Image quality jitter detection method, electronic device, storage medium and computer program product
By training a target model based on target value and grayscale projection algorithm, edge information in images is eliminated, solving the accuracy problem of image quality jitter detection in complex environments, achieving efficient image quality jitter detection, and improving user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2026-03-27
AI Technical Summary
Existing image jitter detection technologies perform poorly when dealing with non-rigid deformations or complex motion patterns, and are easily affected by noise and image features, resulting in inaccurate detection results.
A target model based on target value and grayscale projection algorithm is adopted, which is trained by training samples. This eliminates the mean of image edge information, reduces the impact of noise and complex motion patterns on detection, and improves detection accuracy.
It improves the accuracy of image quality jitter detection, can intelligently identify image quality problems, and enhances the user's viewing experience.
Smart Images
Figure CN121750852A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and in particular to a method for detecting image jitter, an electronic device, a storage medium, and a computer program product. Background Technology
[0002] The development of Internet Protocol Television (IPTV) and Over-The-Top (OTT) technologies has enabled people to watch TV programs and movies anytime, anywhere via the Internet, no longer restricted by traditional television broadcast times and channels.
[0003] However, if users encounter image quality issues while watching videos, and the operator is unaware of the problem and does not take corrective action, this will negatively impact the user's viewing experience. Therefore, a new image quality jitter detection technology has been proposed, using a grayscale projection algorithm for image quality detection. However, the grayscale projection algorithm simply reports image quality anomalies in its detection process, and during detection, it may be affected by noise and other image features, leading to inaccurate results, especially when dealing with non-rigid deformations or complex motion patterns. Summary of the Invention
[0004] This application provides a method for detecting image quality jitter, an electronic device, a storage medium, and a computer program product to improve the accuracy of image quality jitter detection.
[0005] In a first aspect, a method for detecting image quality jitter is provided, comprising: extracting keyframe images from a played video at a predetermined frequency; inputting the keyframe images into a target model for detection, and obtaining a detection result; wherein the target model is used to detect image quality jitter in the input images, the target model is trained using a target value and a grayscale projection algorithm through at least one set of training samples, the target value is the mean value used to eliminate edge information of the image, and the detection result indicates whether image quality jitter exists in the keyframe images.
[0006] Secondly, a method for obtaining a model for image quality jitter detection is provided, comprising: training a target model using at least one set of training samples based on a target value and a grayscale projection algorithm until the training termination condition is met, thereby obtaining the target model, wherein the target model is used to perform image quality jitter detection on the input image, and the target value is the mean value used to eliminate edge information of the image.
[0007] Thirdly, an electronic device is provided, the electronic device including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions being executed by the processor to implement the steps of the method described in the first aspect above, or to implement the steps of the method described in the second aspect above.
[0008] Fourthly, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect or the steps of the method described in the second aspect.
[0009] Fifthly, a computer program product is provided, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to perform the steps of the method described in the first aspect, or to perform the steps of the method described in the second aspect.
[0010] In this embodiment, keyframe images are extracted from the played video at a predetermined frequency, and then the keyframe images are input into a target model for detection to obtain detection results. The target model is trained using at least one set of training samples based on a target value and a grayscale projection algorithm. The target value is the mean used to eliminate edge information in the image, thereby avoiding large differences in the projected waveform at the edges, reducing the impact of image noise or complex motion patterns on image quality jitter detection, and improving the accuracy of image detection.
[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description
[0012] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0013] Figure 1 A flowchart illustrating a method for obtaining a model for image quality jitter detection provided in an exemplary embodiment of this application is shown. Figure 2 A flowchart illustrating a model acquisition method for image quality jitter detection provided in another exemplary embodiment of this application is shown. Figure 3 The illustration shows a flowchart of an exemplary embodiment of the present application providing a method for detecting image quality jitter; Figure 4 A flowchart illustrating another exemplary embodiment of this application provides a method for detecting image quality jitter; Figure 5 This is a structural block diagram of an electronic device according to an exemplary embodiment. Detailed Implementation
[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0015] With the advent of the artificial intelligence (AI) era, the rapid development of technology is profoundly changing the way we live and work. Against this backdrop, the application of big data and technologies such as IPTV and OTT is becoming increasingly widespread. These technologies not only provide people with more convenient and personalized services, but also bring enormous commercial value to operators and equipment manufacturers.
