An image anomaly marking method and system based on image feature analysis

Abstract

The application relates to the technical field of image processing, and provides an image abnormality marking method and system based on image feature analysis. A first abnormality stage set and a second abnormality stage set are obtained by performing intersection operation on a texture reconstruction record set based on an analysis texture reconstruction process; an image abnormality is obtained by combining the first stage of the abnormality stage set with the texture reconstruction record; an abnormality feature set is obtained by performing multi-feature collection on the image abnormality; a model marking result of a target image of a target stage is obtained based on an intelligent marking model, and then abnormality information is obtained. The technical problem that the existing technology has insufficient abnormality detection effectiveness for three-dimensional reconstruction application images, resulting in insufficient appearance authenticity of a restored modeling object in three-dimensional geometric modeling is solved, the technical effects of improving the abnormality detection effectiveness for three-dimensional reconstruction application images, reducing the influence of an image abnormal area on three-dimensional reconstruction, and improving the appearance authenticity of the restored modeling object are achieved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image anomaly labeling method and system based on image feature analysis. Background Technology

[0002] With the continuous development of image processing technology, significant progress has been made in the field of reconstructed geometry, particularly in image-based 3D reconstruction technology. Among these advancements, texture reconstruction technology, used to restore the appearance of 3D models and enhance their realism, has garnered widespread attention from professionals in the economic market and scientific and academic fields.

[0003] Currently, various methods for restoring the appearance of 3D geometric models based on texture reconstruction technology suffer from insufficient realism in the restored appearance of the modeled object. The reason for this lies in the deviation occurring during the image acquisition stage of texture reconstruction. Specifically, due to the combined effects of differences in the object's material and structure, as well as the complex lighting conditions during image acquisition, the actual acquired image of the object often contains abnormal areas such as shadows and dark zones.

[0004] Applying images with abnormal areas that have not been processed directly to 3D model texture reconstruction severely interferes with the texture reconstruction effect, causing damage to the outline of the 3D geometric model and surface color deviation, resulting in insufficient realism of the appearance of the modeled object restored by 3D geometric modeling.

[0005] In summary, existing technologies suffer from insufficient effectiveness in detecting anomalies in 3D reconstruction application images, resulting in inadequate realism in the appearance of the modeled objects reconstructed by 3D geometric modeling. Summary of the Invention

[0006] Therefore, it is necessary to provide an image anomaly labeling method and system based on image feature analysis to address the above-mentioned technical problems. This method and system can improve the effectiveness of anomaly detection in 3D reconstruction applications, reduce the impact of abnormal regions in images on 3D reconstruction, and improve the realism of the appearance of the restored modeled object.

[0007] An image anomaly labeling method based on image feature analysis includes: analyzing the texture reconstruction process and constructing a first anomaly stage set; analyzing texture reconstruction records and constructing a second anomaly stage set; performing an intersection operation on the first and second anomaly stage sets to obtain a target anomaly stage set; extracting a first stage from the target anomaly stage set and combining it with the texture reconstruction records to obtain a first image anomaly of the first stage; obtaining a first anomaly type of the first image anomaly and performing multi-feature acquisition on the first anomaly type to obtain a first anomaly feature set; constructing an intelligent labeling model and storing the first stage, the first anomaly feature set, and their corresponding relationships in the intelligent labeling model; acquiring a target image of the target stage, acquiring the target feature set of the target image, and inputting it into the intelligent labeling model to obtain a model labeling result; and analyzing the model labeling result to obtain target anomaly information of the target image.

[0008] An image anomaly labeling system based on image feature analysis is disclosed. The system comprises: a texture analysis processing module for analyzing the texture reconstruction process and constructing a first anomaly stage set, and analyzing texture reconstruction records and constructing a second anomaly stage set; an anomaly stage analysis module for performing an intersection operation on the first anomaly stage set and the second anomaly stage set to obtain a target anomaly stage set; an image anomaly acquisition module for extracting a first stage of the target anomaly stage set and combining it with the texture reconstruction records to obtain a first image anomaly of the first stage; an anomaly feature acquisition module for acquiring a first anomaly type of the first image anomaly and performing multi-feature acquisition on the first anomaly type to obtain a first anomaly feature set; a labeling model construction module for constructing an intelligent labeling model and storing the first stage, the first anomaly feature set, and their corresponding relationships in the intelligent labeling model; a model labeling execution module for acquiring a target image of the target stage, acquiring a target feature set of the target image, and inputting it into the intelligent labeling model to obtain a model labeling result; and a target anomaly acquisition module for analyzing the model labeling result to obtain target anomaly information of the target image.

[0009] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps:

[0010] Analyze the texture reconstruction process and construct the first abnormal stage set; analyze the texture reconstruction records and construct the second abnormal stage set.

