interval of the unit) will also affect the accuracy of the detection results, the analysis module will first set the appropriate time resolution based on the length of the video or the frame rate of the live photo and the movement speed of the feature points. At the same time, the time axis resolution can also be set by the technician based on experience.

[0077] S1.6. Extract static images from non-abnormal intervals or pre-selected processing intervals, and analyze the distribution of feature points; the analysis module extracts each frame of static images within the non-abnormal interval, non-overlapping abnormal interval, or abnormal interval with the least number of overlapping feature points, and uses the deep learning model in step S1.2 to classify the image feature points (such as head, hands, and other parts of the body), performs feature point distribution composition analysis on each element in the image, and outputs one or more static images with the most reasonable feature point distribution to obtain a clear static image or a static image to be processed.

[0078] The specific steps are as follows:

[0079] S1.6.1 Extract static images from non-abnormal intervals or pre-selected processing intervals, integrate the feature points in all static images according to their classification labels, divide them into several feature point elements, and form a feature point element database. For each image, record the number, type and position of its feature point elements in the picture.

[0080] S1.6.2 Utilizes a fully trained AIGC technology module to analyze the composition of each image, involving evaluation of multiple aspects such as spatial distribution of feature points, symmetry, balance, visual guides, and focal points, and obtains evaluation results such as scoring or ranking as evaluation criteria to measure the composition quality of the image.

[0081] S1.6.3 Based on the score or ranking of the composition analysis, select the one with the most perfect composition from all the still images as the clear still image or the still image to be processed.

[0082] In the actual processing process, if the time axis overlap query of each feature point is performed to obtain a non-abnormal interval / pre-selected processing interval, a static image is extracted from the non-abnormal interval / pre-selected processing interval and a feature point analysis is performed. At this time, a non-abnormal interval / pre-selected processing interval is selected from the video clip, and all static images in the interval are extracted. A deep learning model is used to perform a composition analysis of the feature point elements on each image (or each image at a specified unit time interval in the non-abnormal interval). The composition analysis includes the symmetry, balance, and visual guide lines (such as the sight line of a person, the direction of action) of multiple feature point elements on the image, and a score or ranking is generated for each image as an evaluation criterion for measuring the composition quality. According to the score or ranking of the composition analysis, the one with the most perfect composition is selected from all static images. This image may have the best visual balance, the most attractive focus, and the clearest visual guide line, and it is used as a clear static image or a static image to be processed.

[0083] S1.7. If the result is a clear static image, the clear static image is directly stored as an intermediate processed image; if it is a static image to be processed, the AIGC technology is used to automatically repair the static image to be processed, and the repaired image is stored as an intermediate processed image.

[0084] Specifically, when the result is a static image to be processed, the fuzzy feature points in the static image to be processed are extracted, and the range of the fuzzy feature points is extracted to obtain a fuzzy image block. The extraction of the fuzzy image block is used to provide necessary data support for subsequent targeted processing and optimization. The original image block of the non-abnormal interval on the time axis corresponding to the feature point in the fuzzy feature block is used to generate a replacement image block after adaptation processing to replace the fuzzy feature block (fuzzy part); the specific steps are as follows:

[0085] S1.7.1. The system extracts the time axis data of the feature point in the fuzzy feature block area and queries the non-abnormal interval on the time axis; if there is a non-abnormal interval, the image of the feature point in the non-abnormal interval (i.e., the image block within a certain range around the feature point) is extracted as the original image block to replace the fuzzy feature block;

[0086] In other words, when there is an image in which the feature point is not blurred in the time axis, the blurred part on the static image to be processed is replaced by the original image block obtained from the part (ie, the non-abnormal interval).

[0087] When the original image block is obtained using the non-abnormal interval, the time interval between the original image block and the interval where the blur point is located is evaluated. At the same time, tools such as SSIM and PSNR are used to calculate the similarity measurement between the area around the blur point and the original image block in the non-abnormal interval, or to determine whether the time interval is appropriate based on the set threshold. If the similarity measurement result or the time interval is greater than the threshold, the picture fragment is reselected or the original image block obtained without using the non-abnormal interval is not used.

[0088] S1.7.2. When the system queries the time axis data of the feature point and finds that there is no non-abnormal interval on the time axis or the system chooses not to use the original image block obtained from the non-abnormal interval, it uses an external reference image (the deep learning system uses the existing network image) for repair. Specifically, the system uses AIGC technology to automatically repair the blurred feature block.

[0089] S1.7.3. When using the original image block obtained from the non-abnormal interval, the original image block is optimized; the original image block is micro-processed using AIGC technology; the original image block obtained from the non-abnormal interval is micro-processed using AIGC technology (such as intelligent sharpening, color correction, texture synthesis, etc.) so that the processed image block better matches the size, shape and color of the blurred feature block.