[0016] Large models are a natural language processing technology based on deep learning. They can understand and generate human language, enabling various functions such as intelligent dialogue, automatic translation, and text generation. In the IPTV and OTT fields, large models can be used for image quality detection, content recommendation, advertising, and user behavior analysis, helping operators better understand user needs and provide more accurate and high-quality services. This application provides an optimized detection algorithm based on large models to achieve real-time monitoring of image quality jitter. The technical solution provided in this application embodiment is described below with reference to the accompanying drawings.
[0017] Figure 1 This illustration shows a flowchart of a method for obtaining a model for image quality jitter detection according to an exemplary embodiment of this application. This method can be executed by an electronic device, which can be a server or a terminal. For example, the method can be executed by a server. After obtaining a target model for image quality jitter detection using this method, the server sends the target model to the terminal. When the terminal plays a video, it activates the target model to perform image quality jitter detection on the played video. Figure 1 As shown, the method mainly includes the following steps: S110, Obtain at least one set of training samples.
[0018] S112, Based on the target value and grayscale projection algorithm, the target model is trained using at least one set of training samples until the training termination condition is met, and the target model is obtained. The target model is used to perform image quality jitter detection on the input image, and the target value is the mean value used to eliminate image edge information.
[0019] Assuming the entire image is moved upwards by dh, the algorithm used in related technologies is as follows: (1) in, and These represent the row grayscale projections of the image before and after the displacement, respectively; m is the search range of this algorithm for the offset. This indicates that the hm values of the two images are in... The sum of squares of the differences in the grayscale projections of the rows; if there is a minimum value Make If the minimum value is found, it indicates that the projections between rows of the two images are relatively similar, and therefore the image displacement in the h direction is considered to be... However, when the image displacement is large, the edge information of each image is different, resulting in significant differences in the projected waveform at the edges. This leads to errors in the calculation of cross-correlation peaks, resulting in a large detection error. In the technical solution provided in this application, the target model is trained based on the mean used to eliminate image edge information. This avoids the problem of large differences in the projected waveform at the edges, reduces the impact of image noise or complex motion patterns on image quality jitter detection, and improves the accuracy of the target model in detecting image jitter.
[0020] In some embodiments, training the target model using at least one set of training samples based on the target value and grayscale projection algorithm may include the following steps: Step 1: Obtain the feature data corresponding to each training sample; In this embodiment of the application, a training sample may include a sample image and the corresponding label data of the sample image, and the label data may indicate whether the sample image has jitter.
[0021] In step 1, each training sample can be characterized. For example, feature data can be extracted from the sample image, which includes, but is not limited to, grayscale projection data.
[0022] Optionally, this feature data can be obtained from pixel sampling of the sample image. Alternatively, the entire pixel count of the sample image can be omitted. For example, approximately M% of the sample image's center and N% of each of the four edges can be sampled, resulting in a pixel count of the sample image (M+4N≤100, M>N). This approach allows for focusing on the core content of the image while reducing the computational load on large models.
[0023] Step 2: Based on each feature data, the weight corresponding to each feature data, and the model correction parameters, the above target value is obtained, wherein the model correction parameters are used to correct the target model; In these optional embodiments, weights can be assigned to each feature data point. These weights can be determined based on the training samples corresponding to each feature data point, and the weights represent the importance of each training sample. Model calibration parameters can be used to calibrate the target model. A target value is obtained using the feature data, weights, and model calibration parameters, and this target value can be adjusted during subsequent training of the target model by adjusting the model calibration parameters.
[0024] In some optional implementations, the model correction parameters may include multiple first correction parameters and one second correction parameter, and the target value can be obtained in step 2 by following these steps: Step 21: Calculate the average of the target differences of multiple feature data, wherein the target difference is the difference between the first value and the first correction parameter corresponding to the feature data, and the first value is the product of the feature data and the weight corresponding to the feature data; Step 22: Calculate the difference between the average of the above target differences and the second correction parameter to obtain the target value.
[0025] For example, the target value can be calculated using the following formula. : (2) in, For the feature data of the (i+1)th training sample, The weights corresponding to the (i+1)th feature data are... The first correction parameter is the one corresponding to the (i+1)th feature data. The second correction parameter is , and the total number of training samples is m+1.
[0026] Using the above technical solution, with a sufficient number of training samples, the following can be calculated: It can eliminate the effects of image edges.
[0027] Step 3: Based on the target value and the grayscale projection algorithm, train the target model using at least one set of training samples until the training termination condition is met, and obtain the target model.
