Yuan blue floral pattern gene library construction and ai derived design system and method

Abstract

The application relates to the technical field of decoration gene library construction, and particularly discloses a Yuan blue and white decoration gene library construction and AI derivative design system and method, the method comprises the following steps: carrying out decoration feature extraction on a surface graph, partitioning the surface graph according to the decoration feature extraction result, identifying each region based on AI, and outputting an identification result; the partitioning result and the identification result of the surface graph of each sample are counted, a sample sample is created; the sample samples are compared, the information acquisition sequence is determined, sample information is acquired, and a decoration gene library is constructed; the surface of a sample model is identified based on AI, an identification result is obtained, the sample is classified according to the identification process and the identification result, the importance of each sample is determined according to the classification result, the information acquisition sequence is further determined, the information of more important samples is acquired earlier, and the data quality of the gene library is improved, so that the gene library can be put into use earlier.

Description

Technical Field

[0001] This invention relates to the field of decorative pattern gene bank construction technology, specifically a Yuan blue and white porcelain decorative pattern gene bank construction and AI-derived design system and method. Background Technology

[0002] The "Yuan Blue and White Porcelain Pattern Gene Bank" typically refers to a database obtained by systematically sorting, classifying, digitally modeling, and database-organizing the decorative elements on Yuan Dynasty blue and white porcelain. This database is used for research in various fields such as art, archaeology, artifact authentication, and cultural heritage preservation. Currently, pattern identification and archiving are all done manually, which is tedious and requires highly skilled personnel. With the development of AI technology, AI can be used to identify and archive patterns. However, in this type of artistic identification process, AI can only obtain most of the main information, and many details will inevitably be missing. In this case, manual supplementation is required, but the manual supplementation process is time-consuming, making it difficult to put the gene bank into use. Therefore, how to improve the data quality of the gene bank so that it can be put into use as soon as possible is the technical problem that this invention aims to solve. Summary of the Invention

[0003] The purpose of this invention is to provide a system and method for constructing a gene bank of Yuan blue and white porcelain patterns and for AI-derived design, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A system and method for constructing a gene library of Yuan blue and white porcelain decorative patterns and for AI-derived design, the method comprising:

[0006] The sample image is acquired according to the preset shooting trajectory, the sample model is constructed according to the sample image, and the surface of the three-dimensional model is simultaneously extended in two dimensions to obtain the surface map.

[0007] The surface map is subjected to texture feature extraction. Based on the texture feature extraction results, the surface map is divided into regions. Each region is identified based on AI, and the identification results are output.

[0008] The zoning and identification results of the surface map for each sample are statistically analyzed to create a sample sample.

[0009] By comparing sample samples, determining the order of information acquisition, obtaining sample information, and constructing a pattern gene library;

[0010] Guiding information for the design process generated based on the pattern gene library.

[0011] Preferably, the steps of acquiring sample images according to a preset shooting trajectory, constructing a sample model based on the sample images, and simultaneously performing two-dimensional extension on the surface of the three-dimensional model to obtain a surface map include:

[0012] An initial sample image is acquired based on a preset standard viewpoint, and a basic model of the sample is created based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the dimensions of the circumscribed cuboid of the basic model are preset values.

[0013] Obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the camera.

[0014] The sphere is divided into latitude and longitude sections according to a preset step size to obtain spherical sub-regions;

[0015] Connect the spherical sub-region and the center of the sphere to determine the detection direction, send the detection direction to the imaging end, and acquire the final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance;

[0016] Based on the final-state sample image correction model, a sample model is constructed.

[0017] A cross section is created based on the preset direction corresponding to the standard viewpoint, and the boundary line between the cross section and the sample model is obtained. The sample image surface is then extended in two dimensions based on one of the boundary lines to obtain a surface map.