[0011] The first abnormal stage set and the second abnormal stage set are intersected to obtain the target abnormal stage set.

[0012] Extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage;

[0013] Obtain the first anomaly type of the first image anomaly, and perform multi-feature acquisition on the first anomaly type to obtain the first anomaly feature set;

[0014] Construct an intelligent tagging model, and store the first stage, the first abnormal feature set and their corresponding relationships into the intelligent tagging model;

[0015] The target image of the target stage is acquired, the target feature set of the target image is collected and input into the intelligent labeling model to obtain the model labeling result;

[0016] By analyzing the model labeling results, target anomaly information of the target image is obtained.

[0017] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0018] Analyze the texture reconstruction process and construct the first abnormal stage set; analyze the texture reconstruction records and construct the second abnormal stage set.

[0019] The first abnormal stage set and the second abnormal stage set are intersected to obtain the target abnormal stage set.

[0020] Extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage;

[0021] Obtain the first anomaly type of the first image anomaly, and perform multi-feature acquisition on the first anomaly type to obtain the first anomaly feature set;

[0022] Construct an intelligent tagging model, and store the first stage, the first abnormal feature set and their corresponding relationships into the intelligent tagging model;

[0023] The target image of the target stage is acquired, the target feature set of the target image is collected and input into the intelligent labeling model to obtain the model labeling result;

[0024] By analyzing the model labeling results, target anomaly information of the target image is obtained.

[0025] The above-mentioned image anomaly marking method and system based on image feature analysis solves the technical problem in the prior art that the effectiveness of anomaly detection in images used for 3D reconstruction is insufficient, resulting in insufficient realism of the appearance of the modeled object restored by 3D geometric modeling. It achieves the technical effect of improving the effectiveness of anomaly detection in images used for 3D reconstruction, reducing the impact of abnormal regions in images on 3D reconstruction, and improving the realism of the restored appearance of the modeled object.

[0026] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating an image anomaly labeling method based on image feature analysis in one embodiment;

[0028] Figure 2 This is a flowchart illustrating the process of constructing a smart tagging model in an image anomaly tagging method based on image feature analysis in one embodiment;

[0029] Figure 3 This is a structural block diagram of an image anomaly labeling system based on image feature analysis in one embodiment;

[0030] Figure 4 This is an internal structural diagram of a computer device in one embodiment.

[0031] Figure labeling: Texture analysis and processing module 1, anomaly stage analysis module 2, image anomaly acquisition module 3, anomaly feature acquisition module 4, labeling model construction module 5, model labeling execution module 6, target anomaly acquisition module 7. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0033] like Figure 1 As shown, this application provides an image anomaly labeling method based on image feature analysis, including:

[0034] S100: Analyze the texture reconstruction process and build the first abnormal stage set; analyze the texture reconstruction record and build the second abnormal stage set.

[0035] In one embodiment, the method step S100 of the method provided in this application, which analyzes the texture reconstruction process and assembles a first abnormal stage set, further includes:

[0036] S111: Obtain the texture reconstruction stage set according to the texture reconstruction process;

[0037] S112: Wherein, the texture reconstruction stage set includes a visibility judgment stage, a projection stage, a feature selection stage, and a feature mapping stage;

[0038] S113: Form an expert group, and have the expert group perform mechanism analysis on each stage of the texture reconstruction stage set in sequence, and form the first abnormal stage set based on the mechanism analysis results;

[0039] The first abnormal stage set refers to the reconstruction processing stage where abnormal features will theoretically occur.

[0040] In one embodiment, the method step S100 of the present application, which involves analyzing texture reconstruction records and assembling a second anomaly stage set, further includes:

[0041] S121: The texture reconstruction record includes records of multiple reconstructed image anomalies;

[0042] S122: Obtain the first reconstructed image anomaly record from the multiple reconstructed image anomaly records, wherein the first reconstructed image anomaly record includes a first stage;

[0043] S123: Count the number of first anomalies in the first stage and determine whether the number of first anomalies meets the preset anomaly threshold;

[0044] S124: If the condition is met, add the first stage to the second abnormal stage set.

[0045] Specifically, in this embodiment, texture reconstruction is an image processing technique that uses texture technology to register texture images with geometric models, thereby seamlessly pasting the texture images onto the three-dimensional geometric surface, restoring the real appearance and color of the modeled object on the three-dimensional geometric surface, and enhancing the visibility and realism of the three-dimensional geometric model.

