CN114022912B - Palm recognition model training method, device, equipment and readable storage medium - Google Patents
Palm recognition model training method, device, equipment and readable storage medium Download PDFInfo
- Publication number
- CN114022912B CN114022912B CN202111327846.2A CN202111327846A CN114022912B CN 114022912 B CN114022912 B CN 114022912B CN 202111327846 A CN202111327846 A CN 202111327846A CN 114022912 B CN114022912 B CN 114022912B
- Authority
- CN
- China
- Prior art keywords
- palm
- clustered
- image
- images
- recognition model
- 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.)
- Active
Links
- 238000012549 training Methods 0.000 title claims abstract description 88
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000010586 diagram Methods 0.000 claims description 26
- 238000000605 extraction Methods 0.000 claims description 17
- 238000012795 verification Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 6
- 125000004122 cyclic group Chemical group 0.000 abstract 1
- 230000008569 process Effects 0.000 description 25
- 238000004891 communication Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 239000013598 vector Substances 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a palm recognition model training method, a device, equipment and a readable storage medium, wherein the method comprises the following steps: extracting image features of palm images to be clustered in the palm image set to be clustered by using the initial palm recognition model; carrying out connected domain clustering on the palm images to be clustered according to the image characteristics to obtain clustering results of each palm image set to be clustered; extracting representative images in each clustering result; taking each representative image as a training sample to train a target palm recognition model; judging whether the performance parameters of the target palm recognition model meet the set requirements or not; if yes, taking the target palm recognition model as a final palm recognition model; if not, taking the target palm recognition model as a new initial palm recognition model, and executing the subsequent steps. Obviously, each training is to extract representative images from the palm image set to be clustered and train a target palm recognition model by the representative images, and the repeated cyclic training can improve the palm recognition model recognition precision and the palm clustering effect.
Description
Technical Field
The present application relates to the technical field of neural network models, and in particular, to a palm recognition model training method, apparatus, device, and readable storage medium.
Background
With the development of technology, it is very common to verify identity by using biometric identification, and the technologies of face recognition and fingerprint recognition are widely used, so that more biometric identification modes are needed to meet different requirements, including palm recognition technology.
The existing palm recognition technology generally adopts a neural network model to extract the characteristics of the palm, so as to analyze the characteristics of the palm, and then clusters the palm. However, the clustering effect depends on the accuracy of the extracted palm features, and the training mode of the recognition model adopted by the existing palm recognition technology generally adopts a large number of samples, so that the accuracy of the extracted palm features is required to be improved, and the palm clustering effect is further affected.
Therefore, how to train the palm recognition model so that the extraction accuracy of the palm features can be improved in the palm recognition process, so that the improvement of the palm clustering effect is an important problem.
Disclosure of Invention
In view of the above, the present application provides a palm recognition model training method, device, apparatus and readable storage medium, which are used for improving the extraction precision of palm features, thereby improving the efficiency of palm clustering.
In order to achieve the above object, the following solutions have been proposed:
a palm recognition model training method, comprising:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using an initial palm recognition model to obtain image features of each palm image to be clustered;
carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
extracting representative images in the clustering result to be used as representative images corresponding to the palm image set to be clustered;
training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples;
Determining performance parameters of the target palm recognition model on a verification set, and judging whether the performance parameters meet a set requirement or not;
If yes, the target palm recognition model is used as a final palm recognition model;
And if not, taking the target palm recognition model as a new initial palm recognition model, and executing the step of extracting image features from the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model.
Preferably, for each palm image set to be clustered, performing connected domain clustering on the palm images to be clustered according to the image features to obtain a clustering result corresponding to the palm image set to be clustered, including:
aiming at each palm image set to be clustered, calculating similarity scores among different palm images to be clustered according to the image features;
and carrying out connected domain clustering on the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered.
Preferably, the clustering of connected domains of the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered includes:
Establishing a first undirected graph and a second undirected graph aiming at each palm image set to be clustered, wherein the numbers of vertexes contained in the two undirected graphs are the same as the number of the palm images to be clustered in the palm images to be clustered, each vertex corresponds to one palm image to be clustered, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a first score threshold in the first undirected graph are connected through undirected edges, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a second score threshold in the second undirected graph are connected through undirected edges, and each vertex connected through undirected edges in the two undirected graphs forms a vertex group, wherein the first score threshold is lower than the second score threshold;
The vertex group which is communicated through undirected edges in the second undirected graph is used as a cluster, and each cluster forms a second initial cluster result;
and determining a clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result.
Preferably, the determining, according to the first initial clustering result and the second initial clustering result, a clustering result corresponding to the palm image set to be clustered includes:
Inquiring the corresponding vertex group numbers of all vertexes in the first initial clustering result in the second initial clustering result, and replacing all vertex numbers in the first initial clustering result with the corresponding vertex group numbers, wherein the vertex numbers are consistent with the palm image numbers to be clustered corresponding to the vertexes, each vertex group in the second initial clustering result corresponds to a group number, and the group numbers of different vertex groups are different;
And carrying out de-duplication treatment on vertexes with the same vertex numbers in each vertex group of the first initial clustering result, wherein the vertex group after the de-duplication treatment is used as a clustering cluster, and each clustering cluster forms a clustering result corresponding to the palm image set to be clustered.
Preferably, the extracting a representative image in the clustering result as a representative image corresponding to the palm image set to be clustered includes:
Calculating average image characteristics of all palm images to be clustered in each cluster in the cluster result;
And determining the palm images to be clustered corresponding to the image features closest to the average image features in the clustering cluster, and taking the palm images to be clustered as representative images corresponding to the palm image sets to be clustered.