[0018] Preferably, ambient light parameters are recorded during the acquisition of the final state sample image containing the detection direction by the imaging end; color value restoration is performed on the acquired final state sample image based on the ambient light parameters;

[0019] The process of constructing the sample model based on the final state sample image correction base model includes:

[0020] For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction;

[0021] The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

[0022] Preferably, the steps of extracting decorative features from the surface image, dividing the surface image into regions based on the extracted decorative features, identifying each region based on AI, and outputting the identification results include:

[0023] Iterate through each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel.

[0024] The boundaries of the decorative pattern are located based on the horizontal and vertical gradients;

[0025] Select boundary points on the boundary according to the preset step size, calculate the curvature at the boundary points, and mark the boundary points whose curvature is greater than the preset curvature threshold.

[0026] Cluster the marked boundary points, and partition the boundary points according to the clustering results;

[0027] Each region is identified using AI, and the identification results are output.

[0028] Preferably, the steps of comparing sample samples, determining the information acquisition order, acquiring sample information, and constructing a pattern gene library include:

[0029] Compare the partitioning results of the sample samples and calculate the first distance between the sample samples;

[0030] When the first distance is less than a preset first distance threshold, compare the recognition results and calculate the second distance;

[0031] Samples whose second distance is less than the preset second distance threshold are grouped into one category;

[0032] Calculate the priority score for each sample, and use the descending order of the priority scores as the information acquisition order;

[0033] Based on the order of information acquisition, sample information acquisition requests are sent, and the partitioning results, identification results, and sample information of the acquired samples are statistically analyzed to construct a pattern gene library.

[0034] Preferably, the method further includes:

[0035] The calculation process for the first distance is as follows:

[0036] Calculate the area of ​​each sub-region, and statistically analyze the area of ​​each sub-region based on its location to obtain an area matrix. Compare the area matrices to calculate the first distance.

[0037] The calculation process for the second distance is as follows:

[0038] The recognition results are converted into a word library. The intersection and union of the word libraries of the two samples are calculated. The intersection-union ratio is calculated. The second distance is determined based on the inverse ratio of the intersection-union ratio.

[0039] The priority score is calculated as follows:

[0040] Obtain the total number of samples in the class to which the sample belongs, and the number of samples in the class to which the sample belongs whose sample information has been obtained. Use the total number of samples as an intermediate indicator and the number of samples as an inverse indicator to calculate the priority score together.

[0041] The present invention also provides a system for constructing a gene bank for Yuan blue and white porcelain patterns and for AI-derived design, the system comprising:

[0042] The modeling module is used to acquire sample images according to a preset shooting trajectory, construct a sample model based on the sample images, and simultaneously extend the surface of the three-dimensional model in two dimensions to obtain a surface map.

[0043] The pattern extraction and recognition module is used to extract pattern features from the surface map, divide the surface map into regions based on the pattern feature extraction results, recognize each region based on AI, and output the recognition results.

[0044] The sample creation module is used to statistically analyze the zoning and identification results of the surface map of each sample and create the sample.

[0045] The gene bank construction module is used to compare sample samples, determine the order of information acquisition, obtain sample information, and construct a pattern gene bank.

[0046] The guidance information generation module is used to generate guidance information for the design process based on the pattern gene library.

[0047] Preferably, the modeling module includes:

[0048] The basic model creation unit is used to acquire an initial sample image based on a preset standard viewpoint and create a basic model of the sample based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the size of the circumscribed cuboid of the basic model is a preset value.

[0049] The sphere creation unit is used to obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the shooting end;

[0050] The spherical segmentation unit is used to segment the sphere according to the latitude and longitude according to the preset step size to obtain spherical sub-regions;

[0051] The image acquisition unit is used to connect the spherical sub-region and the center of the sphere, determine the detection direction, send the detection direction to the imaging end, and acquire a final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance;

[0052] The sample model building unit is used to construct a sample model based on the final state sample image and the correction of the base model.

[0053] The extension unit is used to create a cross section based on a preset direction corresponding to the standard viewpoint, obtain the boundary line between the cross section and the sample model, and perform two-dimensional extension on the sample image surface based on one of the boundary lines to obtain a surface map.

