CN117197000A - Quick grid denoising method and device and electronic equipment - Google Patents
Quick grid denoising method and device and electronic equipment Download PDFInfo
- Publication number
- CN117197000A CN117197000A CN202311461903.5A CN202311461903A CN117197000A CN 117197000 A CN117197000 A CN 117197000A CN 202311461903 A CN202311461903 A CN 202311461903A CN 117197000 A CN117197000 A CN 117197000A
- Authority
- CN
- China
- Prior art keywords
- characteristic
- patches
- correction information
- patch
- grid
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000012937 correction Methods 0.000 claims abstract description 144
- 238000013135 deep learning Methods 0.000 claims abstract description 26
- 238000004364 calculation method Methods 0.000 claims abstract description 18
- 238000000547 structure data Methods 0.000 claims description 18
- 238000011176 pooling Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- 230000004927 fusion Effects 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 5
- 230000002093 peripheral effect Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000007796 conventional method Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000011960 computer-aided design Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a rapid grid denoising method, a device and electronic equipment, comprising the following steps: acquiring grid data; carrying out characteristic surface patch calculation on the grid data to obtain the surface variation degree of each surface patch in the grid data, setting a surface variation degree threshold, marking the surface patch higher than the surface variation degree threshold as a characteristic surface patch, and taking other surface patches as non-characteristic surface patches; performing normal correction on the characteristic patches and the neighborhood patches based on a deep learning method to obtain characteristic correction information, performing normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information; and correcting and denoising the grid data according to the normal correction information. In conclusion, the invention realizes the accurate correction of the characteristic patches by a deep learning method, realizes the rapid correction of the non-characteristic patches by a traditional method, effectively reduces the calculated amount on the premise of ensuring the correction precision, and realizes the rapid and accurate denoising of the grid data.
Description
Technical Field
The invention relates to the field of computer images, in particular to a method and a device for removing noise from a quick grid and electronic equipment.
Background
In the fields of computer aided design, engineering application and the like, three-dimensional data acquired by a scanner or other equipment are required to be reconstructed and processed, and noise is usually contained in the reconstructed three-dimensional grid data, so that the accuracy of modeling and analysis results is reduced. Therefore, the denoising processing of the grid data is very important, the data quality can be effectively improved, and the accuracy of modeling and analysis results can be improved.
The current widely used grid denoising method is divided into two steps, namely, correcting the normal vector of the grid surface patch, and updating the vertex position according to the corrected normal vector. Conventional methods attempt to maintain grid feature information by using various anisotropic operators, but the conventional methods rely on complex parameter settings for denoising effects, which are general; the deep learning method can effectively solve various complex parameter adjustment problems, and has good denoising effect and characteristic maintaining effect, but because of complex network framework, a large amount of hardware resources are required in the calculation process, and the calculation time cost is too large to meet the industrial efficiency requirement. Therefore, in the prior art, both the traditional method and the deep learning method are difficult to realize rapid and accurate grid denoising, and improvement is needed.
Disclosure of Invention
In view of the foregoing, it is necessary to provide a fast grid denoising method, apparatus and electronic device, which are used for solving the technical problem that the conventional method and the deep learning method in the prior art are difficult to implement fast and accurate grid denoising.
In order to solve the above problems, in one aspect, the present invention provides a fast grid denoising method, including:
acquiring grid data, wherein the grid data comprises a plurality of patches;
carrying out characteristic surface patch calculation on the grid data to obtain the surface variation degree of each surface patch in the grid data, setting a surface variation degree threshold value, marking the surface patch with the surface variation degree higher than the surface variation degree threshold value as a characteristic surface patch, and taking other surface patches as non-characteristic surface patches;
performing normal correction on the characteristic patches and the neighborhood patches of the characteristic patches based on a deep learning method to obtain characteristic correction information, performing normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information;
and correcting and denoising the grid data according to the normal correction information.
Further, the grid data further includes a plurality of vertices, and the computing of the feature patches for the grid data to obtain the surface variability of each patch in the grid data includes:
constructing the topological relation between the vertexes of the grid data and the patches, and determining the neighborhood relation between the vertexes and the patches according to the topological relation;
calculating the surface information of the vertex and the surface patch according to the neighborhood relation of the vertex and the surface patch;
the degree of surface variation of each patch is determined from the surface information.