[0046] The texture reconstruction process refers to the collective steps necessary for achieving 3D geometric texture reconstruction. The process includes a visibility assessment stage, a projection stage, a feature selection stage, and a feature mapping stage. The visibility assessment stage establishes the visual relationship between multiple pre-acquired entity images and the 3D geometric model, obtaining the coordinate positions of the entity images within the 3D geometric model. The projection stage projects the entity images onto the 3D geometric model based on their coordinate positions, thereby obtaining visible texture image patches representing each component of the 3D geometric model's appearance. The feature selection stage performs feature selection analysis on each of the multiple visible texture image patches that make up the appearance of the 3D geometric model, obtaining the spectral features, texture features, gradient intensity features, and optical geometric features of the multiple visible texture image patches. The feature mapping stage is used to perform feature mapping based on spectral features, texture features, gradient intensity features, and optical geometric features to obtain multiple sets of abnormal regions corresponding to multiple visible texture image blocks (abnormal regions are regions with abnormal features such as highlights, shadows, dark areas, blur, and broken lines). Based on multiple visible texture image blocks, the apparent texture of the three-dimensional geometric model is reconstructed, and multiple sets of abnormal regions are used to mark the abnormal regions in the three-dimensional texture reconstruction results.

[0047] The visibility judgment stage, projection stage, feature selection stage, and feature mapping stage constitute the texture reconstruction stage set of the texture reconstruction process. Any stage in the texture reconstruction stage set may contain abnormal region images that could affect the texture reconstruction effect.

[0048] Based on multiple expert components in the field of measurement / image modeling, expert technicians conduct mechanistic analysis on each stage of the texture reconstruction process based on their research experience, work experience, and texture reconstruction principles. The mechanistic analysis results determine whether a stage is a reconstruction process stage that is prone to producing or introducing abnormal features. The mechanistic analysis results include "yes" and "no".

[0049] Based on the reconstruction processing stages that are "yes" in the mechanism analysis results given by each expert for each stage, the frequency of each stage in the texture reconstruction stage set is obtained. The stages with a non-zero frequency are constituted as the first abnormal stage set. The first abnormal stage set refers to the reconstruction processing stages that theoretically will have abnormal features.

[0050] Specifically, in this embodiment, the texture reconstruction record is a record of multiple texture reconstructions performed in historical 3D geometric modeling. This historical reconstruction record includes multiple records of reconstructed image anomalies. These anomaly records are obtained by detecting image anomalies before and after each reconstruction processing stage during a single texture reconstruction. Each reconstructed image anomaly record typically contains 0 to 4 of the following stages: visibility judgment stage, projection stage, feature selection stage, and feature mapping stage. The frequency of occurrence of the visibility judgment stage, projection stage, feature selection stage, and feature mapping stage is statistically obtained based on these multiple reconstructed image anomaly records, constituting the first reconstructed image anomaly record.

[0051] The first stage is any reconstruction processing stage. Based on the first reconstructed image anomaly record, the first anomaly count of the first stage is obtained. The first anomaly count is the frequency of occurrence of the first stage in multiple reconstructed image anomaly records, that is, the frequency of image anomalies that occurred in the history of the first stage.

[0052] A preset anomaly threshold is used to determine whether an image anomaly occurring during the reconstruction process is an occasional anomaly or a normal occurrence. The specific value of the preset anomaly threshold can be set according to actual conditions, and this embodiment does not impose any restrictions. It is determined whether the number of the first anomalies meets the preset anomaly threshold. If it does, the first stage is added to the second anomaly stage set. The same processing method as the first stage is used to determine the anomaly in each reconstruction process stage in the texture reconstruction stage set, thus completing the construction of the second anomaly stage set. The second anomaly stage set consists of multiple reconstruction processes that are actually prone to exhibiting anomaly features.

[0053] This embodiment provides effective constraints for combining theory with practice to improve the targeting of texture reconstruction image anomaly analysis by obtaining reconstruction processing stages that theoretically tend to have image anomalies based on expert experience, and reconstruction processing stages that actually tend to have image anomalies based on historical data.

[0054] S200: Perform an intersection operation on the first abnormal stage set and the second abnormal stage set to obtain the target abnormal stage set;

[0055] Specifically, in this embodiment, the intersection of the first abnormal stage set and the second abnormal stage set is performed to obtain the target abnormal stage set, which is a reconstruction processing stage that is theoretically and practically prone to abnormal features.

[0056] Targeted analysis of multiple reconstruction processing stages within the target anomaly stage set can improve the efficiency of image anomaly analysis during reconstruction processing stages and shorten the time spent on image anomaly labeling compared to analyzing each reconstruction processing stage in the texture reconstruction stage set in a uniform manner.