Preferably, after the extracting the representative image in the clustering result, the method further includes:
aiming at the palm image sets to be clustered in pairs, calculating similarity scores between representative images of clustering results corresponding to the two palm image sets to be clustered;
And performing de-duplication treatment on the clustering result corresponding to the representative image exceeding the third score threshold and the palm image set to be clustered, and removing the clustering result corresponding to the removed palm image set to be clustered together.
Preferably, the extracting image features of the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model to obtain the image features of each palm image to be clustered includes:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using a plurality of different initial palm recognition models to obtain image features of each palm image to be clustered, wherein the image features are formed by fusing a plurality of different initial palm recognition models to a plurality of different features extracted from the same palm image to be clustered;
The training of the target palm recognition model by taking the representative images corresponding to each palm image set to be clustered as training samples comprises the following steps:
And training target palm recognition models with the same number as the initial palm recognition models by taking representative images corresponding to each palm image set to be clustered as training samples.
A palm recognition model training device, comprising:
The characteristic extraction unit is used for extracting image characteristics of the palm images to be clustered in the plurality of palm images to be clustered by utilizing the initial palm recognition model to obtain the image characteristics of each palm image to be clustered;
The image clustering unit is used for carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
the representative image extraction unit is used for extracting representative images in the clustering result and taking the representative images as representative images corresponding to the palm image set to be clustered;
The model training unit is used for taking representative images corresponding to each palm image set to be clustered as training samples and training a target palm recognition model;
The parameter verification unit is used for determining the performance parameters of the target palm recognition model on a verification set and judging whether the performance parameters meet a set requirement or not;
The model determining unit is used for taking the target palm recognition model as a final palm recognition model if the performance parameter meets a set requirement;
And the model retraining unit is used for taking the target palm recognition model as a new initial palm recognition model and returning to execute the feature extraction unit if the performance parameters do not meet the set requirements.
A palm recognition model training device comprises a memory and a processor;
The memory is used for storing programs;
the processor is used for executing the program and realizing each step of the palm recognition model training method.
A readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the palm recognition model training method described above.
From the above scheme, the palm recognition model training method provided by the application comprises the following steps: extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using an initial palm recognition model to obtain image features of each palm image to be clustered; carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered; extracting representative images in the clustering result to be used as representative images corresponding to the palm image set to be clustered; training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples; determining performance parameters of the target palm recognition model on a verification set, and judging whether the performance parameters meet a set requirement or not; if yes, the target palm recognition model is used as a final palm recognition model; and if not, taking the target palm recognition model as a new initial palm recognition model, and executing the step of extracting image features from the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model. Obviously, each time, the representative image extracted after each palm image set to be clustered is used for training the target palm recognition model, a large number of palm images to be clustered can be prevented from being extracted, so that training efficiency is improved, the image sets are extracted again after each time of training of the palm recognition model, the image features are extracted from the image sets with higher extraction precision, then the representative features of each palm image set to be clustered are extracted again, further, the palm recognition model is trained again, and the palm recognition precision can be improved through multiple times of training, so that the palm clustering efficiency is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a palm recognition model training method according to an embodiment of the present application;
fig. 2 is a schematic diagram of a first initial clustering result of palm images to be clustered according to an embodiment of the present application;
fig. 3 is a schematic diagram of a second initial clustering result of palm images to be clustered according to an embodiment of the present application;
fig. 4 is a schematic diagram of a clustering result of palm images to be clustered according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a palm recognition model training device according to an embodiment of the present application;
Fig. 6 is a block diagram of a hardware structure of a palm recognition model training device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Next, the palm recognition model training method of the present application will be described in detail, referring to fig. 1, fig. 1 is a schematic flow chart of a palm recognition model training method provided in an embodiment of the present application, where the method includes:
Step S100: extracting image features of the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model to obtain the image features of each palm image to be clustered.
Specifically, the starting palm recognition model may be a trained palm recognition model that may extract image features of a palm image.
The palm image set to be clustered can be a palm image set which is not subjected to category labeling, wherein a palm video can be used as a palm image set, and a folder containing a plurality of palm images can also be used as a palm image set.
In addition, the process of extracting the image features may include: and extracting the features of the palm image by using a feature extraction function of the palm recognition model, and taking the image features as output in the form of matrix vectors.
Step S110: and carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered, and obtaining a clustering result corresponding to the palm image set to be clustered.
Specifically, for each palm image set to be clustered, according to image features of each palm image to be clustered, connected domain clustering can be performed on the palm images to be clustered, wherein the connected domain clustering process can include determining relevance among the palm images according to the image features, and clustering the palm images according to the relevance.
In addition, the clustering result of each palm image set to be clustered can have a plurality of different clustering clusters.
Step S120: and extracting representative images in the clustering result to serve as representative images corresponding to the palm image set to be clustered.
Specifically, the palm image which can most represent the clustering result can be selected from the clustering result and used as the representative image of the palm image set to be clustered.
Step S130: and training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples.
Specifically, the representative images of each palm image set to be clustered are used as training samples to train a target palm recognition model, wherein the target palm recognition model can be different from the initial palm recognition model in the previous step, or can be the same as the model structure of the initial palm recognition model but different in parameters.
Step S140: and determining the performance parameters of the target palm recognition model on the verification set, judging whether the performance parameters meet the set requirements, if so, executing the step S150, and if not, executing the step S160.
Specifically, performance parameters of the target palm recognition model in the training process can be verified, for example: and verifying the error rate of the target palm recognition model on the verification set, and judging whether the error rate meets the set requirement.