Further, performing normal correction on the feature patches and the neighborhood patches of the feature patches based on a deep learning method to obtain feature correction information, including:
constructing a graph neural network, wherein the graph neural network comprises a feature extraction module, a pooling layer and a correction information output module;
performing topology reconstruction on the neighborhood patches of the characteristic patches to obtain a reconstructed topology relationship;
constructing graph structure data according to the reconstruction topological relation;
the method comprises the steps of inputting graph structure data into a graph neural network, carrying out feature extraction on the graph structure data based on a feature extraction module to obtain graph convolution features, carrying out cascade weighting on the graph convolution features based on a pooling layer to obtain cascade features, and carrying out network inference on the cascade features based on a correction information output module to obtain feature correction information.
Further, the reconstruction topology includes a relationship of the face sheet and the vertex, a relationship of the edge and the vertex, and a relationship of the vertex and the face sheet.
Further, constructing graph structure data according to the reconstructed topological relation includes:
and constructing graph structure data by taking the characteristic patches and the neighborhood patches of the characteristic patches as nodes of the graph structure and taking inter-patch connection as edges of the graph structure, wherein for each patch, all patches in the neighborhood where the patch is located share one vertex.
Further, cascade weighting is performed on the graph convolution feature based on the pooling layer to obtain a cascade feature, which comprises the following steps:
obtaining the geometric weight and the feature weight of the graph convolution feature, and combining the geometric weight and the feature weight to obtain a cascade edge weight;
and carrying out cascade weighting on the graph convolution characteristics according to the cascade edge weight to obtain cascade characteristics.
Further, fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information, including:
marking the most peripheral neighborhood surface patches in the neighborhood surface patches of the characteristic surface patches as surface patches to be fused;
and carrying out normal fusion on the characteristic correction information and the non-characteristic correction information of the to-be-fused patch to obtain fusion correction information, and splicing and fusing the characteristic correction information and the non-characteristic correction information according to the fusion correction information to obtain normal correction information.
Further, the correction denoising processing is performed on the grid data according to the normal correction information, including:
calculating a coordinate residual error according to the normal correction information;
and updating the position information of the vertex according to the coordinate residual error.
In another aspect, the present invention provides a fast grid denoising apparatus, comprising:
the grid data acquisition unit is used for acquiring grid data, and the grid data comprises a plurality of patches;
the characteristic patch marking unit is used for carrying out characteristic patch calculation on the grid data to obtain the surface variation degree of each patch in the grid data, setting a surface variation degree threshold value, marking patches with the surface variation degree higher than the surface variation degree threshold value as characteristic patches, and other patches as non-characteristic patches;
the normal correction unit is used for carrying out normal correction on the characteristic patches and the neighborhood patches of the characteristic patches based on a deep learning method to obtain characteristic correction information, carrying out normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information;
and the correction denoising unit is used for performing correction denoising processing on the grid data according to the normal correction information.
In another aspect, the invention provides an electronic device comprising a memory and a processor, wherein,
a memory for storing a computer program;
a processor coupled to the memory for executing a computer program to implement the steps in the fast grid denoising method of any one of the above.
Compared with the prior art, the beneficial effects of adopting the embodiment are as follows: according to the invention, the surface patches are divided into the characteristic surface patches and the non-characteristic surface patches according to the surface variation, the accurate correction of the characteristic surface patches is realized by a deep learning method, the rapid correction of the non-characteristic surface patches is realized by a traditional method, the data quantity calculated by a network is effectively reduced on the premise of ensuring the correction precision, and the rapid and accurate denoising of the grid data is realized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being evident that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of an embodiment of a fast grid denoising method according to the present invention;
FIG. 2 is a schematic diagram illustrating the structure of an embodiment of a fast grid denoising apparatus according to the present invention;
fig. 3 is a schematic structural diagram of an embodiment of an electronic device provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be understood that the drawings of the schematic drawings are not drawn to scale. A flowchart, as used in this disclosure, illustrates operations implemented according to some embodiments of the present invention. It should be appreciated that the operations of the flow diagrams may be implemented out of order and that steps without logical context may be performed in reverse order or concurrently. Moreover, one or more other operations may be added to or removed from the flow diagrams by those skilled in the art under the direction of the present disclosure.
Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor systems and/or microcontroller systems.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
Fig. 1 is a flow chart of an embodiment of a fast grid denoising method according to the present invention, where, as shown in fig. 1, the fast grid denoising method includes:
s101, acquiring grid data, wherein the grid data comprises a plurality of patches;
s102, calculating characteristic patches of the grid data to obtain the surface variation degree of each patch in the grid data, setting a surface variation degree threshold, marking patches with the surface variation degree higher than the surface variation degree threshold as characteristic patches, and taking other patches as non-characteristic patches;
s103, carrying out normal correction on the characteristic patches and the neighborhood patches of the characteristic patches based on a deep learning method to obtain characteristic correction information, carrying out normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information;
s104, correcting and denoising the grid data according to the normal correction information.
Specifically, the invention divides the surface patch into the characteristic surface patch and the non-characteristic surface patch according to the surface variation, realizes the accurate correction of the characteristic surface patch by a deep learning method, realizes the rapid correction of the non-characteristic surface patch by a traditional method, effectively reduces the data volume of network calculation on the premise of ensuring the correction precision, and realizes the rapid and accurate denoising of the grid data.
In a specific embodiment of the present invention, the mesh data further includes a plurality of vertices, and performing feature patch calculation on the mesh data to obtain a surface variation degree of each patch in the mesh data, including:
constructing the topological relation between the vertexes of the grid data and the patches, and determining the neighborhood relation between the vertexes and the patches according to the topological relation;
calculating the surface information of the vertex and the surface patch according to the neighborhood relation of the vertex and the surface patch;
the degree of surface variation of each patch is determined from the surface information.
Specifically, in order to implement rapid and accurate denoising of the grid data, the grid data needs to be classified first, a deep learning method is used for the noise points with larger errors so as to ensure the denoising accuracy, and a traditional method is used for the noise points with smaller errors so as to implement rapid denoising. In the process of calculating the surface variation degree of the surface patch according to the embodiment, firstly, a topological relation is constructed for the vertex of the grid data and the surface patch, the neighborhood relation of the vertex and the surface patch can be quickly obtained according to the topological relation, and the surface information is calculated according to the neighborhood relation, wherein the surface information comprises but is not limited to curvature and the like, and the curvature is taken as an example in the embodiment.
There are many methods of calculating curvature, but they are susceptible to the dimensions of the mesh, i.e. as the resolution of the mesh changes, so does the resulting surface curvature. Therefore, the embodiment uses a surface change degree calculation method insensitive to the grid scale, and uses the product of the average value of the edge length connected by the vertex and the curvature calculated by the traditional method as a measure for the degree of the grid surface change, and the formula is as follows:
wherein the method comprises the steps ofRepresenting vertices in the mesh +_>Representing the degree of surface variation of the grid data,/->Curvature information calculated for the conventional method, +.>Representing the average of the edge lengths of the edges to which the vertices are attached.
After the surface variation degree is calculated, the embodiment marks the characteristic state of the surface patch with the surface variation degree higher than the surface variation degree threshold according to the preset surface variation degree threshold to obtain the characteristic surface patch, and other surface patches are used as non-characteristic surface patches. In addition, in order to improve the grid denoising effect and ensure the overall consistency of the grid data denoising result, the embodiment in the subsequent process jointly selects marked characteristic patches and neighborhood patches thereof to be transmitted into a deep learning network for normal correction.
In particular embodiments of the present invention, the reconstruction topology includes a relationship of a patch and a vertex, a relationship of an edge and a vertex, and a relationship of a vertex and a patch.
Specifically, to meet the input requirement of the designed deep learning network framework, topology reconstruction is required to be performed on the selected feature patches and the neighborhood patches, that is, each part of patches is converted into a new small grid model. For one grid data, the data information required after reconstruction should include the center three-dimensional coordinates of the surface patch, the normal vectors of the surface patch, the three-dimensional coordinates of the vertex and the normal vectors of the vertex, and the topological relation required after reconstruction should include the connection relation between the surface patch and the vertex, the connection relation between the edge and the vertex and the connection relation between the vertex and the surface patch.
In the embodiment, taking a triangular mesh as an example, the relationship between the surface patch and the vertex includes a correspondence relationship between the index ordering of the surface patch and the index ordering of the corresponding vertex. The embodiment firstly traverses the surface sheets, sequences from zero according to the access sequence, establishes access labels for the vertexes at the same time, then traverses the vertexes, does not change the access labels of the vertexes which have been accessed in the previous step, and adds one to the index sequence according to the sequence of the vertexes which have not been accessed.