[0054] Preferably, ambient light parameters are recorded during the acquisition of the final state sample image containing the detection direction by the imaging end; color value restoration is performed on the acquired final state sample image based on the ambient light parameters;

[0055] The process of constructing the sample model based on the final state sample image correction base model includes:

[0056] For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction;

[0057] The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

[0058] Preferably, the pattern extraction and recognition module includes:

[0059] The gradient calculation unit is used to traverse each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel.

[0060] Boundary positioning unit, used to locate the decorative boundary according to the horizontal and vertical gradients;

[0061] The boundary point marking unit is used to select boundary points on the boundary according to a preset step size, calculate the curvature at the boundary points, and mark boundary points whose curvature is greater than a preset curvature threshold.

[0062] Clustering units are used to cluster the marked boundary points and partition the boundary points based on the clustering results;

[0063] The recognition execution unit is used to identify each region based on AI and output the recognition results.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] This invention captures sample images, constructs sample models based on the images, identifies the surface of the sample models using AI, obtains identification results, categorizes the samples based on the identification process and results, determines the importance of each sample based on the categorization results, and thus determines the order of information acquisition, allowing information from more important samples to be acquired first, thereby indirectly improving the data quality of the gene bank and enabling it to be put into use earlier. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0067] Figure 1The flowchart illustrates the overall process of constructing a gene bank for Yuan blue-and-white porcelain patterns and the AI-derived design method.

[0068] Figure 2 The diagram shows the structure of the Yuan blue and white porcelain pattern gene bank construction and AI-derived design system. Detailed Implementation

[0069] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0070] Figure 1 This is a flowchart illustrating the overall process of constructing a Yuan blue-and-white porcelain pattern gene bank and an AI-derived design system and method. In this embodiment of the invention, a method for constructing a Yuan blue-and-white porcelain pattern gene bank and an AI-derived design method includes:

[0071] Step S100: Acquire sample images according to the preset shooting trajectory, construct sample models based on sample images, and simultaneously extend the surface of the three-dimensional model in two dimensions to obtain a surface map;

[0072] For blue and white porcelain pattern samples, the images obtained by shooting them according to the preset shooting trajectory are called sample images. The shooting trajectory is used to characterize how the shooting end takes pictures. The conventional shooting trajectory is to shoot around the sample. Combined with the top view and the bottom view, after obtaining the sample image, the sample model can be constructed based on the sample image. Then, the surface of the sample model can be extended to obtain the surface map. This is a conventional function in existing 3D modeling software.

[0073] Step S200: Extract texture features from the surface map, divide the surface map into regions based on the texture feature extraction results, identify each region based on AI, and output the identification results;

[0074] By identifying the surface map, the textured area can be located. Features are extracted from the textured area, and the surface map is divided into multiple sub-regions based on the extracted features. Then, AI is applied to identify each sub-region, and the identification results can be output. The identification results are generally in text format and are descriptive text. Existing AI has extremely strong identification capabilities and the identification results are very rich.

[0075] Step S300: Analyze the zoning and identification results of the surface map for each sample, and create a sample sample;

[0076] After the surface map of each sample is divided into regions for identification, the identification results of each sub-region are statistically analyzed and used as a sample, called a sample sample.

[0077] Step S400: Compare sample samples, determine the order of information acquisition, obtain sample information, and construct a pattern gene library;

[0078] The sample samples are compared and clustered to group similar samples into one class. After the clustering is completed, the priority of each sample sample is determined, and the information acquisition order is determined accordingly. The subsequent sample information is acquired in sequence according to the information acquisition order, which takes a long time. After the sample information is acquired, the acquired sample information and the sample samples themselves are statistically analyzed to construct the pattern gene library.

[0079] Step S500: Generate guiding information for the design process based on the pattern gene library;

[0080] Once the pattern gene library is generated, it can guide the design process, obtain design information in real time, match similar patterns in the pattern gene library based on the design information, and provide feedback to the designer. The designer can then adjust the design process based on the existing patterns.