[0057] S300: Extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage;

[0058] Specifically, it should be understood that in this embodiment, the target anomaly stage set includes multiple reconstruction processing stages, and the first stage is any random reconstruction processing stage in the target anomaly stage set. The first stage of the target anomaly stage set is extracted, and a search instruction is generated based on the actual reconstruction processing stage name of the first stage. The texture reconstruction records are traversed to obtain the first image anomaly of the first stage. The first image anomaly consists of multiple images with anomalies generated in the first stage, each image possessing any anomaly features such as highlights, shadows, dark areas, blur, and broken lines.

[0059] S400: Obtain the first abnormality type of the first image abnormality, and perform multi-feature acquisition on the first abnormality type to obtain the first abnormality feature set;

[0060] Specifically, in this embodiment, each reconstruction process stage may lead to image anomalies. The specific types of anomalies include highlights, shadows, dark areas, blur, and break lines. It should be understood that highlights are the specular reflections of a smooth object onto a light source; dark areas are regions with insufficient illumination on the object's surface, where the brightness is lower than the actual texture brightness; shadows are produced when a light source encounters an opaque object; blur refers to areas in the image where objects are indistinct or unclear, lacking definition; and break lines refer to the boundary between foreground objects and the background in the image, which can easily cause texture seams on the model's surface.

[0061] Based on the first image anomaly, a first anomaly type is obtained by manually dividing the image anomaly region and marking the anomaly features. The first anomaly type is the specific anomaly type present in the multiple images containing anomalies included in the first image anomaly. For example, among the multiple images containing anomalies generated through the projection stage, a certain anomaly image may have a region containing both shadow and blurry image anomalies, as well as two types of image anomalies and two types of image anomaly features.

[0062] Based on the first anomaly type, the system divides and labels the anomaly regions containing various anomaly features to obtain image features for each anomaly type, thus forming the first anomaly feature set. This first anomaly feature set represents the anomaly types that are prone to occur during the first stage of reconstruction processing, as well as the specific image state representation of each anomaly type.

[0063] S500: Construct a smart tagging model and store the first stage, the first abnormal feature set and their corresponding relationships into the smart tagging model;

[0064] In one embodiment, such as Figure 2 As shown, the method step S500 provided in this application further includes: constructing a smart tagging model and storing the first stage, the first abnormal feature set, and their corresponding relationships into the smart tagging model;

[0065] S510: Obtain the abnormal data of each abnormal stage in the target abnormal stage set and form a target abnormal stage-feature list.

[0066] S520: Using the target anomaly stage-feature list as the first training data, train the first labeled model;

[0067] S530: Adjust the first training data to obtain the second training data;

[0068] S540: Based on the second training data, a second labeled model is trained, and the process continues iterating until the Nth labeled model is obtained;

[0069] S550: The smart tagging model is obtained by superimposing the first tagging model, the second tagging model, and up to the Nth tagging model.

[0070] In one embodiment, the method step S530 of adjusting the first training data to obtain the second training data further includes:

[0071] S531: Acquire detection data, wherein the detection data includes detection stage image, detection stage image features, and detection stage image anomaly markers;

[0072] S532: The detection stage image and its features are processed using the first labeling model to obtain the detection labeling result;

[0073] S533: Compare the detection marker result with the image anomaly marker in the detection stage to obtain a first comparison result;

[0074] S534: Adjust the first training data according to the first comparison result to obtain the second training data.

[0075] Specifically, in this embodiment, the smart labeling model is a data processing model used to identify image anomalies and label specific image anomaly feature types in images that have entered any reconstruction processing stage. The input data of the smart labeling model is the image after processing through any reconstruction processing stage, and the output results are the region division labels of the specific anomaly features in the image and the anomaly feature type labels.

[0076] The smart tagging model includes a data input layer, a reconstruction processing stage identification layer, a smart tagging layer, and a data output layer. Abnormal data for each reconstruction processing stage (i.e., anomaly stage) in the target anomaly stage set is acquired, and multiple anomaly feature sets corresponding to each anomaly stage in the target anomaly stage set are obtained using the same method as steps S300-S400. The first stage, the first anomaly feature set, and their corresponding relationships are stored in the reconstruction processing stage identification layer of the smart tagging model. Multiple anomaly feature sets corresponding to each anomaly stage in the target anomaly stage set are stored in the reconstruction processing stage identification layer of the smart tagging model using the same storage method as the first stage and the first anomaly feature set.

[0077] When an image processed by a certain reconstruction stage is input through the data input layer, the reconstruction stage recognition layer calls the corresponding abnormal feature set according to the reconstruction stage. The smart labeling layer labels the abnormal features present in the image based on the image and the types of abnormal features present in the abnormal feature set. It should be understood that the smart labeling model essentially only requires model training for the smart labeling layer to improve the accuracy of abnormal feature recognition and labeling.