To ensure unification of index relationships, the relationships of edges and vertices are calculated based on the relationships between the faces and vertices. The relationship between an edge and a vertex can be understood as the index ordering of the edge and the index ordering of two corresponding vertices, the edge is provided with an access label, the edges are ordered according to the access order, each surface piece has three edges, and if the edge is not accessed, the vertex connected with the edge can be compared with the vertex of the surface piece to obtain the connection relationship between the edge and the vertex because the relationship between the surface piece and the vertex is known.
The relationship between a vertex and a patch can be understood as the index ordering of the vertex and the index ordering of the patch to which it is connected. Unlike the first two relations, the relation between the vertices and the faces is uncertain, and the number of faces to which each vertex is connected cannot be determined in advance, so that calculation is also performed on the basis of the relation between the faces and the vertices. The connection relationship between the surface pieces and the vertex is rearranged according to the sequence of the vertex indexes, and the surface pieces containing the same vertex index are arranged together to form the connection relationship between the vertex and the surface pieces.
In a specific embodiment of the present invention, reconstructing topological relation construction graph structure data includes:
and constructing graph structure data by taking the characteristic patches and the neighborhood patches of the characteristic patches as nodes of the graph structure and taking inter-patch connection as edges of the graph structure, wherein for each patch, all patches in the neighborhood where the patch is located share one vertex.
Specifically, after the topology reconstruction of the feature patches and the neighborhood patches is completed, the graph structure data is constructed according to the reconstruction result to serve as the input of the deep learning network. Wherein each patch serves as a node of the graph structure data, the connection of the center of each patch to its neighborhood serves as an edge of the graph structure data, each patch is connected to all faces where vertices of its neighborhood are located, i.e. in a neighborhood, all faces share a vertex, such a design provides a balance between local geometry analysis and computational efficiency.
In a specific embodiment of the present invention, performing normal correction on a feature patch and a neighborhood patch of the feature patch based on a deep learning method to obtain feature correction information includes:
constructing a graph neural network, wherein the graph neural network comprises a feature extraction module, a pooling layer and a correction information output module;
performing topology reconstruction on the neighborhood patches of the characteristic patches to obtain a reconstructed topology relationship;
constructing graph structure data according to the reconstruction topological relation;
the method comprises the steps of inputting graph structure data into a graph neural network, carrying out feature extraction on the graph structure data based on a feature extraction module to obtain graph convolution features, carrying out cascade weighting on the graph convolution features based on a pooling layer to obtain cascade features, and carrying out network inference on the cascade features based on a correction information output module to obtain feature correction information.
In a specific embodiment of the present invention, cascade weighting is performed on graph convolution features based on a pooling layer to obtain cascade features, including:
obtaining the geometric weight and the feature weight of the graph convolution feature, and combining the geometric weight and the feature weight to obtain a cascade edge weight;
and carrying out cascade weighting on the graph convolution characteristics according to the cascade edge weight to obtain cascade characteristics.
Specifically, the embodiment uses a graph neural network (GNN, graph Neural Network) as a network framework for deep learning normal correction, the feature extraction module is partially designed into a three-layer U-Net architecture to gradually extract grid multi-scale information from input graph data, and the pooling layer is designed with a cascading weight estimation strategy to improve the robustness and denoising effect of the method.
In an embodiment, the graph neural network is constructed based on a graph structure, and potential mapping relations are learned in a space and normal domain. The structural properties of a graph are generally described in terms of degrees of vertices, where degrees of vertices refer to the number of edges in the graph that are connected to the vertices, and for a graph structure made up of three-dimensional irregular data, the degrees of each point are not fixed, thus providing an allocation functionAnd soft connection is established between the input data and the weight matrix, so that the number of the weight matrix is not influenced by the input data, and the robustness is realized.
In the space diagram convolution of three-dimensional shape analysis, each node in a local area is assigned a fixed numberThe weighted sum of the individual feature transformations learns the information aggregation. Arbitrary node of the graph->The explicit form of the graph convolution is:
wherein,for deviation term->For node->Is a neighborhood node set, ">Representing node->Is a neighborhood node of%>Is->Layer feature transform weight matrix, < >>The allocation function is expressed as:
wherein,、/>and->Is a parameter of the linear transformation.