It should be noted that, if the training with fewer times is performed, the target palm recognition model with better performance parameters may be obtained, in order to avoid errors and obtain the target palm recognition model with stable performance parameters, the target palm recognition model may still be used as the initial palm recognition model again at this time, that is, step S160 may be executed, and the subsequent steps may be executed continuously, then the above-mentioned process of the user designated times may be repeatedly executed, and after the execution times reach the user requirement, the performance parameters of the target palm recognition model may be considered to reach the enabled setting requirement, so that step S150 may be executed, and the obtained final palm recognition model may be applied to palm recognition.
In addition, if the target palm recognition model with poor performance parameters is obtained after training, the performance parameters of the target palm recognition model can be considered to meet the setting requirement of abandon at the moment, and step S150 can be executed at the moment, but the final palm recognition model obtained at the moment can be eliminated, so that the new initial palm recognition model can be reused to execute the palm recognition model training method provided by the application.
Further, if the error rate on the verification set reaches the set requirement, step S150 may be executed, and if the error rate does not reach the set requirement, step S160 may be executed.
Step S150: and taking the target palm recognition model as a final palm recognition model.
Specifically, if the performance parameter of the trained target palm recognition model on the verification set meets the set requirement, training may be stopped and the target palm recognition model may be determined as the final palm recognition model.
Step S160: taking the target palm recognition model as a new initial palm recognition model, and then executing the step S100 and subsequent steps.
Specifically, if the performance parameters of the trained target palm recognition model on the verification set do not meet the set requirements, the target palm recognition model may be used as a new initial palm recognition model, and the subsequent steps may be performed.
According to the scheme, the representative images extracted after the palm image sets to be clustered are clustered each time to train the target palm recognition model, so that training efficiency is improved, the image sets are extracted again after each time of training of the palm recognition model, the image features are extracted from the image sets with higher extraction precision, then the representative features of the palm image sets to be clustered are extracted again and clustered, the palm recognition model is trained by using the obtained clustering result again, and the palm recognition model recognition precision can be improved through multiple times of training, and meanwhile, the palm clustering effect is improved.
In some embodiments of the present application, the step S110 is introduced, and for each palm image set to be clustered, a process of performing connected domain clustering on the palm images to be clustered according to the image features to obtain a clustering result corresponding to the palm image set to be clustered is further described below.
Specifically, the process may include the steps of:
s1, calculating similarity scores among different palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered.
Specifically, for each palm image set to be clustered, calculating similarity scores of image features among different palm images to be clustered, wherein the similarity scores can represent similarity among different images.
S2, carrying out connected domain clustering on the palm images to be clustered according to the similarity score, and obtaining a clustering result corresponding to the palm image set to be clustered.
Specifically, according to similarity scores among different images, all palm images to be clustered are subjected to connected domain clustering under each palm image set to be clustered, and a clustering result corresponding to each palm image set to be clustered can be obtained.
According to the scheme, the similarity among the different palm images to be clustered is determined according to the similarity score, and the connected domain clustering is carried out on the palm images to be clustered according to the similarity, so that the palms to be clustered with higher relevance can be classified, and the palm images to be clustered of each palm image set to be clustered are clustered into corresponding categories.
In some embodiments of the present application, the step S2 is introduced, and the process of clustering connected domains of the palm images to be clustered according to the similarity score to obtain the clustering result corresponding to the palm image set to be clustered is further described below.
The specific process can comprise the following steps:
S21, establishing a first undirected graph and a second undirected graph aiming at each palm image set to be clustered, wherein the numbers of vertexes contained in the two undirected graphs are the same as the number of the palm images to be clustered in the palm images to be clustered, each vertex corresponds to one palm image to be clustered, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a first score threshold in the first undirected graph are connected through undirected edges, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a second score threshold in the second undirected graph are connected through undirected edges, and each vertex connected through undirected edges in the two undirected graphs forms a vertex group, wherein the first score threshold is lower than the second score threshold.
Specifically, the vertex for establishing connection in the first undirected graph may be represented as a palm image corresponding to different features of the same palm, and the vertex for establishing connection in the second undirected graph may be represented as a palm image corresponding to the same feature of the same palm. Therefore, in order to achieve the above-described effect, the first score threshold may be set lower, the second score threshold may be set relatively higher, the two palm images may be considered to be images of the same palm when the first score threshold is reached, and the two palm images may be considered to be images of the same palm when the second score threshold is higher.
S22, the vertex group which is communicated through undirected edges in the second undirected graph is used as a cluster, and each cluster is used as a cluster, so that a second initial cluster result is formed.
Specifically, in the first undirected graph, the clusters with fewer vertex groups than the set vertex number requirement can be regarded as noisy picture clusters, so that the noise picture clusters can be removed, and the rest of clusters are used as the first initial clustering result.
In the second undirected graph, establishing a vertex group connected by undirected edges as a cluster, wherein each cluster forms a second initial cluster result, and because the second score threshold is set relatively high, palm images corresponding to the vertices connected by undirected edges are very similar, so that noise removal operation in the first undirected graph is not needed, and each cluster can be directly used as the second initial cluster result.
S23, determining a clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result.
Specifically, a clustering result corresponding to the palm image set to be clustered can be determined by combining the first initial clustering result and the second initial clustering result.
According to the scheme, two undirected graphs are established, two initial clustering results of the same palm image set to be clustered are obtained, one initial clustering result is used as an initial clustering result of different characteristics of the same palm, the other initial clustering result is used as an initial clustering result of the same characteristics of the same palm, and the clustering result without repeated palm images can be obtained by combining the two initial clustering results.