[0081] Regarding step S100, the steps of acquiring a sample image according to a preset shooting trajectory, constructing a sample model based on the sample image, and simultaneously performing two-dimensional extension on the surface of the three-dimensional model to obtain a surface map include:

[0082] An initial sample image is acquired based on a preset standard viewpoint, and a basic model of the sample is created based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the dimensions of the circumscribed cuboid of the basic model are preset values.

[0083] Obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the camera.

[0084] The sphere is divided into latitude and longitude sections according to a preset step size to obtain spherical sub-regions;

[0085] Connect the spherical sub-region and the center of the sphere to determine the detection direction, send the detection direction to the imaging end, and acquire the final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance;

[0086] Based on the final-state sample image correction model, a sample model is constructed.

[0087] A cross section is created based on the preset direction corresponding to the standard viewpoint, and the boundary line between the cross section and the sample model is obtained. The sample image surface is then extended in two dimensions based on one of the boundary lines to obtain a surface map.

[0088] The above describes the modeling process. Initial sample images are acquired based on a preset standard viewpoint, and a basic model of the sample is created based on these images. The standard viewpoint refers to the standard six views: front view, back view, left view, right view, bottom view, and top view. One of these six directions needs to be specified in advance, for example, facing due west is the front view, which is the preset direction mentioned above. The process of 3D modeling using the six views is not complicated. First, a 3D model (shape is not important) can be created, and then simulated images of the model in the six directions can be acquired. The simulated images are compared with the actual images, and continuous corrections are made to obtain the final basic model.

[0089] In the modeling software, the circumscribed cuboid of the base model is obtained. The center of the circumscribed cuboid is selected as the center of a sphere to create a sphere. The radius of the sphere is a preset value, determined by the shooting distance of the camera, and is usually set to a constant value. The spherical sub-region and the center of the sphere are connected to determine the detection direction. The detection direction is sent to the camera to obtain the final sample image containing the detection direction. At this time, the shooting trajectory is limited to different shooting directions (detection directions). The images obtained by the camera on these shooting trajectories are called final sample images. During acquisition, the detection direction needs to be recorded. The detection direction is the direction of the center line of the image. The distance of the detection direction is used to determine the shooting position of the image. Based on the scale of the base model, the final sample image can be overlaid on the base model to obtain the sample model.

[0090] The surface of the sample model is extended to obtain a surface map.

[0091] In a preferred embodiment of the technical solution of the present invention, the ambient light parameters are recorded during the process of the imaging end acquiring the final state sample image containing the detection direction; the color value is restored based on the ambient light parameters of the acquired final state sample image.

[0092] In one example of the technical solution of this invention, when capturing an image at the shooting end, on the one hand, the color value of each pixel in the image is recorded, and on the other hand, the ambient light parameters also need to be recorded. The ambient light parameters are equivalent to a filter. Based on the ambient light parameters, the color value of each pixel in the image is adjusted, which is the color value restoration process.

[0093] The process of constructing the sample model based on the final state sample image correction base model includes:

[0094] For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction;

[0095] The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

[0096] In one example of the technical solution of the present invention, the insertion process of the final sample image is described. For any final sample image, the detection direction is read, and the final sample image is inserted into the sample model based on the detection direction. The final sample image is wrapped around the sample model. Since there are many final sample images captured, each position may correspond to multiple pixels. The average value of all corresponding pixels is calculated as the color value of a certain point on the final sample model.

[0097] The encapsulation process is a standard procedure in existing 3D graphics processing software and will not be elaborated upon here.

[0098] Regarding step S200, the steps of extracting texture features from the surface image, dividing the surface image into regions based on the texture feature extraction results, identifying each region based on AI, and outputting the identification results include:

[0099] Iterate through each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel.

[0100] The boundaries of the decorative pattern are located based on the horizontal and vertical gradients;

[0101] Select boundary points on the boundary according to the preset step size, calculate the curvature at the boundary points, and mark the boundary points whose curvature is greater than the preset curvature threshold.