[0090] S1.7.4. Store the repaired static image to be processed as an intermediate processing image.

[0091] S2. Use AIGC technology to enhance clarity; use image enhancement or super-resolution reconstruction technology in AIGC technology to improve the resolution of the image. The principle of super-resolution reconstruction technology is to generate a higher-resolution image by learning the relationship between high-resolution images and low-resolution images, and to process the intermediate processed images to obtain a clearer image. Use image enhancement algorithms (such as contrast enhancement, sharpening, etc.) to improve the visual effect of the image, and improve the clarity and details of the image by adjusting the pixel value of the image.

[0092] S3. Perform color correction through AIGC technology to adjust the color balance and saturation of the image to make the image look more natural and vivid.

[0093] This image processing method combines computer vision technology, AIGC technology and deep learning technology to extract the clearest static images through steps such as feature point detection, classification, displacement comparison, image block matching, rapid processing of blurred points and overall image processing. After obtaining these images, AIGC technology is used for further image enhancement, super-resolution reconstruction, denoising and color correction to improve the quality and clarity of the image. This method is not only suitable for the processing of video clips and live photos, but can also be extended to other image processing fields, such as image restoration and image synthesis.

[0094] In response to the above method, this technical solution provides a processing system based on AIGC data that can use the above method to process images and quickly generate images. The system aims to extract, select, and process the clearest static images from video clips or live photos, and perform further image enhancement, color correction and other processing to improve the quality and clarity of the image. The system includes: feature point detection module, feature point database, classification module, displacement calculation module, anomaly detection module, anomaly information database, timeline construction module, timeline database, interval recognition module, static image extraction module, static image database, processing result database, composition analysis module, fuzzy feature block processing module, image clarity enhancement module and color correction module;

[0095] The feature point detection module can use feature point detection algorithms (such as SIFT, SURF, ORB, etc.) to pick up key feature points in live images and store feature point data in each image, including the coordinates, types, stability indicators and other information of the feature points in the feature point database.

[0096] The classification module uses deep learning models (such as convolutional neural networks) to classify feature points, identify different objects and parts (such as hands, feet, body, objects, etc.), and simultaneously enter the classification information into the feature point database to match it with the corresponding feature points.

[0097] The displacement calculation module compares the displacement of the feature points within the set time period of the live image. Then the anomaly detection module compares the displacement change value with the set anomaly threshold, marks the anomaly and builds an anomaly information database. The anomaly information database is used to store the time axis data and anomaly information of each feature point, including the time axis start point, end point, anomaly time list, etc., to facilitate subsequent query and analysis. At the same time, the displacement calculation module can also generate feature point trajectories, calculate and count the statistical information such as the displacement distribution and speed change of the feature points.

[0098] The timeline construction module establishes a data structure corresponding to the timeline for each feature point, constructs a timeline database, and marks abnormal intervals on the timeline. The interval recognition module uses the overlapping recognition method to identify non-abnormal intervals or pre-selected processing intervals, providing a basis for subsequent static image extraction.

[0099] The static image extraction module extracts static images from non-abnormal intervals or pre-selected processing intervals and builds a static image database. The composition analysis module uses a deep learning model to perform composition analysis on image feature points, evaluates the image's symmetry, balance, visual guide lines and other composition qualities, selects images with the most perfect composition, and builds a processing result database to store images.

[0100] The fuzzy feature block processing module processes the fuzzy feature blocks in the static image to be processed, and replaces them with the original image blocks in the non-abnormal interval, or uses AIGC technology for micro-processing. If there is no non-abnormal interval, an external reference image is used for repair.

[0101] The image clarity enhancement module and color correction module use the image enhancement or super-resolution reconstruction technology in AIGC technology to improve the resolution and clarity of the image, improve the visual effect of the image, adjust the color balance and saturation of the image, and make the image look more natural and vivid.

[0102] Therefore, this method can quickly generate an image processing system based on AIGC data, and achieve efficient and fast image processing through scientific module division and database design.

[0103] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, technicians familiar with the field may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for quickly generating images based on AIGC technology, characterized in that Follow these steps: S1. Generate a static image based on a video clip or a live image; the specific steps are: S1.

1. Pick up feature points of live images; record the picked feature points in the computer analysis module inside the system, the computer analysis module sets the coordinate system according to the frame, and records the coordinates of each feature point on the coordinate system; S1.

2. Use the deep learning model to classify the feature points, and assign corresponding abnormal thresholds to different types of feature points; S1.