The above-described procedure for determining two initial clustering results will be described next in connection with specific examples, with reference to fig. 2 and 3 for details.
Specifically, the first score threshold in the first undirected graph may be set to 100, and if the similarity score between the palm images to be clustered exceeds 100, the palm images may be regarded as different images of the same palm, so as to obtain a first initial clustering result shown in fig. 2, where the first initial clustering result may include a vertex group formed by connecting vertices with vertex numbers 1,2, 9, 11, and 18, a vertex group formed by connecting vertices with vertex numbers 3,7, 10, and 5, and a palm image to be clustered corresponding to each vertex.
The second score threshold in the second undirected graph may be set to 2000, and if the similarity score between the palm images to be clustered exceeds 2000, the palm images may be regarded as the same image of the same palm, so as to obtain a second initial clustering result shown in fig. 3, where the second initial clustering result may include a vertex group formed by connecting vertices with vertex numbers 3 and 7, a vertex group formed by connecting vertices with vertex numbers 1, 2 and 9, a vertex group formed by connecting vertices with vertex numbers 11 and 18, and a palm image to be clustered corresponding to each vertex.
It should be noted that, the above two initial clustering results are only for illustrating an exemplary case of the present application, and are not necessarily the same as an actual case, and a plurality of clusters may be obtained by both the two initial clustering results in the actual clustering process.
Through the connected domain clustering, different images of the same palm and the same image of the same palm can be clustered respectively, and the connected domain clustering method can be used for a subsequent clustering result determining process.
In some embodiments of the present application, the above step S23 is introduced, and a process of determining the clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result is further described below.
Specifically, the process may include the steps of:
step S231, inquiring the corresponding vertex group numbers of all the vertexes in the first initial clustering result in the second initial clustering result, and replacing all the vertex numbers in the first initial clustering result with the corresponding vertex group numbers, wherein the vertex numbers are consistent with the palm image numbers to be clustered corresponding to the vertexes, each vertex group in the second initial clustering result corresponds to a group number, and the group numbers of different vertex groups are different.
Specifically, each vertex may be given a number when the undirected graph is built, different vertices have different numbers and the vertex numbers in the two undirected graphs are the same with a single palm image.
Therefore, the number of the vertex group corresponding to each vertex in the first initial clustering result in the second initial clustering result can be queried, and in an optional manner, the group number of each vertex group in the second initial clustering result can be the number corresponding to the vertex with the smallest number in the vertex group.
After the group number corresponding to each vertex number in the second initial clustering result is determined, replacing the vertex with the same number in the first initial clustering result with the group number corresponding to the vertex number in the second initial clustering result.
And S232, carrying out de-duplication treatment on vertexes with the same vertex numbers in each vertex group of the first initial clustering result, wherein the vertex group after the de-duplication treatment is used as a clustering cluster, and each clustering cluster forms a clustering result corresponding to the palm image set to be clustered.
Specifically, one vertex with the same vertex number in each vertex group of the first initial clustering result is reserved, the rest vertices with the same numbers and the corresponding palm images are removed, and the vertex group after the duplicate removal treatment is used as a clustering cluster to form a clustering result corresponding to the palm image set to be clustered.
According to the scheme, the characteristic that the second initial clustering result is high in similarity and can be regarded as the same palm image is utilized, the second initial clustering result is mapped into the first initial clustering result, repeated palm images in the first initial clustering result are removed, redundancy of the clustering result can be reduced, and compactness is improved.
The process of determining the clustering result will be described with reference to fig. 4.
Specifically, taking the two initial clustering results described in fig. 2 and fig. 3 as an example.
For the second initial clustering result, each vertex group may have a corresponding group number, where the smallest vertex number in each vertex group is taken as the corresponding group number, as can be seen from fig. 3, the group number of the first vertex group is 3, the group number of the second vertex group is 1, and the group number of the second vertex group is 11.
After determining the group numbers in the second initial clustering result, the group numbers corresponding to the vertexes in the first initial clustering result in the second initial clustering result can be queried, wherein the group numbers corresponding to the vertexes with the vertex numbers of 3 and 7 are 3, the group numbers corresponding to the vertex numbers of 1,2 and 9 are 1, and the group numbers corresponding to the vertex numbers of 11 and 18 are 11, so that the vertex numbers in the first initial clustering result are replaced by the corresponding vertex group numbers, and the vertex groups with the vertex numbers of 1, 11, 3, 10 and 5 can be obtained.
And performing de-duplication treatment on the first initial clustering result with the replacement number to obtain a vertex group with vertex numbers of 1, 11, 3, 10 and 5, wherein the de-duplication treated first initial clustering result is used as a clustering result.
At this time, a clustering result that does not include a duplicate image as shown in fig. 4 can be obtained.
In some embodiments of the present application, the above step S120 is introduced, and a process of extracting a representative image in the clustering result as a representative image corresponding to the palm image set to be clustered is further described below.
Specifically, the process may include the steps of:
s1, calculating average image characteristics of all palm images to be clustered in each clustering cluster in the clustering result.
Specifically, the similarity of the palm image features to be clustered in each cluster is higher, so that the average image features of all the palm images to be clustered in each cluster can be calculated.
S2, determining palm images to be clustered corresponding to the image features closest to the average image features in the clustering clusters, and taking the palm images to be clustered as representative images corresponding to the palm image sets to be clustered.
Specifically, palm images to be clustered corresponding to the image features closest to the average image features are determined in each cluster and used as representative images of the clusters, and further representative images of all the clusters can be combined to form representative images of a palm image set to be clustered.