[0102] Cluster the marked boundary points, and partition the boundary points according to the clustering results;

[0103] Each region is identified using AI, and the identification results are output.

[0104] In one example of the technical solution of this invention, the process of identifying and extracting pattern features is described. The process involves traversing each pixel of the surface image, calculating the horizontal and vertical gradients at each pixel, and identifying the pixel as a pattern boundary when the horizontal or vertical gradient reaches a preset threshold. Boundary points are then selected on the boundary according to a preset step size, which simplifies the pattern boundary. For example, a point is selected every 10 pixels. The curvature at the boundary point is calculated, and when the curvature is sufficiently large (greater than the preset threshold), the boundary point is marked, indicating that it corresponds to a corner point (corresponding to a pattern). The greater the curvature, the more clustering is performed on the marked boundary points. Existing clustering algorithms can be used for clustering. The parameter for the clustering process is the distance between points. After clustering, for each type of point, a circular region is created with each point as the center and the curvature as the radius. At this time, one type of point corresponds to one type of circular region. The union of multiple circular regions is a whole irregular region, but each of its edges is the boundary of a circle, and the tangent can be obtained. Therefore, the circumscribed rectangle of this irregular region can be easily obtained. Each circumscribed rectangle corresponds to a sub-region, and multiple circumscribed matrices correspond to the partitioning results.

[0105] Regarding step S400, the steps of comparing sample samples, determining the information acquisition order, acquiring sample information, and constructing the pattern gene library include:

[0106] Compare the partitioning results of the sample samples and calculate the first distance between the sample samples;

[0107] When the first distance is less than a preset first distance threshold, compare the recognition results and calculate the second distance;

[0108] Samples whose second distance is less than the preset second distance threshold are grouped into one category;

[0109] Calculate the priority score for each sample, and use the descending order of the priority scores as the information acquisition order;

[0110] Based on the order of information acquisition, sample information acquisition requests are sent, and the partitioning results, identification results, and sample information of the acquired samples are statistically analyzed to construct a pattern gene library.

[0111] In one example of the technical solution of this invention, the construction process of the tattoo gene library is described. For any two samples, the partitioning results of the samples are compared, the degree of difference is calculated, and a first distance is obtained. When the first distance is small enough, it means that the two samples may be classified into the same category. At this time, the identification results are compared, the degree of difference is calculated, and a second distance is obtained. When the second distance is small enough, the two samples are considered to be in the same category. All samples are compared pairwise. After classification, the priority score of each sample is calculated. The priority score indicates the importance of the sample. The descending order of the priority scores is used as the information acquisition order. At this time, based on the information acquisition order, the sample information acquisition request is sent. The sample information of more important samples can be acquired first. The partitioning results, identification results and sample information of the acquired samples are statistically analyzed to construct the tattoo gene library.

[0112] It should be noted that the zoning results and recognition results have been explained above. The sample information is the information of a regular sample, including the type of object, shape characteristics, type of use, place of excavation and export region, etc. With the extremely high AI recognition capability, many parts of the sample information are included in the recognition results. Therefore, the sample information can be regarded as supplementary content to the recognition results.

[0113] As a preferred embodiment of the technical solution of the present invention, the method further includes:

[0114] The calculation process for the first distance is as follows:

[0115] Calculate the area of ​​each sub-region, and statistically analyze the area of ​​each sub-region based on its location to obtain an area matrix. Compare the area matrices to calculate the first distance.

[0116] The calculation process for the second distance is as follows:

[0117] The recognition results are converted into a word library. The intersection and union of the word libraries of the two sample samples are calculated. The intersection-union ratio is calculated, and the second distance is determined based on the inverse ratio of the intersection-union ratio.

[0118] The comparison process of the area matrix is ​​not complicated. Calculate the data difference at each row and column position, and then accumulate the data differences to obtain the first distance. Since the partitioning process is different, the size of the matrix is ​​different. However, each matrix has a first row and a first column. For positions with fewer rows or columns, there is data in the area matrix, and the data difference is easy to calculate. However, as the number of rows and columns increases, there will be positions without data. At this time, you can directly use the maximum difference, or you can fill the position with the default area and then calculate. Both of these are feasible. The key is that there needs to be a filling process.