[0028] In this embodiment of the application, the training termination condition may include at least one of the following: the detection results of N consecutive training samples are consistent with the actual results of the training samples, or a preset number of training rounds has been reached.
[0029] In some embodiments, the process of training the target model using at least one set of training samples based on the target value and grayscale projection algorithm may include the following steps: Step 31: Calculate the grayscale projection of each training sample in the current group before and after displacement using the target model. Step 32: Based on the sum of the squares of the first differences of all rows within the search range of the offset, obtain the displacement corresponding to the training sample, wherein the first difference is the difference between the second difference and the target value mentioned above, and the second difference is the difference between the grayscale projection of the row after the displacement and the grayscale projection of the row before the displacement of the training sample. Step 33: Based on the relationship between the displacement corresponding to each training sample in the current group and the target threshold corresponding to each training sample, obtain the detection result of each training sample; In this embodiment of the application, since the criteria for judging whether different sample images have jitter may be different, different training samples may correspond to different thresholds. The target threshold corresponding to each training sample can be stored in the feature vector data corresponding to the training sample. The target thresholds stored in different feature vector data may be different.
[0030] In this embodiment of the application, if the displacement corresponding to a certain training sample is greater than or equal to the target threshold corresponding to the training sample, the detection result of the training sample is that there is image quality jitter; if the displacement corresponding to a certain training sample is less than the target threshold corresponding to the training sample, the detection result of the training sample is that there is no image quality jitter.
[0031] For example, the target model can use the following algorithm to calculate the displacement corresponding to the training samples: (3) in, and These represent the row grayscale projections of the image before and after the displacement, respectively; m is the search range of this algorithm for the offset. This indicates that the hm values of the two images are in... The sum of squares of the differences in the grayscale projections of the rows. If the above target value has a minimum value... Make If the minimum value is found, it indicates that the projections between rows of the two images are relatively similar, and therefore the image displacement in the h direction is considered to be... ,if If the value is greater than or equal to the target threshold corresponding to the training sample, then the training sample is determined to have image quality jitter, and furthermore, it can be determined based on... The difference between the training sample's score and the target threshold is used to score the degree of jitter in the training sample. Compared to algorithms in related technologies, this method can reduce the impact of image noise or complex motion patterns on image jitter detection, thereby improving the recognition rate of image detection.
[0032] Step 34: In response to the fact that all the detection results of the current group are consistent with the actual results corresponding to the training samples, the target model is trained again using the next group of training samples until the training termination condition is met. The training termination condition includes: the detection results of N consecutive groups of training samples are consistent with the actual results of the training samples.
[0033] In some embodiments, in step 34 above, if at least one detection result in the current group is inconsistent with the actual result corresponding to the training sample, it indicates that the model correction parameters are not suitable. The model correction parameters can be adjusted, and based on the adjusted model correction parameters, the target value is recalculated. Using the recalculated target value, the process jumps to step 31 above. Based on the target value, the process calculates the row grayscale projection of each training sample in the current group before and after the displacement. The current group training samples are used again for training until all detection results in the current group are the same as the actual results.
[0034] In some embodiments, if the detection result is consistent with the actual result corresponding to the training sample and the detection result indicates that there is image quality jitter, the feature data of the training sample is stored in the vector database. Optionally, when storing the feature data in the vector database, the jitter severity indication information corresponding to the feature data, such as the jitter severity score, can also be stored in the feature vector corresponding to the feature data.
[0035] In some embodiments, to ensure the robustness of the target model, after obtaining the target model, the method may further include the following steps: Step 1: Input a set of verification samples into the target model to detect image quality jitter; The target model can obtain feature data corresponding to the input test sample. For example, it can extract feature data from the input test sample to obtain the feature data corresponding to the test sample. The obtained feature data is matched with the feature vectors in the vector database. If a feature vector is matched, the test sample is output as a detection result of image quality jitter. If no feature vector is matched, the test sample is output as a detection result of no image quality jitter.
[0036] Step 2: In response to the discrepancy between the detection results of the verification sample and the actual results of the verification sample, update the above model correction parameters; If the test results of the test sample are inconsistent with the actual results of the test sample, it indicates that the target model's detection is inaccurate, and the model calibration parameters can be further updated.
[0037] Step 3: Update the target value based on the updated model correction parameters; Step 4: Based on the updated target value and grayscale projection algorithm, use a set of validation samples to train the target model until the training termination condition is met, and obtain the target model.