According to the scheme, the representative image is selected from each cluster and used as the representative image of the palm image set to be clustered, so that the representative image can replace the whole palm image set to be clustered to carry out subsequent steps, the sample size can be reduced, and the accuracy of training samples can be improved.
In consideration of that the palm image sets to be clustered may have the same or similar relationship, in step S120, a process of removing the repeated palm image set to be clustered may be added after the representative image in the clustering result is extracted.
Specifically, the process may include the steps of:
S1, calculating similarity scores between representative images of clustering results corresponding to two palm image sets to be clustered according to the two palm image sets to be clustered.
Specifically, the similarity score between each palm image set to be clustered can be calculated according to the image features of the representative images of each palm image set to be clustered.
S2, performing de-duplication processing on the clustering result corresponding to the representative image exceeding the third score threshold and the palm image set to be clustered, and removing the clustering result corresponding to the removed palm image set to be clustered.
Specifically, if the similarity score between the palm image sets to be clustered exceeds the third score threshold, the palm image sets to be clustered may be considered as the same or similar relationship. In order to distinguish the palm image sets to be clustered more easily, the third score threshold may be set lower than the second score threshold but higher than the first score threshold, that is, the two palm image sets to be clustered may be regarded as the same or similar as long as a certain similarity is achieved, without achieving a very high similarity.
After the repeated palm image sets to be clustered are determined, only one of the palm image sets to be clustered can be reserved, and the rest palm image sets to be clustered which are the same or similar can be removed, so that clustering results corresponding to the removed palm image sets to be clustered can be removed together.
According to the scheme, repeated palm image sets to be clustered are removed, repeated feature extraction and clustering on the same palm can be avoided, and training efficiency of a target palm recognition model can be improved by reducing the repeated palm image sets to be clustered.
In order to acquire more accurate features of the palm images to be clustered, further, the images to be clustered can be clustered more accurately, and the target palm recognition model can be trained later, the image features of the palm images to be clustered can be extracted by using a plurality of initial palm recognition models, and a process of extracting the image features by using the plurality of initial palm recognition models is described next.
Specifically, the process may include the steps of:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using a plurality of different initial palm recognition models to obtain the image features of each palm image to be clustered, wherein the image features are formed by fusing a plurality of different features extracted from the same palm image to be clustered by using a plurality of different initial palm recognition models.
Specifically, the image features focused by different initial palm recognition models may be different, and the multiple initial palm recognition models may extract different image features from the same palm image to be clustered, where the different image features have a certain complementary effect, so that the different image features of the same palm image to be clustered may be fused, for example: the extracted image features are feature vectors, a plurality of feature vectors can be spliced together after normalization, and the spliced feature vectors are used as the image features of the palm images to be clustered.
Also, in order to obtain the image features of the palm images to be clustered more comprehensively, the step S100 of training the target palm recognition model by using the representative images corresponding to the palm image sets to be clustered as training samples may include:
And training target palm recognition models with the same number as the initial palm recognition models by taking representative images corresponding to each palm image set to be clustered as training samples.
Specifically, the same training sample can be used to train a plurality of different target palm recognition models, and the image features focused by the different target palm recognition models can be different.
According to the scheme, different palm recognition models can pay attention to different image features, the image features of the palm images to be clustered can be obtained more accurately after the different image features are fused, and the speed of the palm clustering process in the palm recognition process can be improved due to the image features with higher accuracy.
It can be understood that the final palm recognition model can be obtained through training through the embodiment, and on the basis of the final palm recognition model, the embodiment of the application can further apply the trained final palm recognition model to recognize the palm image to be recognized.
Specifically, inputting the palm image to be identified into a final palm identification model, and obtaining an identification result output by the model, namely the type of the palm to be identified.
Taking an unlabeled palm video as an example, inputting an unlabeled palm video composed of a plurality of palm images to be identified into a final palm identification model, if a palm image matched with the palm image to be identified exists in a palm image library, it can be considered that a person corresponding to the palm in the palm video has already undergone information registration, and further, a corresponding service can be provided for the person, for example: opening access control, confirming sign-in and the like; if the palm image matched with the palm image to be identified does not exist in the palm image library, the person corresponding to the palm in the palm video can be considered to be not subjected to information registration, and information input guidance or corresponding problem feedback service can be provided for the person.
The palm recognition model training device provided by the embodiment of the application is described below, and the palm recognition model training device described below and the palm recognition model training device described above can be referred to correspondingly.
First, referring to fig. 5, a palm recognition model training device is described, as shown in fig. 5, the palm recognition model training device may include:
The feature extraction unit 100 is configured to extract image features of palm images to be clustered in a plurality of palm images to be clustered by using a starting palm recognition model, so as to obtain image features of each palm image to be clustered;
The image clustering unit 110 is configured to perform connected domain clustering on the palm images to be clustered according to the image features for each palm image set to be clustered, so as to obtain a clustering result corresponding to the palm image set to be clustered;
A representative image extracting unit 120, configured to extract a representative image in the clustering result, as a representative image corresponding to the palm image set to be clustered;
the model training unit 130 is configured to train a target palm recognition model by using representative images corresponding to each palm image set to be clustered as training samples;
A parameter verification unit 140, configured to determine a performance parameter of the target palm recognition model on a verification set, and determine whether the performance parameter meets a set requirement;
The model determining unit 150 is configured to take the target palm recognition model as a final palm recognition model if the performance parameter meets a set requirement;
And a model retraining unit 160, configured to return to executing the feature extraction unit 100 by using the target palm recognition model as a new initial palm recognition model if the performance parameter does not meet the set requirement.