[0119] The priority score is calculated as follows:

[0120] Obtain the total number of samples in the class to which the sample belongs, and the number of samples in the class to which the sample belongs whose sample information has been obtained. Use the total number of samples as an intermediate indicator and the number of samples as an inverse indicator to calculate the priority score together.

[0121] In the calculation of priority scores, the total number of samples in the class to which the sample belongs is obtained. The closer the total number of samples is to a preset threshold, the higher the priority score. The threshold is preset by the staff and is generally an intermediate value. Therefore, if there are too many samples of the same class, it means that the sample may be relatively common and does not need to be analyzed first. If there are too few samples of the same class, it means that the sample is relatively niche or needs to be analyzed manually. This is beyond the application scope of the technical solution of this invention. Therefore, the best approach is that the closer the sample is to a certain intermediate value, the more necessary it is to obtain sample information first and the higher the priority score. That is, the total number of samples is an intermediate indicator.

[0122] Similarly, when the number of samples in the same class that have obtained sample information is larger, it means that the sample information of the same class has been obtained, and the remaining sample information can be obtained without rushing to obtain sample information. At this time, the priority score becomes lower. That is, the number of samples in the same class that have obtained sample information is a reverse indicator. The reverse indicator corresponds to the positive indicator. The positive indicator is better the larger it is, and the reverse indicator is better the smaller it is.

[0123] Figure 2A structural diagram of a Yuan blue-and-white porcelain pattern gene bank construction and AI-derived design system is shown. In a preferred embodiment of the technical solution of the present invention, a Yuan blue-and-white porcelain pattern gene bank construction and AI-derived design system is also provided, the system 10 comprising:

[0124] Modeling module 11 is used to acquire sample images according to a preset shooting trajectory, construct a sample model based on the sample images, and simultaneously extend the surface of the three-dimensional model in two dimensions to obtain a surface map.

[0125] The pattern extraction and recognition module 12 is used to extract pattern features from the surface map, divide the surface map into regions based on the pattern feature extraction results, recognize each region based on AI, and output the recognition results.

[0126] The sample creation module 13 is used to statistically analyze the zoning and identification results of the surface map of each sample and create sample data.

[0127] Gene bank construction module 14 is used to compare sample samples, determine the information acquisition order, acquire sample information, and construct a pattern gene bank.

[0128] The guidance information generation module 15 is used to generate guidance information for the design process based on the pattern gene library.

[0129] Furthermore, the modeling module 11 includes:

[0130] The basic model creation unit is used to acquire an initial sample image based on a preset standard viewpoint and create a basic model of the sample based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the size of the circumscribed cuboid of the basic model is a preset value.

[0131] The sphere creation unit is used to obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the shooting end;

[0132] The spherical segmentation unit is used to segment the sphere according to the latitude and longitude according to the preset step size to obtain spherical sub-regions;

[0133] The image acquisition unit is used to connect the spherical sub-region and the center of the sphere, determine the detection direction, send the detection direction to the imaging end, and acquire a final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance;

[0134] The sample model building unit is used to construct a sample model based on the final state sample image and the correction of the base model.

[0135] The extension unit is used to create a cross section based on a preset direction corresponding to the standard viewpoint, obtain the boundary line between the cross section and the sample model, and perform two-dimensional extension on the sample image surface based on one of the boundary lines to obtain a surface map.

[0136] Specifically, the ambient light parameters are recorded during the process of acquiring the final state sample image containing the detection direction at the camera end; color value restoration is performed on the acquired final state sample image based on the ambient light parameters;

[0137] The process of constructing the sample model based on the final state sample image correction base model includes:

[0138] For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction;

[0139] The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

[0140] Furthermore, the pattern extraction and recognition module 12 includes:

[0141] The gradient calculation unit is used to traverse each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel.