[0038] The specific training process is the same as the process described above for training with training samples, and will not be repeated here.
[0039] In some embodiments, when a user encounters image quality jitter during playback, they can provide feedback to the target model's client by taking screenshots or other means to obtain sample data. After accumulating a certain number of samples, the target model's client can retrain the image quality detection algorithm to generate new feature vectors. We continue to perform jitter detection on the video frames during playback. This ensures the accuracy of the target model's detection results.
[0040] Figure 2 A flowchart illustrating a model acquisition method for image quality jitter detection provided in another exemplary embodiment of this application is shown, such as... Figure 2 As shown, the method may include the following steps: S201, Prepare sufficient training samples and store them in the image quality jitter detection training sample library. S202, Obtain training samples from the image quality jitter detection training sample library, calculate the target value using the above formula 2, train the target model using the training samples, and generate the feature vector of the training samples and store it in the vector database. S203, Inject new samples, detect the target model, and obtain the detection results output by the target model. These detection results indicate whether there is image quality jitter in the new samples. S204, determine whether the test results meet expectations, that is, determine whether the test results are consistent with the actual results of the new sample. If they meet expectations, proceed to S205; otherwise, proceed to S206.
[0041] For example, if the actual result of the new sample is that there is image quality jitter, and the detection result is also that there is image quality jitter, then the detection result is determined to be consistent with the actual result of the new sample; otherwise, the detection result is determined to be inconsistent with the actual result of the new sample.
[0042] S205, The algorithm for determining the target value meets the requirements, and the feature vectors in the feature vector library meet the requirements.
[0043] S206, optimize the target value algorithm by adjusting the first and second correction parameters in the target value algorithm, and repeat the training in S202.
[0044] Figure 3The illustration shows a flowchart of an exemplary embodiment of the image quality jitter detection method provided in this application. This method can be executed by an electronic device, which may be an electronic device for playing video, such as an electronic device playing IPTV video or OTT video. Figure 3 As shown, the method mainly includes the following steps.
[0045] S310 extracts keyframe images from the playing video at a predetermined frequency.
[0046] In this embodiment, keyframe images can be extracted from the video being played on an electronic device at a set frequency. For example, one keyframe image can be extracted every 10 frames, and the specific frequency can be determined according to the actual required detection accuracy.
[0047] S312, the above keyframe images are input into the target model for detection to obtain the detection results; wherein, the target model is used to detect image quality jitter in the input images. The target model is trained by at least one set of training samples based on the target value and grayscale projection algorithm. The above target value is the mean value used to eliminate image edge information. The detection results indicate whether there is image quality jitter in the keyframe images.
[0048] In this embodiment, based on the aforementioned target value and grayscale projection algorithm, a target model for detecting image quality jitter in the input image can be obtained through at least one set of training samples. During video playback, the extracted keyframe images are input into this target model for image quality jitter detection. Since the target model is trained based on the aforementioned target value...
[0049] In some embodiments, S312 may include the following steps: Step 1: Obtain the keyframe data corresponding to the keyframe image using the target model described above; Step 2: Using the target model described above, match the keyframe data with the feature vectors in the vector database, which stores the feature vectors corresponding to images with image quality jitter. Step 3: Based on whether there is a feature vector in the vector database that matches the keyframe data, obtain the corresponding detection result. If there is a feature vector in the vector database that matches the keyframe data, the detection result indicates that there is image quality jitter. If there is no feature vector in the vector database that matches the keyframe data, the detection result indicates that there is no image quality jitter.
[0050] In the above embodiments, the target model can store feature vectors of images with image quality jitter in the vector database during the training process. These feature vectors can record one or more pieces of information such as whether jitter exists, the severity of jitter, and pixel displacement. When detecting, the target model can determine whether the input keyframe image has image quality jitter based on whether there is a feature vector in the vector database that matches the keyframe data corresponding to the input keyframe image, thereby improving the detection efficiency.
[0051] In some embodiments, after obtaining the detection result in S312, in response to the detection result indicating that the keyframe image has image quality jitter, playback anomaly information can be reported to the Content Delivery Network (CDN) corresponding to the video. This playback anomaly information is used to indicate video playback anomalies. In these embodiments, when video jitter is detected, playback anomaly information can be reported to the CDN, thereby enabling the CDN to be aware of the video playback anomaly. Furthermore, based on the playback anomaly information, the CDN can diagnose the problem and determine the next steps to be taken to improve video playback quality.