Optionally, the image clustering unit 110 may include:
the similarity score calculation unit is used for calculating similarity scores among different palm images to be clustered according to the image characteristics for each palm image set to be clustered;
And the clustering result determining unit is used for carrying out connected domain clustering on the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered.
Optionally, the clustering result determining unit may include:
The device comprises a non-directional diagram establishing unit, a non-directional diagram establishing unit and a non-directional diagram establishing unit, wherein the non-directional diagram establishing unit is used for establishing a first non-directional diagram and a second non-directional diagram aiming at each palm image set to be clustered, the number of vertexes contained in the two non-directional diagrams is the same as that of the palm images to be clustered in the palm image set to be clustered, each vertex corresponds to one palm image to be clustered, the vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a first score threshold value in the first non-directional diagram are connected through non-directional edges, the vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a second score threshold value in the second non-directional diagram are connected through non-directional edges, and each vertex connected through non-directional edges in the two non-directional diagrams forms a vertex group, wherein the first score threshold value is lower than the second score threshold value;
The initial clustering result determining unit is used for taking the vertex groups which reach the set vertex number requirement in the first undirected graph and are communicated through undirected edges as clustering clusters, wherein each clustering cluster forms a first initial clustering result, and the vertex groups which are communicated through undirected edges in the second undirected graph are taken as clustering clusters, and each clustering cluster forms a second initial clustering result;
And the image clustering result determining unit is used for determining a clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result.
Optionally, the image clustering result determining unit may include:
the vertex group number determining unit is used for inquiring the vertex group numbers corresponding to the vertexes in the first initial clustering result in the second initial clustering result and replacing the vertex numbers in the first initial clustering result with the corresponding vertex group numbers, wherein the vertex numbers are consistent with the palm image numbers to be clustered corresponding to the vertexes, each vertex group in the second initial clustering result corresponds to a group number, and the group numbers of different vertex groups are different;
and the vertex de-duplication unit is used for performing de-duplication treatment on the vertexes with the same vertex numbers in each vertex group of the first initial clustering result, wherein the vertex groups after the de-duplication treatment are used as clustering clusters, and each clustering cluster forms a clustering result corresponding to the palm image set to be clustered.
Alternatively, the representative image extraction unit 120 may include:
The average image feature calculation unit is used for calculating average image features of all palm images to be clustered in each cluster in the cluster result;
and the representative image determining unit is used for determining the palm image to be clustered corresponding to the image feature closest to the average image feature in the cluster, and taking the palm image to be clustered as the representative image corresponding to the palm image set to be clustered.
Optionally, the palm recognition model training device may further include:
The representative image similarity score calculation unit is used for calculating similarity scores between representative images of clustering results corresponding to the two palm image sets to be clustered aiming at the two palm image sets to be clustered;
The image set de-duplication unit is used for de-duplication processing the clustering result corresponding to the representative image exceeding the third score threshold and the palm image set to be clustered, and removing the clustering result corresponding to the removed palm image set to be clustered together.
Optionally, the feature extraction unit 100 may include:
The comprehensive feature extraction unit is used for extracting image features of the palm images to be clustered in the plurality of palm images to be clustered by using a plurality of different initial palm recognition models to obtain image features of each palm image to be clustered, wherein the image features are formed by fusing a plurality of different features extracted from the same palm image to be clustered by using a plurality of different initial palm recognition models;
the model training unit 130 may include:
the multi-model training unit is used for training target palm recognition models with the same number as the initial palm recognition models by taking representative images corresponding to each palm image set to be clustered as training samples.
The palm recognition model training device provided by the embodiment of the application can be applied to palm recognition model training equipment. Fig. 6 is a block diagram of a hardware structure of the palm recognition model training device, and referring to fig. 6, the hardware structure of the palm recognition model training device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;
In the embodiment of the application, the number of the processor 1, the communication interface 2, the memory 3 and the communication bus 4 is at least one, and the processor 1, the communication interface 2 and the memory 3 complete the communication with each other through the communication bus 4;
The processor 1 may be a central processing unit CPU, or an Application-specific integrated Circuit ASIC (Application SPECIFIC INTEGRATED Circuit), or one or more integrated circuits configured to implement embodiments of the present invention, etc.;
the memory 3 may comprise a high-speed RAM memory, and may further comprise a non-volatile memory (non-volatile memory) or the like, such as at least one magnetic disk memory;
wherein the memory stores a program, the processor is operable to invoke the program stored in the memory, the program operable to:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using an initial palm recognition model to obtain image features of each palm image to be clustered;
carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
extracting representative images in the clustering result to be used as representative images corresponding to the palm image set to be clustered;
training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples;
Determining performance parameters of the target palm recognition model on a verification set, and judging whether the performance parameters meet a set requirement or not;
If yes, the target palm recognition model is used as a final palm recognition model;
And if not, taking the target palm recognition model as a new initial palm recognition model, and executing the step of extracting image features from the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model.
Alternatively, the refinement function and the extension function of the program may be described with reference to the above.
The embodiment of the present application also provides a storage medium storing a program adapted to be executed by a processor, the program being configured to:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using an initial palm recognition model to obtain image features of each palm image to be clustered;
carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
extracting representative images in the clustering result to be used as representative images corresponding to the palm image set to be clustered;
training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples;
Determining performance parameters of the target palm recognition model on a verification set, and judging whether the performance parameters meet a set requirement or not;
If yes, the target palm recognition model is used as a final palm recognition model;
And if not, taking the target palm recognition model as a new initial palm recognition model, and executing the step of extracting image features from the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model.