[0142] Boundary positioning unit, used to locate the decorative boundary according to the horizontal and vertical gradients;

[0143] The boundary point marking unit is used to select boundary points on the boundary according to a preset step size, calculate the curvature at the boundary points, and mark boundary points whose curvature is greater than a preset curvature threshold.

[0144] Clustering units are used to cluster the marked boundary points and partition the boundary points based on the clustering results;

[0145] The recognition execution unit is used to identify each region based on AI and output the recognition results.

[0146] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for constructing a gene library of Yuan blue-and-white porcelain patterns and for AI-derived design, characterized in that, The method includes: The sample image is acquired according to the preset shooting trajectory, the sample model is constructed according to the sample image, and the surface of the three-dimensional model is simultaneously extended in two dimensions to obtain the surface map. The surface map is subjected to texture feature extraction. Based on the texture feature extraction results, the surface map is divided into regions. Each region is identified based on AI, and the identification results are output. The zoning and identification results of the surface map for each sample are statistically analyzed to create a sample sample. By comparing sample samples, determining the order of information acquisition, obtaining sample information, and constructing a pattern gene library; Guiding information for generating the design process based on the pattern gene library; The steps of acquiring sample images according to a preset shooting trajectory, constructing a sample model based on the sample images, and simultaneously performing two-dimensional extension on the surface of the three-dimensional model to obtain a surface map include: An initial sample image is acquired based on a preset standard viewpoint, and a basic model of the sample is created based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the dimensions of the circumscribed cuboid of the basic model are preset values. Obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the camera. The sphere is divided into latitude and longitude sections according to a preset step size to obtain spherical sub-regions; Connect the spherical sub-region and the center of the sphere to determine the detection direction, send the detection direction to the imaging end, and acquire the final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance; Based on the final-state sample image correction model, a sample model is constructed. A cross section is created based on the preset direction corresponding to the standard viewpoint, and the boundary line between the cross section and the sample model is obtained. The sample image surface is then extended in two dimensions based on one of the boundary lines to obtain a surface map.

2. The method for constructing a Yuan blue-and-white porcelain pattern gene library and designing AI derivatives according to claim 1, characterized in that, Ambient light parameters are recorded during the process of acquiring a final sample image containing the detection direction at the camera end; Color value restoration is performed on the final state sample image obtained based on ambient light parameters; The process of constructing the sample model based on the final state sample image correction base model includes: For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction; The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

3. The method for constructing a Yuan blue-and-white porcelain pattern gene library and designing AI derivatives according to claim 2, characterized in that, The steps of extracting texture features from the surface image, dividing the surface image into regions based on the extracted texture features, identifying each region based on AI, and outputting the identification results include: Iterate through each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel. The boundaries of the decorative pattern are located based on the horizontal and vertical gradients; Select boundary points on the boundary according to the preset step size, calculate the curvature at the boundary points, and mark the boundary points whose curvature is greater than the preset curvature threshold. Cluster the marked boundary points, and partition the boundary points according to the clustering results; Each region is identified using AI, and the identification results are output.

4. The method for constructing a Yuan blue-and-white porcelain pattern gene bank and designing AI derivatives according to claim 1, characterized in that, The steps of comparing sample samples, determining the order of information acquisition, acquiring sample information, and constructing a pattern gene library include: Compare the partitioning results of the sample samples and calculate the first distance between the sample samples; When the first distance is less than a preset first distance threshold, compare the recognition results and calculate the second distance; Samples whose second distance is less than the preset second distance threshold are grouped into one category; Calculate the priority score for each sample, and use the descending order of the priority scores as the information acquisition order; Based on the order of information acquisition, sample information acquisition requests are sent, and the partitioning results, identification results, and sample information of the acquired samples are statistically analyzed to construct a pattern gene library.