3. Compare the displacement of the feature point within the set time period of the live image, compare the displacement change value of the feature point with the abnormal threshold of the classification, and mark the abnormality; S1.

4. Establish a time axis corresponding to the total video or live time for each feature point, and mark the abnormal interval of the feature point on the time axis of each feature point according to the abnormal displacement; S1.

5. Perform a coincidence query on the time axis of each feature point. If there is a common non-abnormal interval on the time axis of each feature point, then take this interval as the non-abnormal interval; if there is no common non-abnormal interval on the time axis of each feature point, then take the abnormal interval with the least number of non-overlapping or overlapping feature points as the pre-selected processing interval; S1.

6. Extract a static image from a non-abnormal interval or a pre-selected processing interval, and analyze the distribution of feature points; output one or more static images with the most reasonable distribution of feature points to obtain a clear static image or a static image to be processed; S1.

7. If the result is a clear static image, the clear static image is directly stored as an intermediate processed image; if the result is a static image to be processed, the AIGC technology is used to automatically repair the static image to be processed, and the repaired image is stored as an intermediate processed image; S2. Use AIGC technology to enhance clarity; S3. Color correction through AIGC technology.

2. The method for rapidly generating images based on AIGC technology according to claim 1, characterized in that: The step S1.1 comprises the following steps: S1.1.

1. Set the upper limit of the number of feature points in the plane space range. When the number of picked feature points exceeds the upper limit, the feature points will be excluded according to the rules; S1.1.

2. Set the picking range of feature points and restrict the area of ​​feature points; S1.1.

3. Calculate the stability index of each feature point and select the feature point with higher stability; S1.1.

4. Based on the machine learning method, a certain number of highly homogeneous feature points are excluded.

3. The method for rapidly generating images based on AIGC technology according to claim 1, characterized in that: The step S1.3 comprises the following steps: S1.3.

1. Generate feature point trajectory; Use optical flow method to track the trajectory of feature points in the entire video or live photo sequence, and generate the feature point trajectory by calculating the displacement vector of each feature point from the initial frame to the subsequent frame; S1.3.

2. Traverse each feature point trajectory and calculate its total displacement within the set time period, and at the same time count the displacement distribution and speed change information of the feature points; S1.3.

3. Detect abnormal displacement of feature points by statistical methods.

4. The method for rapidly generating images based on AIGC technology according to claim 1, characterized in that: The step S1.6 comprises the following steps: S1.6.1 Extract static images from non-abnormal intervals or pre-selected processing intervals, integrate the feature points in all static images according to their classification labels, divide them into a number of feature point elements, and form a feature point element database; for each image, record the number, type and position of its feature point elements in the picture; S1.6.2 Use a fully trained AIGC technology module to analyze the composition of each image and obtain an evaluation result with a score or ranking as the evaluation standard; S1.6.3 Based on the score or ranking of the composition analysis, select the one with the highest composition evaluation from all the still images as the clear still image or the still image to be processed.

5. The method for rapidly generating images based on AIGC technology according to claim 1, characterized in that: The specific processing method of the static image to be processed in step S1.7 is: Extract the fuzzy feature points in the static image to be processed, and extract the range of the fuzzy feature points to obtain a fuzzy image block; use the original image block of the non-abnormal interval on the time axis corresponding to the feature points in the fuzzy feature block, and generate a replacement image block after adaptation processing to replace the fuzzy feature block.

6. The method for rapidly generating images based on AIGC technology according to claim 5, characterized in that: The processing of the static image to be processed comprises the following steps: S1.7.

1. The system extracts the time axis data of the feature point in the fuzzy feature block area and queries the non-abnormal interval on the time axis; if there is a non-abnormal interval, the image of the feature point in the non-abnormal interval is extracted as the original image block; S1.7.

2. When the system queries the time axis data of the feature point and finds that there is no non-abnormal interval on the time axis or the system chooses not to use the original image block obtained from the non-abnormal interval, the AIGC technology is used to automatically repair the blurred feature block; S1.7.

3. Use AIGC technology to optimize the size, shape and color of the fuzzy feature blocks when matching the original image blocks; S1.7.

4. Store the repaired static image to be processed as an intermediate processing image.

7. The method for rapidly generating images based on AIGC technology according to claim 6, characterized in that: When the original image block is obtained using the non-abnormal interval, the time interval between the original image block and the interval where the blurred point is located is evaluated, and a computer tool is used to calculate the similarity measure between the area around the blurred point and the original image block in the non-abnormal interval. If the similarity measure result or the time interval is greater than a threshold, the original image block is reselected or the original image block obtained without using the non-abnormal interval is not used.

Images