[0052] Optionally, the above playback error information may include at least one of the following: 1) First-level information includes the video identifier and jitter severity indicator. For IPTV videos, the video identifier can be the channel number; for OTT videos, it can be the video name. However, it's not limited to these; the video identifier can also be other information that identifies the video. The jitter severity indicator can be a jitter severity score. For example, different scores can represent different levels of jitter severity, with higher values indicating greater jitter severity. This first-level information allows the CDN to identify videos with abnormal playback and their jitter severity, thus determining the appropriate actions to take.
[0053] 2) Second-level information: This includes information about the jittered pixel displacement, such as jitter deviation. This second-level information allows the CDN to understand the displacement of the jittered pixels and thus determine subsequent operations to be performed on the video.
[0054] 3) Third-level information, which includes the time information of the jitter occurrence. For example, the third-level information may include the timestamp of the jitter occurrence, which can be in milliseconds, so that the CDN can locate the specific time of the video anomaly.
[0055] It should be noted that although the above provides the specific levels of information included in playback error messages, it is not limited to these. In practical applications, the content of the reported playback error messages can be adjusted according to actual needs.
[0056] In some embodiments, prior to S310, the procedure described above can be followed. Figure 1 or Figure 2 The method shown is used to train the target model. For the specific training process, please refer to the above section on... Figure 1 or Figure 2 To avoid repetition, the relevant descriptions will not be repeated here.
[0057] Figure 4 A flowchart illustrating a method for detecting image jitter according to another exemplary embodiment of this application is shown. This method can be executed by an electronic device, which may be an electronic device for playing video, such as an electronic device playing IPTV video or OTT video. Figure 4 As shown, the method mainly includes the following steps.
[0058] S401, play IPTV or OTT video; S402, Target model initiated; S403, Image quality detection is performed using an image quality jitter detection algorithm optimized by the target model. In this step, the target model extracts keyframes from the video at a predetermined frequency and matches the keyframe data processed by the algorithm with the feature vectors in the vector database. S404, determine whether a feature vector is matched. If so, it is determined that there is an image quality jitter problem. The severity of the jitter is scored according to the matched feature vector, and S405 is executed. If no feature vector is matched, it is considered that there is no image quality jitter problem, and the process returns to step S403.
[0059] S405 reports playback error information to the operator's CDN.
[0060] Playback anomaly information can include multiple levels. The first level includes the channel number (IPTV), video name (OTT), and jitter severity score; the second level includes pixel displacement information of the jitter, such as jitter deviation; and the third level includes the timestamp of the jitter occurrence (accurate to milliseconds). The reporting method for playback anomaly information is not limited to the above hierarchical relationship; any similar representation is acceptable. Operators can use the anomaly information for diagnosis to determine the next steps required.
[0061] S406, User feedback regarding image quality judder. If users encounter image quality judder during video playback, they can use screenshots or other methods to report this issue to the target model.
[0062] S407 After receiving the returned video frame data, the target model extracts feature vectors after accumulating a certain number of samples and retrains the target model.
[0063] S408: Update the target model and target value calculation algorithm based on the training results to generate new feature vectors. Then return to S403 to continue jitter detection on the video frames during playback.
[0064] The technical solution provided in this application, based on a large model, for detecting image quality jitter in IPTV and OTT, can intelligently and accurately identify image quality problems, helping operators improve the user's viewing experience.
[0065] Figure 5 A structural block diagram of an electronic device 500 illustrated in an exemplary embodiment of this application is shown. The electronic device includes a Central Processing Unit (CPU) 501, a system memory 504 including Random Access Memory (RAM) 502 and Read-Only Memory (ROM) 503, and a system bus 505 connecting the system memory 504 and the CPU 501. The electronic device 500 also includes a mass storage device 506 for storing an operating system 509, a client 510, and other program modules 511.
[0066] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state storage technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage media are not limited to the above-mentioned types. The system memory 504 and mass storage device 506 described above can be collectively referred to as memory.
[0067] According to various embodiments of this disclosure, the electronic device 500 can also be connected to a remote computer on a network, such as the Internet. That is, the electronic device 500 can be connected to a network 508 via a network interface unit 507 connected to the system bus 505, or it can use the network interface unit 507 to connect to other types of networks or remote computer systems (not shown).