Alternatively, the refinement function and the extension function of the program may be described with reference to the above.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A palm recognition model training method, comprising:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using an initial palm recognition model to obtain image features of each palm image to be clustered;
carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
extracting representative images in the clustering result to be used as representative images corresponding to the palm image set to be clustered;
training a target palm recognition model by taking representative images corresponding to each palm image set to be clustered as training samples;
Determining performance parameters of the target palm recognition model on a verification set, and judging whether the performance parameters meet a set requirement or not;
If yes, the target palm recognition model is used as a final palm recognition model;
if not, taking the target palm recognition model as a new initial palm recognition model, and executing the step of extracting image features from the palm images to be clustered in the plurality of palm images to be clustered by using the initial palm recognition model;
and for each palm image set to be clustered, performing connected domain clustering on the palm images to be clustered according to the image features to obtain a clustering result corresponding to the palm image set to be clustered, wherein the clustering result comprises:
aiming at each palm image set to be clustered, calculating similarity scores among different palm images to be clustered according to the image features;
carrying out connected domain clustering on the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered;
and performing connected domain clustering on the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered, wherein the clustering result comprises:
Establishing a first undirected graph and a second undirected graph aiming at each palm image set to be clustered, wherein the numbers of vertexes contained in the two undirected graphs are the same as the number of the palm images to be clustered in the palm images to be clustered, each vertex corresponds to one palm image to be clustered, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a first score threshold in the first undirected graph are connected through undirected edges, vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a second score threshold in the second undirected graph are connected through undirected edges, and each vertex connected through undirected edges in the two undirected graphs forms a vertex group, wherein the first score threshold is lower than the second score threshold;
The vertex group which is communicated through undirected edges in the second undirected graph is used as a cluster, and each cluster forms a second initial cluster result;
and determining a clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result.
2. The method according to claim 1, wherein the determining the clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result includes:
Inquiring the corresponding vertex group numbers of all vertexes in the first initial clustering result in the second initial clustering result, and replacing all vertex numbers in the first initial clustering result with the corresponding vertex group numbers, wherein the vertex numbers are consistent with the palm image numbers to be clustered corresponding to the vertexes, each vertex group in the second initial clustering result corresponds to a group number, and the group numbers of different vertex groups are different;
And carrying out de-duplication treatment on vertexes with the same vertex numbers in each vertex group of the first initial clustering result, wherein the vertex group after the de-duplication treatment is used as a clustering cluster, and each clustering cluster forms a clustering result corresponding to the palm image set to be clustered.
3. The method according to claim 1, wherein the extracting the representative image in the clustering result as the representative image corresponding to the palm image set to be clustered includes:
Calculating average image characteristics of all palm images to be clustered in each cluster in the cluster result;
And determining the palm images to be clustered corresponding to the image features closest to the average image features in the clustering cluster, and taking the palm images to be clustered as representative images corresponding to the palm image sets to be clustered.
4. The method according to claim 1, further comprising, after said extracting the representative image in the clustering result:
aiming at the palm image sets to be clustered in pairs, calculating similarity scores between representative images of clustering results corresponding to the two palm image sets to be clustered;
And performing de-duplication treatment on the two palm image sets to be clustered corresponding to the representative images exceeding the third score threshold, and removing the clustering results corresponding to the removed palm image sets to be clustered together.
5. The method according to any one of claims 1-4, wherein the extracting image features of the palm images to be clustered in the plurality of palm images to be clustered using the initial palm recognition model to obtain the image features of each of the palm images to be clustered includes:
Extracting image features of palm images to be clustered in a plurality of palm images to be clustered by using a plurality of different initial palm recognition models to obtain image features of each palm image to be clustered, wherein the image features are formed by fusing a plurality of different initial palm recognition models to a plurality of different features extracted from the same palm image to be clustered;
The training of the target palm recognition model by taking the representative images corresponding to each palm image set to be clustered as training samples comprises the following steps:
And training target palm recognition models with the same number as the initial palm recognition models by taking representative images corresponding to each palm image set to be clustered as training samples.
6. A palm recognition model training device, comprising:
The characteristic extraction unit is used for extracting image characteristics of the palm images to be clustered in the plurality of palm images to be clustered by utilizing the initial palm recognition model to obtain the image characteristics of each palm image to be clustered;
The image clustering unit is used for carrying out connected domain clustering on the palm images to be clustered according to the image characteristics aiming at each palm image set to be clustered to obtain a clustering result corresponding to the palm image set to be clustered;
the representative image extraction unit is used for extracting representative images in the clustering result and taking the representative images as representative images corresponding to the palm image set to be clustered;
The model training unit is used for taking representative images corresponding to each palm image set to be clustered as training samples and training a target palm recognition model;
The parameter verification unit is used for determining the performance parameters of the target palm recognition model on a verification set and judging whether the performance parameters meet a set requirement or not;
The model determining unit is used for taking the target palm recognition model as a final palm recognition model if the performance parameter meets a set requirement;
The model retraining unit is used for taking the target palm recognition model as a new initial palm recognition model if the performance parameters do not meet the set requirements, and returning to execute the feature extraction unit;
wherein, the image clustering unit includes:
the similarity score calculation unit is used for calculating similarity scores among different palm images to be clustered according to the image characteristics for each palm image set to be clustered;
The clustering result determining unit is used for carrying out connected domain clustering on the palm images to be clustered according to the similarity score to obtain a clustering result corresponding to the palm image set to be clustered;
wherein the clustering result determining unit includes:
The device comprises a non-directional diagram establishing unit, a non-directional diagram establishing unit and a non-directional diagram establishing unit, wherein the non-directional diagram establishing unit is used for establishing a first non-directional diagram and a second non-directional diagram aiming at each palm image set to be clustered, the number of vertexes contained in the two non-directional diagrams is the same as that of the palm images to be clustered in the palm image set to be clustered, each vertex corresponds to one palm image to be clustered, the vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a first score threshold value in the first non-directional diagram are connected through non-directional edges, the vertexes corresponding to the two palm images to be clustered with similarity scores exceeding a second score threshold value in the second non-directional diagram are connected through non-directional edges, and each vertex connected through non-directional edges in the two non-directional diagrams forms a vertex group, wherein the first score threshold value is lower than the second score threshold value;
The initial clustering result determining unit is used for taking the vertex groups which reach the set vertex number requirement in the first undirected graph and are communicated through undirected edges as clustering clusters, wherein each clustering cluster forms a first initial clustering result, and the vertex groups which are communicated through undirected edges in the second undirected graph are taken as clustering clusters, and each clustering cluster forms a second initial clustering result;
And the image clustering result determining unit is used for determining a clustering result corresponding to the palm image set to be clustered according to the first initial clustering result and the second initial clustering result.