[0078] A retrieval instruction is generated based on the actual reconstruction processing stage name of the first stage. The texture reconstruction record is traversed to obtain the first image anomaly of the first stage. The first image anomaly consists of multiple images with anomalies generated in the first stage. Based on the first image anomaly, anomaly feature marking is performed manually to obtain the anomaly feature marking of the first image anomaly.

[0079] Using the same method as the abnormal feature labeling of the first abnormal image in the first stage, multiple sets of multiple abnormal images corresponding to each reconstruction processing stage in the target abnormal stage set are obtained, forming a target abnormal stage-feature list. The feature list in the target abnormal stage-feature list specifically includes multiple abnormal images and feature type labels of specific abnormal features in the abnormal images.

[0080] The target anomaly stage-feature list is used for model training of the smart tagging layer in the smart tagging model. The recognition layer of the reconstruction processing stage calls the corresponding anomaly feature set according to the target anomaly stage. The anomaly feature images in this anomaly feature set are the types of anomalies that are likely to be caused to the image during the processing of the reconstruction processing stage, and the image features of the corresponding anomaly feature types.

[0081] The intelligent labeling layer of the intelligent labeling model is constructed based on a backpropagation (BP) neural network. Multiple anomaly feature types and their corresponding specific images and labels are extracted from the target anomaly stage-feature list. Multiple sets of anomaly feature images and labels are used as the first training data. This first training data is divided into a training set and a test set with a 9:1 data volume ratio. The intelligent labeling model is trained multiple times based on the training and test sets. Training stops when half of the data in the first training data has participated in the training of the intelligent labeling model, and the current intelligent labeling model is labeled as the first labeling model.

[0082] Acquire detection data, which includes detection stage images, detection stage image features, and detection stage image anomaly markers. Directly input the monitoring data into the intelligent recognition layer of the first marker model. The intelligent recognition layer processes the detection stage images and detection stage image features to obtain the detection marker results.

[0083] By comparing the detection labeling results with the image anomaly labels of the detection stage, a first comparison result is obtained. The first comparison result characterizes the image anomaly labeling accuracy of the actual first labeling machine model. The method for obtaining the first comparison result is to count the number of labeling result data that match the image anomaly labels of the detection stage in the detection labeling results, and use the percentage of this data in the number of labeling data of the image anomaly labels of the detection stage as the first comparison result.

[0084] When the first comparison result is less than or equal to 98%, the first training data is adjusted according to the first comparison result. Half of the data is extracted from the remaining half of the first training data that did not participate in model training, that is, half of the total amount of the first training data is added to the first training data that has participated in model training to obtain the second training data.

[0085] Based on the second training data, a second labeled model is trained. A second alignment result is then obtained based on the detection data. When the first alignment result is less than or equal to 98%, the first training data is adjusted according to the first alignment result. Half of the data (i.e., 1 / 8 of the total first training data) that was not used in model training is extracted and added to the second training data already used in model training to obtain the third training data. This process continues iterating until the Nth labeled model is obtained, where the Nth alignment result of the Nth labeled model is greater than or equal to 98%.

[0086] The smart tagging model is obtained by superimposing the first tagging model, the second tagging model, and up to the Nth tagging model, wherein the smart tagging layer of the smart tagging model is the smart tagging layer of the Nth tagging model.

[0087] This embodiment achieves the technical effect of quickly, accurately, and efficiently identifying and marking image anomalies in the current texture reconstruction image by constructing an intelligent tagging model after obtaining the texture reconstruction image and the corresponding reconstruction processing stage. This is achieved by calling the image anomaly features and anomaly feature images that are prone to be generated in the reconstruction processing stage within the intelligent tagging model.

[0088] S600: Acquire the target image of the target stage, collect the target feature set of the target image and input it into the intelligent labeling model to obtain the model labeling result;

[0089] In one embodiment, the method step S600 of the method provided in this application further includes: acquiring the target image in the target stage, collecting the target feature set of the target image and inputting it into the smart labeling model to obtain the model labeling result.

[0090] S610: Collect the target Haar features of the target image and filter the target Haar features to obtain the target feature set;

[0091] S620: Obtain the first target feature in the target feature set, and analyze the first target feature through the smart tagging model to obtain the first analysis result;

[0092] S630: Analyze the first analysis result and determine whether the first target feature is abnormal;

[0093] S640: If an exception occurs, obtain the first exception flag instruction;

[0094] S650: Wherein, the first anomaly marking instruction is used to mark the first target feature as an anomaly.

[0095] In one embodiment, after analyzing the first analysis result and determining whether the first target feature is abnormal, the method step S630 provided in this application further includes:

[0096] S631: If there is no abnormality, obtain the first normal flag instruction;

[0097] S632: Wherein, the first normal marking instruction is used to normally mark the first target feature.