[0068] The memory further includes at least one instruction, at least one program, code set, or instruction set, which are stored in the memory. The central processing unit 501 executes the at least one instruction, at least one program, code set, or instruction set to implement all or part of the steps in the image quality jitter detection model acquisition method or image quality jitter detection method shown in the above embodiments.
[0069] In one exemplary embodiment, a readable storage medium is also provided, on which a program or instructions are stored. When executed by a processor, the program or instructions implement all or part of the steps in the above-described image quality jitter detection model acquisition method or image quality jitter detection method. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage device, etc.
[0070] In one exemplary embodiment, a computer program product is also provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, cause the computer to perform all or part of the steps in the above-described image quality jitter detection model acquisition method or image quality jitter detection method.
[0071] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.
[0072] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A method for detecting image jitter, characterized in that, include: Keyframe images are extracted from the playing video at a predetermined frequency. The keyframe image is input into the target model for detection to obtain the detection result; wherein, the target model is used to detect image quality jitter in the input image, and the target model is trained by at least one set of training samples based on the target value and grayscale projection algorithm, the target value is the mean value used to eliminate image edge information, and the detection result indicates whether the keyframe image has image quality jitter.
2. The method according to claim 1, characterized in that, The step of inputting the keyframe image into the target model for detection and obtaining the detection result includes: The keyframe data corresponding to the keyframe image is obtained through the target model; The keyframe data is matched with feature vectors in a vector database using the target model. The vector database stores feature vectors corresponding to images with image quality jitter. Based on whether there is a feature vector in the vector database that matches the keyframe data, a corresponding detection result is obtained. If there is a feature vector in the vector database that matches the keyframe data, the detection result indicates that there is image quality jitter. If there is no feature vector in the vector database that matches the keyframe data, the detection result indicates that there is no image quality jitter.
3. The method according to claim 1, characterized in that, After obtaining the detection result, the method further includes: In response to the detection result indicating that the keyframe image has image quality jitter, a playback anomaly information is reported to the content distribution network corresponding to the video, wherein the playback anomaly information is used to indicate that the video playback is abnormal.
4. The method according to claim 3, characterized in that, The playback error information includes at least one of the following: The first level of information includes the video's identifier and jitter severity indication information; The second level of information includes the pixel displacement information of the jitter; The third level of information includes the time information of the jitter occurrence.
5. The method according to any one of claims 1 to 4, characterized in that, Before extracting keyframe images from the played video at a predetermined frequency, the method further includes: Obtain the feature data corresponding to each of the training samples; The target value is obtained based on each of the feature data, the weight corresponding to each of the feature data, and the model correction parameters, wherein the model correction parameters are used to correct the target model; Based on the target value and the grayscale projection algorithm, the target model is trained using at least one set of training samples until the training termination condition is met, thereby obtaining the target model.
6. The method according to claim 5, characterized in that, The model calibration parameters include: multiple first calibration parameters and one second calibration parameter; obtaining the target value based on each feature data, the weights corresponding to each feature data, and the model calibration parameters includes: Calculate the average of the target differences of multiple feature data, wherein the target difference is the difference between a first value and a first correction parameter corresponding to the feature data, and the first value is the product of the feature data and the weight corresponding to the feature data; The target value is obtained by calculating the difference between the average value of the target difference and the second correction parameter.
7. The method according to claim 5, characterized in that, The step of training the target model using at least one set of training samples based on the target value and grayscale projection algorithm includes: The target model is used to calculate the row grayscale projection of each training sample in the current group before and after the displacement. The displacement corresponding to the training sample is obtained by summing the squares of the first differences of all rows within the search range of the offset, wherein the first difference is the difference between the second difference and the target value, and the second difference is the difference between the grayscale projection of the row after the displacement and the grayscale projection of the row before the displacement of the training sample. Based on the relationship between the displacement corresponding to each training sample in the current group and the target threshold corresponding to each training sample, the detection result of each training sample is obtained; In response to the fact that all the detection results of the current group are consistent with the actual results corresponding to the training samples, the target model is trained again using the next group of training samples until the training termination condition is met. The training termination condition includes: the detection results of N consecutive groups of training samples are consistent with the actual results of the training samples.
8. The method according to claim 7, characterized in that, After obtaining the detection result of each training sample based on the relationship between the displacement corresponding to each training sample in the current group and the target threshold corresponding to each training sample, the method further includes: In response to the existence of at least one detection result in the current group that is inconsistent with the actual result corresponding to the training sample, the model correction parameters are adjusted. Based on the adjusted model correction parameters, the target value is updated, and the process proceeds to the step of calculating the row grayscale projection of each training sample in the current group before and after displacement using the target model.