5. The method for constructing a Yuan blue-and-white porcelain pattern gene library and designing AI derivatives according to claim 4, characterized in that, The method further includes: The calculation process for the first distance is as follows: Calculate the area of ​​each sub-region, and statistically analyze the area of ​​each sub-region based on its location to obtain an area matrix. Compare the area matrices to calculate the first distance. The calculation process for the second distance is as follows: The recognition results are converted into a word library. The intersection and union of the word libraries of the two samples are calculated. The intersection-union ratio is calculated. The second distance is determined based on the inverse ratio of the intersection-union ratio. The priority score is calculated as follows: Obtain the total number of samples in the class to which the sample belongs, and the number of samples in the class to which the sample belongs whose sample information has been obtained. Use the total number of samples as an intermediate indicator and the number of samples as an inverse indicator to calculate the priority score together.

6. A system for constructing a gene bank of Yuan blue-and-white porcelain decorative patterns and for AI-derived design, characterized in that, The system includes: The modeling module is used to acquire sample images according to a preset shooting trajectory, construct a sample model based on the sample images, and simultaneously extend the surface of the three-dimensional model in two dimensions to obtain a surface map. The pattern extraction and recognition module is used to extract pattern features from the surface map, divide the surface map into regions based on the pattern feature extraction results, recognize each region based on AI, and output the recognition results. The sample creation module is used to statistically analyze the zoning and identification results of the surface map of each sample and create the sample. The gene bank construction module is used to compare sample samples, determine the order of information acquisition, obtain sample information, and construct a pattern gene bank. The guidance information generation module is used to generate guidance information for the design process based on the pattern gene library; The modeling module includes: The basic model creation unit is used to acquire an initial sample image based on a preset standard viewpoint and create a basic model of the sample based on the initial sample image; the standard viewpoint is a six-view viewpoint based on a preset direction; the size of the circumscribed cuboid of the basic model is a preset value. The sphere creation unit is used to obtain the circumscribed cuboid of the base model, select the center of the circumscribed cuboid as the center of the sphere, and create a sphere; the radius of the sphere is a preset value, determined by the shooting distance of the shooting end; The spherical segmentation unit is used to segment the sphere according to the latitude and longitude according to the preset step size to obtain spherical sub-regions; The image acquisition unit is used to connect the spherical sub-region and the center of the sphere, determine the detection direction, send the detection direction to the imaging end, and acquire a final sample image containing the detection direction; the detection direction is a vector, the direction of the vector is the direction of the center line of the imaging end, and the magnitude of the vector represents the imaging distance; The sample model building unit is used to construct a sample model based on the final state sample image and the correction of the base model. The extension unit is used to create a cross section based on a preset direction corresponding to the standard viewpoint, obtain the boundary line between the cross section and the sample model, and perform two-dimensional extension on the sample image surface based on one of the boundary lines to obtain a surface map.

7. The Yuan blue-and-white porcelain pattern gene bank construction and AI-derived design system according to claim 6, characterized in that, Ambient light parameters are recorded during the process of acquiring a final sample image containing the detection direction at the camera end; Color value restoration is performed on the final state sample image obtained based on ambient light parameters; The process of constructing the sample model based on the final state sample image correction base model includes: For any final state sample image, read the detection direction and insert the final state sample image into the sample model based on the detection direction; The final sample image is overlaid onto the sample model; where, for any point on the sample model, if there are multiple overlay results for that point, the average of all overlay results is taken.

8. The Yuan blue-and-white porcelain pattern gene bank construction and AI-derived design system according to claim 7, characterized in that, The pattern extraction and recognition module includes: The gradient calculation unit is used to traverse each pixel of the surface map and calculate the horizontal and vertical gradients at each pixel. Boundary positioning unit, used to locate the decorative boundary according to the horizontal and vertical gradients; The boundary point marking unit is used to select boundary points on the boundary according to a preset step size, calculate the curvature at the boundary points, and mark boundary points whose curvature is greater than a preset curvature threshold. Clustering units are used to cluster the marked boundary points and partition the boundary points based on the clustering results; The recognition execution unit is used to identify each region based on AI and output the recognition results.

Images