[0098] Specifically, in this embodiment, the target image is acquired after being processed by the target stage, the target Haar features of the target image are collected, and the target Haar features are filtered to obtain the target feature set, which consists of multiple regions or multiple points in the target image that may have abnormal features.

[0099] The target stage and target image are input into the smart tagging model, and anomaly tagging of the target image is performed based on the smart tagging model to obtain an anomaly feature tagging set corresponding to the target feature set.

[0100] Based on the correspondence between the abnormal feature marker set and the target feature set, the first abnormal feature marker corresponding to the first target feature in the target feature set is obtained. The similarity between the first abnormal feature marker and the first target feature is analyzed using a gray similarity algorithm to obtain a first analysis result, which is a similarity value.

[0101] A preset feature anomaly threshold is defined as a similarity value. When the actual analysis result obtained is higher than the feature anomaly threshold, the corresponding target feature is considered to be an anomaly feature; otherwise, it is considered not to be an anomaly feature.

[0102] The first target feature is determined to be abnormal by referring to the feature anomaly threshold and the first analysis result. If it is abnormal, a first anomaly marking instruction is obtained. The first anomaly marking instruction is used to mark the first target feature as abnormal.

[0103] The first target feature is determined to be abnormal by referring to the feature anomaly threshold and the first analysis result. If it is not abnormal, a first normal marking instruction is obtained. The first normal marking instruction is used to mark the first target feature normally.

[0104] By using the same method as labeling the first target feature as normal / abnormal, all target features in the target feature set are labeled to obtain the model labeling result, thus achieving the technical effect of obtaining image anomaly labeling results that accurately represent the image anomaly generation during the reconstruction processing stage.

[0105] S700: Analyze the model labeling results to obtain target anomaly information of the target image.

[0106] Specifically, in this embodiment, the target anomaly information refers to the reconstruction process stage in the texture reconstruction process that is theoretically and practically prone to anomalies, the type of image anomaly generated in the current texture reconstruction, and the specific location identifier of the anomaly in the image.

[0107] Based on the target anomaly information, the technicians responsible for texture reconstruction can determine the specific stage in the texture reconstruction process where the image anomaly occurred, as well as the specific anomaly feature type and the location marker of the anomaly on the image, and thus perform targeted image processing.

[0108] This embodiment achieves the technical effect of improving the effectiveness of anomaly detection in 3D reconstruction application images, thereby avoiding the use of images with abnormal defects for texture reconstruction, reducing the impact of abnormal areas in images on 3D reconstruction, and improving the realism of the appearance of the modeled object.

[0109] In one embodiment, such as Figure 3 As shown, an image anomaly labeling system based on image feature analysis is provided, including: a texture analysis processing module 1, an anomaly stage analysis module 2, an image anomaly acquisition module 3, an anomaly feature acquisition module 4, a labeling model construction module 5, a model labeling execution module 6, and a target anomaly acquisition module 7, wherein:

[0110] Texture analysis and processing module 1 is used to analyze the texture reconstruction process and build a first abnormal stage set, and to analyze the texture reconstruction record and build a second abnormal stage set.

[0111] Anomaly stage analysis module 2 is used to perform an intersection operation on the first anomaly stage set and the second anomaly stage set to obtain the target anomaly stage set;

[0112] Image anomaly acquisition module 3 is used to extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage;

[0113] The abnormal feature acquisition module 4 is used to acquire the first abnormal type of the first image abnormality, and to acquire multiple features of the first abnormal type to obtain the first abnormal feature set;

[0114] The tagging model construction module 5 is used to construct an intelligent tagging model and store the first stage, the first abnormal feature set and their corresponding relationships into the intelligent tagging model;

[0115] Model labeling execution module 6 is used to acquire the target image in the target stage, collect the target feature set of the target image and input it into the intelligent labeling model to obtain the model labeling result;

[0116] The target anomaly acquisition module 7 is used to analyze the model labeling results and obtain target anomaly information of the target image.

[0117] In one embodiment, the system further includes:

[0118] A texture reconstruction execution unit is used to obtain a texture reconstruction stage set according to the texture reconstruction process;

[0119] A reconstruction stage generation unit is used in which the texture reconstruction stage set includes a visibility judgment stage, a projection stage, a feature selection stage, and a feature mapping stage;

[0120] The mechanism analysis execution unit is used to form an expert group, and the expert group performs mechanism analysis on each stage of the texture reconstruction stage set in sequence, and forms the first abnormal stage set based on the mechanism analysis results;

[0121] A stage reconstruction processing unit is used in which the first abnormal stage set refers to the reconstruction processing stage that theoretically exhibits abnormal characteristics.