9. The method according to claim 7 or 8, characterized in that, Before using the next set of training samples to train the target model for the next round, the method further includes: In response to the detection result being consistent with the actual result corresponding to the training sample, and the detection result indicating the presence of image quality jitter, the feature data of the training sample is stored in the vector database.
10. The method according to claim 5, characterized in that, After obtaining the target model, the method further includes: A set of verification samples is input into the target model to detect image quality jitter. In response to the discrepancy between the detection result of the verification sample and the actual result of the verification sample, the model correction parameters are updated; The target value is updated based on the updated model correction parameters; Based on the updated target value and grayscale projection algorithm, the target model is trained using the set of verification samples until the training termination condition is met, thus obtaining the target model.
11. A method for obtaining a model for image quality jitter detection, characterized in that, include: Based on the target value and grayscale projection algorithm, the target model is trained using at least one set of training samples until the training termination condition is met, thereby obtaining the target model. The target model is used to perform image quality jitter detection on the input image, and the target value is the mean value used to eliminate image edge information.
12. The method according to claim 11, characterized in that, The target value and grayscale projection algorithm, using at least one set of training samples, trains the target model, including: Obtain the feature data corresponding to each of the training samples; The target value is obtained based on each of the feature data, the weight corresponding to each of the feature data, and the model correction parameters, wherein the model correction parameters are used to correct the target model; Based on the target value and the grayscale projection algorithm, the target model is trained using at least one set of training samples until the training termination condition is met, thereby obtaining the target model.
13. The method according to claim 12, characterized in that, The model calibration parameters include: multiple first calibration parameters and one second calibration parameter; obtaining the target value based on each feature data, the weights corresponding to each feature data, and the model calibration parameters includes: Calculate the average of the target differences of multiple feature data, wherein the target difference is the difference between a first value and a first correction parameter corresponding to the feature data, and the first value is the product of the feature data and the weight corresponding to the feature data; The target value is obtained by calculating the difference between the average value of the target difference and the second correction parameter.
14. The method according to claim 12, characterized in that, The step of training the target model using at least one set of training samples based on the target value and grayscale projection algorithm includes: The target model is used to calculate the row grayscale projection of each training sample in the current group before and after the displacement. The displacement corresponding to the training sample is obtained by summing the squares of the first differences of all rows within the search range of the offset, wherein the first difference is the difference between the second difference and the target value, and the second difference is the difference between the grayscale projection of the row after the displacement and the grayscale projection of the row before the displacement of the training sample. Based on the relationship between the displacement corresponding to each training sample in the current group and the target threshold corresponding to each training sample, the detection result of each training sample is obtained; In response to the fact that all the detection results of the current group are consistent with the actual results corresponding to the training samples, the target model is trained again using the next group of training samples until the training termination condition is met. The training termination condition includes: the detection results of N consecutive groups of training samples are consistent with the actual results of the training samples.
15. The method according to claim 12, characterized in that, After obtaining the detection result of each training sample based on the relationship between the displacement corresponding to each training sample in the current group and the target threshold corresponding to each training sample, the method further includes: In response to the existence of at least one detection result in the current group that is inconsistent with the actual result corresponding to the training sample, the model correction parameters are adjusted. Based on the adjusted model correction parameters, the target value is updated, and the process proceeds to the step of calculating the row grayscale projection of each training sample in the current group before and after displacement using the target model.
16. The method according to claim 14 or 15, characterized in that, Before using the next set of training samples to train the target model for the next round, the method further includes: In response to the detection result being consistent with the actual result corresponding to the training sample, and the detection result indicating jitter, the feature data of the training sample is stored in the vector database.
17. The method according to any one of claims 11 to 15, characterized in that, After obtaining the target model, the method further includes: A set of verification samples is input into the target model to detect image quality jitter. In response to the discrepancy between the detection result of the verification sample and the actual result of the verification sample, the model correction parameters are updated; The target value is updated based on the updated model correction parameters; Based on the updated target value and grayscale projection algorithm, the target model is trained using the set of verification samples until the training termination condition is met, thus obtaining the target model.
18. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the steps of the method as described in any one of claims 1 to 17.
19. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 17.
20. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, cause the computer to perform the steps of the method as described in any one of claims 1 to 17.