7. A palm recognition model training device, comprising a memory and a processor;
The memory is used for storing programs;
The processor is configured to execute the program to implement the respective steps of the palm recognition model training method according to any one of claims 1 to 5.
8. A readable storage medium having stored thereon a computer program, which when executed by a processor, performs the steps of the palm recognition model training method according to any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111327846.2A CN114022912B (en) | 2021-11-10 | 2021-11-10 | Palm recognition model training method, device, equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111327846.2A CN114022912B (en) | 2021-11-10 | 2021-11-10 | Palm recognition model training method, device, equipment and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114022912A CN114022912A (en) | 2022-02-08 |
| CN114022912B true CN114022912B (en) | 2024-06-28 |
Family
ID=80063086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111327846.2A Active CN114022912B (en) | 2021-11-10 | 2021-11-10 | Palm recognition model training method, device, equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114022912B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416384A (en) * | 2018-03-05 | 2018-08-17 | 苏州大学 | A kind of image tag mask method, system, equipment and readable storage medium storing program for executing |
| CN109145791A (en) * | 2018-08-09 | 2019-01-04 | 深圳大学 | One kind being based on the contactless fingers and palms recognition methods in mobile terminal and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948542B (en) * | 2019-03-19 | 2022-09-23 | 北京百度网讯科技有限公司 | Gesture recognition method and device, electronic equipment and storage medium |
| CN112183504B (en) * | 2020-11-27 | 2021-04-13 | 北京圣点云信息技术有限公司 | Video registration method and device based on non-contact palm vein image |
| CN113361603B (en) * | 2021-06-04 | 2024-05-10 | 北京百度网讯科技有限公司 | Training method, category identification device, electronic device, and storage medium |
-
2021
- 2021-11-10 CN CN202111327846.2A patent/CN114022912B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416384A (en) * | 2018-03-05 | 2018-08-17 | 苏州大学 | A kind of image tag mask method, system, equipment and readable storage medium storing program for executing |
| CN109145791A (en) * | 2018-08-09 | 2019-01-04 | 深圳大学 | One kind being based on the contactless fingers and palms recognition methods in mobile terminal and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114022912A (en) | 2022-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kyperountas et al. | Weighted piecewise LDA for solving the small sample size problem in face verification | |
| Kumar et al. | A new framework for adaptive multimodal biometrics management | |
| Wu | Power mean SVM for large scale visual classification | |
| US10515265B2 (en) | Generating variations of a known shred | |
| EP3311311A1 (en) | Automatic entity resolution with rules detection and generation system | |
| CN109948735B (en) | Multi-label classification method, system, device and storage medium | |
| JP7024515B2 (en) | Learning programs, learning methods and learning devices | |
| US11062120B2 (en) | High speed reference point independent database filtering for fingerprint identification | |
| McCool et al. | Session variability modelling for face authentication | |
| CN111125658B (en) | Method, apparatus, server and storage medium for identifying fraudulent user | |
| CN110929525B (en) | Network loan risk behavior analysis and detection method, device, equipment and storage medium | |
| CN114332500A (en) | Image processing model training method and device, computer equipment and storage medium | |
| El-Abed et al. | Quality assessment of image-based biometric information | |
| CN114022912B (en) | Palm recognition model training method, device, equipment and readable storage medium | |
| Nandakumar et al. | Biometric fusion: Does modeling correlation really matter? | |
| CN113963235A (en) | Cross-category image recognition model reusing method and system | |
| Kumar et al. | Cross domain descriptor for face sketch-photo image recognition | |
| CN108573212B (en) | Palm feature identity authentication method and device | |
| Meraoumia et al. | An efficient palmprint identification system using multispectral and hyperspectral imaging | |
| JP6988828B2 (en) | Collation processing device, collation processing method, and collation processing program | |
| CN112463964A (en) | Text classification and model training method, device, equipment and storage medium | |
| Attallah et al. | Geometrical local image descriptors for palmprint recognition | |
| Pflug et al. | Binarization of spectral histogram models: An application to efficient biometric identification | |
| Sahoo et al. | Vision-based static hand gesture recognition using dense-block features and svm classifier | |
| Shao et al. | An NMF-based method for the fingerprint orientation field estimation |
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 |