[0122] In one embodiment, the system further includes:

[0123] An anomaly record acquisition unit is used for the texture reconstruction record to include anomaly records of multiple reconstructed images;

[0124] An anomaly record extraction unit is used to obtain a first reconstructed image anomaly record from the multiple reconstructed image anomaly records, wherein the first reconstructed image anomaly record includes a first stage;

[0125] An anomaly count unit is used to count the number of anomalies in the first stage and determine whether the number of anomalies meets a preset anomaly threshold.

[0126] An exception stage adding unit is used to add the first stage to the second exception stage set if the condition is met.

[0127] In one embodiment, the system further includes:

[0128] The feature list construction unit is used to acquire the abnormal data of each abnormal stage in the target abnormal stage set and form a target abnormal stage-feature list.

[0129] The model training execution unit is used to train the first labeled model by using the target anomaly stage-feature list as the first training data.

[0130] The training data adjustment unit is used to adjust the first training data to obtain the second training data;

[0131] The model training iteration unit is used to train the second labeled model based on the second training data, and continue to iterate until the Nth labeled model is obtained;

[0132] The tagging model generation unit is used to obtain the smart tagging model by superimposing the first tagging model, the second tagging model, and up to the Nth tagging model.

[0133] In one embodiment, the system further includes:

[0134] A detection data acquisition unit is used to acquire detection data, wherein the detection data includes detection stage images, detection stage image features, and detection stage image anomaly markers;

[0135] The detection label acquisition unit is used to process the detection stage image and the detection stage image features through the first label model to obtain the detection label result;

[0136] An anomaly labeling execution unit is used to compare the detection labeling result with the image anomaly labeling in the detection stage to obtain a first comparison result;

[0137] The training data acquisition unit is used to adjust the first training data according to the first comparison result to obtain the second training data.

[0138] In one embodiment, the system further includes:

[0139] The target feature acquisition unit is used to acquire the target Haar features of the target image and filter the target Haar features to obtain the target feature set;

[0140] The target feature analysis unit is used to acquire the first target feature in the target feature set, and analyze the first target feature through the smart tagging model to obtain the first analysis result;

[0141] The feature anomaly judgment unit is used to analyze the first analysis result and determine whether the first target feature is abnormal.

[0142] The flag instruction acquisition unit is used to acquire the first exception flag instruction if an exception is found.

[0143] The first anomaly marking instruction is used to mark the first target feature as an anomaly.

[0144] In one embodiment, the system further includes:

[0145] The marking instruction generation unit is used to obtain the first normal marking instruction if there is no abnormality.

[0146] A marking instruction application unit is configured such that the first normal marking instruction is used to normally mark the first target feature.

[0147] For a specific embodiment of an image anomaly labeling system based on image feature analysis, please refer to the embodiment of an image anomaly labeling method based on image feature analysis described above, which will not be repeated here. Each module in the above-described image anomaly labeling system based on image feature analysis can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0148] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores news data and data such as time decay factors. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements an image anomaly labeling method based on image feature analysis.

[0149] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: analyzing a texture reconstruction process and constructing a first abnormal stage set; analyzing texture reconstruction records and constructing a second abnormal stage set; performing an intersection operation on the first abnormal stage set and the second abnormal stage set to obtain a target abnormal stage set; extracting a first stage from the target abnormal stage set and combining it with the texture reconstruction records to obtain a first image abnormality of the first stage; obtaining a first abnormality type of the first image abnormality and performing multi-feature acquisition on the first abnormality type to obtain a first abnormality feature set; constructing a smart labeling model and storing the first stage, the first abnormality feature set, and their corresponding relationships in the smart labeling model; obtaining a target image of the target stage, acquiring the target feature set of the target image and inputting it into the smart labeling model to obtain a model labeling result; and analyzing the model labeling result to obtain target abnormality information of the target image.

[0151] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0152] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An image anomaly labeling method based on image feature analysis, characterized in that, include: Analyze the texture reconstruction process and construct the first abnormal stage set; analyze the texture reconstruction records and construct the second abnormal stage set. The first abnormal stage set and the second abnormal stage set are intersected to obtain the target abnormal stage set. Extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage; Obtain the first anomaly type of the first image anomaly, and perform multi-feature acquisition on the first anomaly type to obtain the first anomaly feature set; Construct an intelligent tagging model, and store the first stage, the first abnormal feature set and their corresponding relationships into the intelligent tagging model; The target image of the target stage is acquired, the target feature set of the target image is collected and input into the intelligent labeling model to obtain the model labeling result; Analyze the model labeling results to obtain target anomaly information of the target image; The texture reconstruction process is to realize three-dimensional geometric texture reconstruction, and the texture reconstruction process includes a visibility judgment stage, a projection stage, a feature selection stage, and a feature mapping stage. The texture reconstruction record is a record of multiple texture reconstructions performed in 3D geometric modeling in history. The historical reconstruction record includes multiple records of reconstructed image anomalies. The reconstructed image anomaly record is obtained by performing image anomaly detection before and after each reconstruction processing stage during a single texture reconstruction. Each reconstructed image anomaly record often contains 0 to 4 of the following stages: visibility judgment stage, projection stage, feature selection stage, and feature mapping stage.