In addition, after feature extraction, the embodiment also enlarges the receptive field by designing a pooling layer to increase the depth of the network. The pooling layer adopts a cascade edge weight estimation scheme, and comprehensively considers the geometric weight and the characteristic weight of the grid; for the firstThe layer is pooled, and the side weight is determined by the following formula:
wherein,representing the edge weight of the current layer calculation, +.>Edge weight delivered for the previous pooling layer,/->And->Respectively represent node->And node->Is described.
The cascade edge weight estimation scheme designed by the embodiment comprehensively considers the geometric weight and the characteristic weight, wherein the geometric weight is beneficial to distributing larger weight to the node pairs with similar normals and smaller Euclidean distance, and the characteristic weight tends to endow larger weight to the nodes with similar network characteristics. The embodiment integrates the advantages of the geometric weight and the feature weight, and propagates the edge weight from the fine graph to the coarsening graph, so that the network can learn the global feature and the shallow feature.
And finally, completing network inference of the correction information through the correction information output module and outputting characteristic correction information.
In a specific embodiment of the present invention, fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information includes:
marking the most peripheral neighborhood surface patches in the neighborhood surface patches of the characteristic surface patches as surface patches to be fused;
and carrying out normal fusion on the characteristic correction information and the non-characteristic correction information of the to-be-fused patch to obtain fusion correction information, and splicing and fusing the characteristic correction information and the non-characteristic correction information according to the fusion correction information to obtain normal correction information.
Specifically, after the normal correction information of the feature patch is pushed out by the deep learning method, the normal correction operation is then performed on the non-feature patch by using a conventional method. The traditional method can be realized in parallel through multithreading, GPU and the like to improve the calculation speed, and specific implementation modes of the traditional method include, but are not limited to, a mean value filtering method, a Laplace filtering method, a normal bilateral filtering method and the like.
And then when feature correction information obtained by deep learning and non-feature correction information obtained by a traditional method are fused, the fact that obvious splicing marks exist in the connected areas of the feature patches and the non-feature correction information due to direct normal information accumulation is considered, and the visual effect is not ideal is considered, so that the embodiment marks the outermost neighborhood patches in the feature patches and the neighborhood patches thereof as patches to be fused. In the previous calculation process, the part participates in the deep learning method calculation and the traditional method calculation, so that the normal fusion of the characteristic correction information and the non-characteristic correction information corresponding to the part can be carried out by adopting a mean value method or setting a weight coefficient, the normal after fusion is normalized, and finally the result is spliced to obtain the integral normal correction information.
In a specific embodiment of the present invention, performing correction denoising processing on mesh data according to normal correction information includes:
calculating a coordinate residual error according to the normal correction information;
and updating the position information of the vertex according to the coordinate residual error.
Specifically, after the normal correction information is obtained, the position information of the grid vertexes is updated under the constraint of the correct normal information, and an iteration scheme is used in the process to realize vertex position updating:
at each iteration process set, each vertexNew position +.>Can be expressed as:
wherein the method comprises the steps ofThe coordinate residual error calculated according to the corrected correct normal information of the patch is as follows:
wherein,is vertex->All patches connected, ->And->The centroid of the patch and the corrected normal are shown, respectively.
Compared with the traditional method, the method can obtain better denoising effect, and meanwhile, compared with the deep learning method, the method greatly improves the grid denoising efficiency. Compared with the prior art, the method adopts a characteristic self-adaptive strategy for grid denoising, divides the surface patch into the characteristic surface patch with large surface variation and the non-characteristic surface patch with small surface variation according to the surface variation, realizes accurate correction of the characteristic surface patch by a deep learning method, realizes quick correction of the non-characteristic surface patch by a traditional method, and finally uses corrected normal information as constraint to finish grid vertex position update. The method and the device for denoising the grid data smoothly reduce the characteristic loss and the time cost of grid denoising and improve the precision and the visual effect of the grid model.
Based on the fast grid denoising method provided by the present invention, the present invention also provides a fast grid denoising apparatus 200, as shown in fig. 2, comprising:
a grid data acquisition unit 201, configured to acquire grid data, where the grid data includes a plurality of patches;
a feature patch marking unit 202, configured to perform feature patch calculation on the grid data to obtain a surface variation degree of each patch in the grid data, set a surface variation degree threshold, mark patches with a surface variation degree higher than the surface variation degree threshold as feature patches, and use other patches as non-feature patches;
the normal correction unit 203 is configured to perform normal correction on the feature patch and a neighborhood patch of the feature patch based on a deep learning method to obtain feature correction information, perform normal correction on the non-feature patch based on a conventional method to obtain non-feature correction information, and fuse the feature correction information and the non-feature correction information to obtain normal correction information;
and a correction denoising unit 204 for performing correction denoising processing on the grid data according to the normal correction information.
The fast grid denoising apparatus 200 provided in the foregoing embodiment may implement the technical solution in the foregoing fast grid denoising method embodiment, and the specific implementation principle of each module or unit may refer to the corresponding content in the foregoing fast grid denoising method embodiment, which is not described herein again.
The present invention also provides an electronic device 300, as shown in fig. 3, fig. 3 is a schematic structural diagram of an embodiment of the electronic device provided by the present invention, where the electronic device 300 includes a processor 301, a memory 302, and a computer program stored in the memory 302 and capable of running on the processor 301, and when the processor 301 executes the program, the above-mentioned fast grid denoising method is implemented.
As a preferred embodiment, the electronic device further comprises a display 303 for displaying the process of the fast grid denoising method performed by the processor 301.
The processor 301 may be an integrated circuit chip, and has signal processing capability. The processor 301 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; also digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC). The disclosed methods, steps, and logic blocks in the embodiments of the present invention may be implemented or performed. A general purpose processor may also be a microprocessor or the processor may be any conventional processor or the like.
The Memory 302 may be, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read Only Memory (ROM), a Secure Digital (SD Card), a Flash Card (Flash Card), etc. The memory 302 is configured to store a program, and the processor 301 executes the program after receiving an execution instruction, and the method for defining a flow disclosed in any of the foregoing embodiments of the present invention may be applied to the processor 301 or implemented by the processor 301.
The display 303 may be an LED display, a liquid crystal display, a touch display, or the like. The display 303 is used to display various information on the electronic device 300.
It is to be understood that the configuration shown in fig. 3 is merely a schematic diagram of one configuration of the electronic device 300, and that the electronic device 300 may also include more or fewer components than those shown in fig. 3. The components shown in fig. 3 may be implemented in hardware, software, or a combination thereof.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (10)
1. A method of fast grid denoising, the method comprising:
acquiring grid data, wherein the grid data comprises a plurality of patches;
performing characteristic surface patch calculation on the grid data to obtain the surface variation degree of each surface patch in the grid data, setting a surface variation degree threshold, marking the surface patch with the surface variation degree higher than the surface variation degree threshold as a characteristic surface patch, and taking other surface patches as non-characteristic surface patches;
performing normal correction on the characteristic patches and the neighborhood patches of the characteristic patches based on a deep learning method to obtain characteristic correction information, performing normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information;
and correcting and denoising the grid data according to the normal correction information.
2. The method for fast grid denoising according to claim 1, wherein the grid data further comprises a plurality of vertices, and performing feature patch calculation on the grid data to obtain a surface variation degree of each patch in the grid data comprises:
constructing a topological relation between vertexes and patches of the grid data, and determining a neighborhood relation between the vertexes and the patches according to the topological relation;
calculating the surface information of the vertex and the surface patch according to the neighborhood relation of the vertex and the surface patch;
and determining the surface variation degree of each patch according to the surface information.
3. The method of fast grid denoising according to claim 1, wherein the performing normal correction on the feature patch and the neighborhood patch of the feature patch based on the deep learning method to obtain feature correction information comprises:
constructing a graph neural network, wherein the graph neural network comprises a feature extraction module, a pooling layer and a correction information output module;
performing topology reconstruction on the neighborhood patches of the characteristic patches to obtain a reconstructed topology relationship;
constructing graph structure data according to the reconstruction topological relation;
inputting the graph structure data into the graph neural network, carrying out feature extraction on the graph structure data based on a feature extraction module to obtain graph convolution features, carrying out cascade weighting on the graph convolution features based on a pooling layer to obtain cascade features, and carrying out network inference on the cascade features based on a correction information output module to obtain feature correction information.
4. A fast grid denoising method according to claim 3, wherein the reconstructed topological relationship comprises a relationship of a face sheet and a vertex, a relationship of an edge and a vertex, and a relationship of a vertex and a face sheet.