2. The image anomaly marking method according to claim 1, characterized in that, The analysis of texture reconstruction and the formation of the first anomaly stage set include: The texture reconstruction stage set is obtained according to the texture reconstruction process; The texture reconstruction stage set includes a visibility judgment stage, a projection stage, a feature selection stage, and a feature mapping stage. An expert group is formed, and the expert group performs mechanism analysis on each stage of the texture reconstruction stage set in sequence. The first abnormal stage set is formed based on the mechanism analysis results. The first abnormal stage set refers to the reconstruction processing stage where abnormal features will theoretically occur.

3. The image anomaly marking method according to claim 2, characterized in that, The analysis of texture reconstruction records and the assembly of a second anomaly stage set include: The texture reconstruction record includes records of multiple reconstructed image anomalies. Obtain the first reconstructed image anomaly record from the multiple reconstructed image anomaly records, wherein the first reconstructed image anomaly record includes a first stage; Count the number of first anomalies in the first stage and determine whether the number of first anomalies meets the preset anomaly threshold; If the condition is met, the first stage is added to the second abnormal stage set.

4. The image anomaly marking method according to claim 1, characterized in that, The step of constructing a smart tagging model and storing the first stage, the first abnormal feature set, and their corresponding relationships into the smart tagging model includes: Obtain the abnormal data of each abnormal stage in the target abnormal stage set and form a target abnormal stage-feature list; The target anomaly stage-feature list is used as the first training data to train the first labeling model; The first training data is adjusted to obtain the second training data; Based on the second training data, a second labeled model is trained, and the process continues iteratively until the Nth labeled model is obtained. The smart tagging model is obtained by superimposing the first tagging model, the second tagging model, and so on up to the Nth tagging model.

5. The image anomaly marking method according to claim 4, characterized in that, The step of adjusting the first training data to obtain the second training data includes: Acquire detection data, wherein the detection data includes images from the detection stage, image features from the detection stage, and anomaly markers from the images from the detection stage; The detection stage image and its features are processed using the first labeling model to obtain the detection labeling result; By comparing the detection labeling results with the image anomaly labels in the detection phase, a first comparison result is obtained; The first training data is adjusted based on the first comparison result to obtain the second training data.

6. The image anomaly marking method according to claim 5, characterized in that, The process of acquiring the target image in the target acquisition stage involves collecting the target feature set of the target image and inputting it into the intelligent labeling model to obtain the model labeling result, including: Collect the target Haar features of the target image, and filter the target Haar features to obtain the target feature set; Obtain the first target feature from the target feature set, and analyze the first target feature using the smart tagging model to obtain the first analysis result; Analyze the first analysis result and determine whether the first target feature is abnormal; If an exception is detected, the first exception flag instruction is obtained; The first anomaly marking instruction is used to mark the first target feature as an anomaly.

7. The image anomaly marking method according to claim 6, characterized in that, After analyzing the first analysis result and determining whether the first target feature is abnormal, the method further includes: If there are no abnormalities, obtain the first normal marking instruction; The first normal marking instruction is used to mark the first target feature normally.

8. An image anomaly labeling system based on image feature analysis, employing the method described in any one of claims 1 to 7, characterized in that, The system includes: The texture analysis and processing module is used to analyze the texture reconstruction process and build a first abnormal stage set, and to analyze the texture reconstruction records and build a second abnormal stage set. An anomaly stage analysis module is used to perform an intersection operation on the first anomaly stage set and the second anomaly stage set to obtain the target anomaly stage set. The image anomaly acquisition module is used to extract the first stage of the target anomaly stage set and combine it with the texture reconstruction record to obtain the first image anomaly of the first stage. An abnormal feature acquisition module is used to acquire the first abnormal type of the first image abnormality and to acquire multiple features of the first abnormal type to obtain a first abnormal feature set. A tagging model construction module is used to construct an intelligent tagging model and store the first stage, the first abnormal feature set and their corresponding relationships into the intelligent tagging model; The model labeling execution module is used to acquire the target image in the target stage, collect the target feature set of the target image and input it into the intelligent labeling model to obtain the model labeling result; The target anomaly acquisition module is used to analyze the model labeling results and obtain target anomaly information of the target image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images