5. A fast grid denoising method according to claim 3, wherein the constructing graph structure data from the reconstructed topological relation comprises:
and constructing graph structure data by taking the characteristic patches and the neighborhood patches of the characteristic patches as nodes of a graph structure and taking inter-patch connection as edges of the graph structure, wherein for each patch, all patches in the neighborhood where the patch is located share one vertex.
6. A fast grid denoising method according to claim 3, wherein cascade weighting the graph convolution feature based on a pooling layer to obtain a cascade feature comprises:
obtaining the geometric weight and the feature weight of the graph convolution feature, and combining the geometric weight and the feature weight to obtain a cascade edge weight;
and carrying out cascade weighting on the graph convolution characteristics according to the cascade edge weight to obtain cascade characteristics.
7. The method of fast grid denoising according to claim 1, wherein the fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information comprises:
marking the most peripheral neighborhood surface patches in the neighborhood surface patches of the characteristic surface patches as surface patches to be fused;
and carrying out normal fusion on the characteristic correction information and the non-characteristic correction information of the to-be-fused surface patch to obtain fusion correction information, and splicing and fusing the characteristic correction information and the non-characteristic correction information according to the fusion correction information to obtain normal correction information.
8. The fast grid denoising method of claim 2, wherein the performing correction denoising processing on the grid data according to the normal correction information comprises:
calculating a coordinate residual error according to the normal correction information;
and updating the position information of the vertex according to the coordinate residual error.
9. A fast grid denoising apparatus, comprising:
the grid data acquisition unit is used for acquiring grid data, and the grid data comprises a plurality of patches;
the characteristic patch marking unit is used for carrying out characteristic patch calculation on the grid data to obtain the surface variation degree of each patch in the grid data, setting a surface variation degree threshold value, marking the patch with the surface variation degree higher than the surface variation degree threshold value as a characteristic patch, and other patches as non-characteristic patches;
the normal correction unit is used for carrying out normal correction on the characteristic patches and the neighborhood patches of the characteristic patches based on a deep learning method to obtain characteristic correction information, carrying out normal correction on the non-characteristic patches based on a traditional method to obtain non-characteristic correction information, and fusing the characteristic correction information and the non-characteristic correction information to obtain normal correction information;
and the correction denoising unit is used for performing correction denoising processing on the grid data according to the normal correction information.
10. An electronic device comprising a memory and a processor, wherein,
the memory is used for storing a computer program;
the processor, coupled to the memory, for executing a computer program to implement the steps of the fast grid denoising method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311461903.5A CN117197000B (en) | 2023-11-06 | 2023-11-06 | Quick grid denoising method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311461903.5A CN117197000B (en) | 2023-11-06 | 2023-11-06 | Quick grid denoising method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117197000A true CN117197000A (en) | 2023-12-08 |
| CN117197000B CN117197000B (en) | 2024-03-19 |
Family
ID=88990926
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311461903.5A Active CN117197000B (en) | 2023-11-06 | 2023-11-06 | Quick grid denoising method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117197000B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104183020A (en) * | 2014-07-09 | 2014-12-03 | 浙江大学 | Terrain grid simplifying method based on local quadric error metric with penalty term |
| CN109242972A (en) * | 2018-08-14 | 2019-01-18 | 重庆大学 | A kind of double normal mesh model method for fairing based on apex feature |
| CN112053296A (en) * | 2020-08-25 | 2020-12-08 | 五邑大学 | Robust geometric mean filtering grid denoising method |
| CN112200733A (en) * | 2020-09-09 | 2021-01-08 | 浙江大学 | Grid denoising method based on graph convolution network |
| CN114240779A (en) * | 2021-12-15 | 2022-03-25 | 成都飞机工业（集团）有限责任公司 | Point cloud denoising method, device, equipment and storage medium |
| CN114693660A (en) * | 2022-04-11 | 2022-07-01 | 中国人民解放军空军工程大学 | Solid rocket engine charging calculation grid generation method based on ICT |
| WO2023024395A1 (en) * | 2021-08-26 | 2023-03-02 | 深圳市慧鲤科技有限公司 | Method and apparatus for model optimization, electronic device, storage medium, computer program, and computer program product |
| CN116468632A (en) * | 2023-04-17 | 2023-07-21 | 武汉中观自动化科技有限公司 | Grid denoising method and device based on self-adaptive feature preservation |
-
2023
- 2023-11-06 CN CN202311461903.5A patent/CN117197000B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104183020A (en) * | 2014-07-09 | 2014-12-03 | 浙江大学 | Terrain grid simplifying method based on local quadric error metric with penalty term |
| CN109242972A (en) * | 2018-08-14 | 2019-01-18 | 重庆大学 | A kind of double normal mesh model method for fairing based on apex feature |
| CN112053296A (en) * | 2020-08-25 | 2020-12-08 | 五邑大学 | Robust geometric mean filtering grid denoising method |
| CN112200733A (en) * | 2020-09-09 | 2021-01-08 | 浙江大学 | Grid denoising method based on graph convolution network |
| LU500415B1 (en) * | 2020-09-09 | 2022-03-09 | Univ Zhejiang | Grid denoising method based on graphical convolution network |
| WO2023024395A1 (en) * | 2021-08-26 | 2023-03-02 | 深圳市慧鲤科技有限公司 | Method and apparatus for model optimization, electronic device, storage medium, computer program, and computer program product |
| CN114240779A (en) * | 2021-12-15 | 2022-03-25 | 成都飞机工业（集团）有限责任公司 | Point cloud denoising method, device, equipment and storage medium |
| CN114693660A (en) * | 2022-04-11 | 2022-07-01 | 中国人民解放军空军工程大学 | Solid rocket engine charging calculation grid generation method based on ICT |
| CN116468632A (en) * | 2023-04-17 | 2023-07-21 | 武汉中观自动化科技有限公司 | Grid denoising method and device based on self-adaptive feature preservation |
Non-Patent Citations (4)
| Title |
|---|
| 吴禄慎 等: "基于特征信息分类的三维点数据去噪", 光学精密工程, pages 1465 - 1473 * |
| 张英奎: "基于深度学习的三维数据去噪", 硕士电子期刊信息科技辑, pages 22 - 28 * |
| 赵勇;李玲;单欣;王胜科;秦洪;: "三维几何模型的L_0去噪算法", 计算机辅助设计与图形学学报, no. 05, pages 28 - 33 * |
| 高伟;李政;康倩;: "利用面片法向均值滤波的混合型网格去噪方法", 工程图学学报, no. 04, pages 88 - 93 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117197000B (en) | 2024-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111795704B (en) | Method and device for constructing visual point cloud map | |
| CN109492666B (en) | Image recognition model training method and device and storage medium | |
| CN111968165B (en) | Dynamic human body three-dimensional model complement method, device, equipment and medium | |
| CN111833393A (en) | Binocular stereo matching method based on edge information | |
| EP3905194A1 (en) | Pose estimation method and apparatus | |
| CN107688824A (en) | Picture match method and terminal device | |
| CN111553985B (en) | O-graph pairing European three-dimensional reconstruction method and device | |
| CN112784874B (en) | Binocular vision stereo matching method and device, electronic equipment and storage medium | |
| CN107507232B (en) | Stereo matching method based on multi-scale iteration | |
| US11443481B1 (en) | Reconstructing three-dimensional scenes portrayed in digital images utilizing point cloud machine-learning models | |
| Molnár et al. | Quadratic transformation for planar mapping of implicit surfaces | |
| CN108122280A (en) | The method for reconstructing and device of a kind of three-dimensional point cloud | |
| CN116310076A (en) | Three-dimensional reconstruction method, device, equipment and storage medium based on nerve radiation field | |
| CN112336342A (en) | Hand key point detection method and device and terminal equipment | |
| CN111461164A (en) | Sample data set capacity expansion method and model training method | |
| CN111583398B (en) | Image display method, device, electronic equipment and computer readable storage medium | |
| CN111369682A (en) | Method, apparatus and computer-readable storage medium for generating leaf-like structures | |
| CN112435193A (en) | Method and device for denoising point cloud data, storage medium and electronic equipment | |
| Aubry et al. | A three-dimensional parametric mesher with surface boundary-layer capability | |
| CN111325697A (en) | Color image restoration method based on tensor eigen transformation | |
| CN110909778A (en) | Image semantic feature matching method based on geometric consistency | |
| CN111626311A (en) | Heterogeneous graph data processing method and device | |
| CN117197000B (en) | Quick grid denoising method and device and electronic equipment | |
| CN113888697A (en) | Three-dimensional reconstruction method under two-hand interaction state | |
| Massart et al. | Matrix geometric means based on shuffled inductive sequences |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |