CN113435249A - Densenet-based convolutional neural network finger vein identification method - Google Patents
Densenet-based convolutional neural network finger vein identification method Download PDFInfo
- Publication number
- CN113435249A CN113435249A CN202110541779.8A CN202110541779A CN113435249A CN 113435249 A CN113435249 A CN 113435249A CN 202110541779 A CN202110541779 A CN 202110541779A CN 113435249 A CN113435249 A CN 113435249A
- Authority
- CN
- China
- Prior art keywords
- finger vein
- neural network
- vein image
- image
- recognition
- 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.)
- Pending
Links
- 210000003462 vein Anatomy 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 23
- 238000000605 extraction Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 230000000694 effects Effects 0.000 claims abstract description 11
- 238000012795 verification Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000003708 edge detection Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000013526 transfer learning Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000008034 disappearance Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 2
- 230000005012 migration Effects 0.000 claims description 2
- 238000005286 illumination Methods 0.000 abstract description 4
- 230000002411 adverse Effects 0.000 abstract description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000002035 prolonged effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 210000004204 blood vessel Anatomy 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 102000001554 Hemoglobins Human genes 0.000 description 1
- 108010054147 Hemoglobins Proteins 0.000 description 1
- 210000001015 abdomen Anatomy 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000012850 discrimination method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides a Densenet-based convolutional neural network finger vein recognition method, which is used for carrying out image preprocessing on an original image in an established finger vein image database; performing feature extraction by using the test set image to train a deep convolution model of the convolutional neural network based on densenert, and after each training, evaluating the model by using a verification set and adjusting hyper-parameters to prevent overfitting; and after the model is trained for a certain number of times, testing the effect of the model by using the test set to perform feature extraction, and determining whether to continue training and verifying according to the quality of the test result until the model can obtain a satisfactory result after retesting. The invention has the beneficial effects that: the adverse effect on the recognition result caused by the factors such as finger vein image rotation, strong and weak illumination, insufficient image data learning and the like is avoided, the model training time is prolonged, and the problems of poor recognition effect caused by insufficient generalization capability of the finger vein recognition method and the image factors are solved.
Description
Technical Field
The invention relates to the field of computer vision and biological characteristics, in particular to a finger vein identification method based on a Densenet convolutional neural network.
Background
With the development of science and technology, biological identification has been widely applied in the field of security protection, and brings great convenience for identity identification and information security guarantee. Finger vein recognition is a biological technology for identity recognition based on uniqueness of finger veins, and has the advantages of non-contact property, vividness and internal characteristics compared with other biological recognition methods. The principle that the finger vein can be used as a biological identification method is that when near-infrared light with a certain wavelength (700 nm-900 nm) irradiates a finger, the near-infrared light can penetrate the finger of a person, but hemoglobin in a blood vessel of the finger has stronger capability of absorbing the near-infrared light than tissues and musculoskeletal bones around the blood vessel, so that an image with difference of light and shade is formed, and the dark part in the finger is the vein. The method mainly comprises a transmission type and a reflection type, the difference of the two types is that the irradiation modes of near infrared light are different, the former irradiates the back of a finger, the latter irradiates the abdomen of the finger, and a camera for collecting images is positioned at the lower side of the finger, and clearer finger vein images can be obtained through the irradiation modes generally, so that the vein images in the invention are all transmission type. However, due to factors such as the acquisition device and the environment, simple image preprocessing is generally required to eliminate the interference. The finger vein is generally considered to have uniqueness, vitality and safety, so that the finger vein has good biological identification characteristics, and has become a mainstream biological identification research subject at present, a plurality of traditional solutions appear, but most of the solutions have no good generalization capability, are limited by a specific data set and a method, and have a prominent problem that the identification effect cannot meet practical requirements due to the fact that a plurality of methods only utilize the vein part in the finger vein image and do not perform sufficient feature learning.
The research of finger vein image recognition has important significance to the fields of biological recognition, information safety and the like, with increasing identity verification activities and rapid development of computer science, the deep learning technology based on the convolutional neural network can directly learn more personalized features of images from image data, the method has the most satisfactory effect in typical visual tasks such as image classification, image segmentation, target detection, biological recognition and the like, and the accuracy and the robustness of the method exceed those of the traditional biological discrimination method.
Disclosure of Invention
Aiming at the technical problems of uneven image illumination, image rotation and insufficient image data learning in the conventional finger vein recognition method, the invention provides a convolutional neural network finger vein recognition method based on densenert, which mainly comprises the following steps of:
s1: establishing a finger vein image database, and performing batch preprocessing on all finger vein images in the image database;
s2: introducing a dense network densenert to construct a fully-connected finger vein image recognition deep neural network, and performing transfer learning on the finger vein image recognition deep neural network by using Imagenet initialization weight values;
s3: respectively obtaining a training set, a verification set and a test set from the image database established in the step S1, loading data in the training set to train the finger vein image deep neural network, and establishing and using the verification set to correct the finger vein image characteristics to extract a learning model;
s4: after the model is tested by the test set, measuring the performance of the finger vein image feature extraction learning model by using the classification recognition evaluation index until the test result reaches the preset recognition effect, ending the training to obtain the final finger vein image feature extraction learning model;
s5: and inputting the actually obtained finger vein image into a final finger vein image feature extraction learning model to obtain a feature matrix of the finger vein image, so that the recognition of the finger vein is completed by utilizing the feature matrix.
Further, the preprocessing comprises ROI interest region extraction and image equalization and normalization processing.
Furthermore, the ROI extraction processing adopts a method of combining sobel operator edge detection and centerline calculation.
Further, the contrast limit adaptive histogram equalization method is adopted in the image equalization processing.
Further, the finger vein image recognition deep neural network framework is improved in that full connection is adopted between layers of the finger vein image recognition deep neural network framework, the dense connection blocks all receive the output of all previous layers to establish connection relations between different layers, the RELU is selected as an activation function to reduce the situation of gradient disappearance, the parameter is rapidly optimized by using an SGD random gradient descent method, the loss value is calculated by using a cross entropy method, and softmax is used as the final classification layer output.
Further, migration learning is carried out on the finger vein image recognition deep neural network by using Imagenet initialization weight, then the finger vein image deep neural network is trained by using a training set, and finally loss values are calculated and finger vein image characteristics are optimized by using a verification set to extract learning model parameters.
Further, the finger vein image feature extraction learning model is tested by using a test set, the used evaluation indexes are an FAR-FRR curve, an FPR-TPR curve and a similarity distance distribution graph, the smaller the EER value of the intersection point of the horizontal and vertical coordinates of the FAR-FRR curve is, the better the performance of the finger vein image feature extraction learning model is, the larger the area under the curve is, the better the performance is represented by an ROC curve formed by the FPR-TPR curve, and the more dispersed positive and negative samples of the similarity distance distribution graph are, the stronger the separability of the finger vein image feature extraction learning model is represented by the more dispersed positive and negative samples.
The technical scheme provided by the invention has the beneficial effects that:
1. the invention provides a Densenet-based convolutional neural network finger vein recognition method, which utilizes the strong learning capability of a convolutional neural network to avoid the adverse effect of factors such as image rotation and illumination intensity on a recognition result.
2. By using a transfer learning method and using Imagenet to initialize the weight of the pre-training, the accuracy of the model is improved; meanwhile, the GPU is matched for acceleration, so that the model training time is greatly prolonged.
3. The image features are extracted by utilizing the convolutional neural network model, and the problems of low identification precision and low robustness caused by insufficient learning of the vein image features are effectively solved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
fig. 1 is a flowchart of a convolutional neural network finger vein recognition method based on densenert in an embodiment of the present invention.
Fig. 2 is a schematic diagram of a region of interest extracted from an original finger vein image through edge detection in the embodiment of the present invention.
Fig. 3 is a finger vein image and a histogram after the extracted region of interest is equalized according to the embodiment of the present invention.
Fig. 4 is a schematic diagram of a densely connected Densenst structure used in an embodiment of the present invention.
Fig. 5 is a FAR-FRR graph of a finger vein image feature extraction learning model in an embodiment of the present invention.
Fig. 6 is a FPR-TPR graph of a finger vein image feature extraction learning model in an embodiment of the present invention.
Fig. 7 is a similarity distance distribution diagram of the finger vein image feature extraction learning model in the embodiment of the invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The embodiment of the invention provides a convolutional neural network finger vein identification method based on densenert.
Referring to fig. 1, fig. 1 is a flowchart of a convolutional neural network finger vein recognition method based on densenert in the embodiment of the present invention, which specifically includes the following steps:
step 1.1, collecting finger vein images, preparing to establish an image database for identification, and forming an experimental data set by using the images in the image database, wherein the experimental data set comprises a training set, a verification set and a test set;
and step 1.2, preprocessing all images in the image database, including ROI interest region extraction, image equalization, normalization and other processing, so that the finger vein image is clearer and simpler. As shown in fig. 2, for the finger region extraction, a method of combining sobel operator edge detection and centerline calculation is adopted. The Sobel operator is an important edge detection method, also called a weighted average difference method, and is a first-order discrete difference operator used for calculating the gray value approximate value of the image brightness function. As shown in fig. 3, the vein image is equalized by using contrast Limited Adaptive histogram equalization clahe (contrast Limited Adaptive histogram equalization), and the mapping function f is constructed by using only the histogram distribution in the local region window. Traversing each pixel point of the image, then calculating the local histogram distribution in a window with the size of M multiplied by M around the pixel point, solving a mapping function f, and mapping the pixel point. Compared with the traditional method, the CLAHE method sets a threshold value in the histogram of the image, cuts the histogram assuming that a certain gray level of the histogram exceeds the threshold value, and then averagely distributes the part exceeding the threshold value to each gray level, so that the method has better adaptability. And zero averaging processing to eliminate the average value of the image data for the whole image data.
the DenseNet based framework modifies the last layer, adding a custom embedding layer in the standard architecture, enabling it to produce more distinctive features from the vein pattern input. The dense connection blocks all accept the outputs of all previous layers to establish the connection relationship between different layers. The specific arrangement of the proposed network is shown in table 1. Each conv layer includes Convolution (Convolution), batch normalization (batch normalization), and RELU activation functions. N in the classification layer is the number of classified objects considered in the training process. As shown in table 1, the system is mainly divided into 3 modules, the feature extraction layer in the front includes constraint and poiling, the Dense Block k 1 to the Dense Block 4 are key Dense-connected network blocks, fig. 4 is a schematic diagram of a Dense-connected densnst structure, and finally, a custom layer.
Table 1 improved DenseNet network architecture parameters
Parameters are optimized rapidly by using an SGD random gradient descent method, loss values are calculated by using a cross entropy mode, and softmax is used as final classification layer output. The user-defined layer firstly carries out 7 multiplied by 7 global average pooling to compress the dimension of the feature matrix, then uses batch standardization, adds a discarding layer dorpout to further delete unimportant features, and then uses a full connection layer and batch standardization operation to be connected to a classifier for classification and identification.
And 3, using the Imagenet initialization weight to carry out transfer learning, so that the model has better initial conditions, and the training time is reduced. Training the network model with a training set in a finger vein image database in recognition mode, and using the output of the custom embedded layer as a verified feature template during testing. In the training process, the batch size is 64, random gradient descent (SGD) is adopted, the learning rate is 0.01, 10 is divided after every 30 epochs, the convergence of gradient vectors is accelerated by momentum of 0.9, and the maximum training epoch number is 120. The complete training of the images of all training sets at once is called an epoch.
DenseNet is to let the input of each layer directly influence all the layers after it, and its output is Xl＝Hl([X0,X1,…,Xl-1]) Wherein [ X ]0,X1,…,Xl-1]The previous feature maps are merged in the dimension of the channel. And since each layer contains the output information of all previous layers, it is onlyIt is sufficient to require few feature maps, which is why the amount of parameters of densenert is greatly reduced compared to other models. The dense connection form is equivalent to that each layer is directly connected with input and loss, so that the gradient disappearance phenomenon can be reduced, a deeper network can be established, and the generalization capability is good.
Step 4, extracting the characteristics of the finger vein image to obtain a learning model for testing; the specific method comprises the following steps:
and testing the established finger vein image feature extraction learning model by using a test set in a database, and classifying by using a softmax function according to the Hamming distance. And (5) counting the classification results, and performing system evaluation by using the evaluation indexes. The rejection rate FRR as shown in fig. 5 refers to a ratio at which fingers actually belonging to the same identity are erroneously recognized as unmatched because they do not satisfy the similarity metric, and the false recognition rate FAR refers to a ratio at which fingers actually belonging to different identities are erroneously recognized as matched because they satisfy the similarity metric; according to the FAR-FRR curve, the smaller the EER value of the horizontal and vertical coordinate intersection point is, the better the model performance is, the EER value obtained in the invention is 0.98%, and the method has obvious advantages compared with other methods. The real rate TPR shown in fig. 6 indicates the proportion of correctly classified finger veins; false positive rate FPR refers to the proportion of different finger veins considered to be the same; the larger the area under the ROC curve of a tested working curve consisting of FPR-TPR, the better the performance, and the invention can reach 0.999 which is very close to 1, which is enough to show that the accuracy is very high. And the similarity distance distribution graph is also shown, the more dispersed the positive and negative samples represent the stronger the separability of the phenotype model, as shown in fig. 7, the left and right sides have good dispersibility, and the distribution is normal.
The invention has the beneficial effects that:
1. the invention provides a Densenet-based convolutional neural network finger vein recognition method, which utilizes the strong learning capability of a convolutional neural network to avoid the adverse effect of factors such as image rotation and illumination intensity on a recognition result.
2. By using a transfer learning method and using Imagenet to initialize the weight of the pre-training, the accuracy of the model is improved; meanwhile, the GPU is matched for acceleration, so that the model training time is greatly prolonged.
3. The image features are extracted by utilizing the convolutional neural network model, and the problems of low identification precision and low robustness caused by insufficient learning of the vein image features are effectively solved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (7)
1. A convolutional neural network finger vein recognition method based on densenert is characterized in that: the method comprises the following steps:
s1: establishing a finger vein image database, and performing batch preprocessing on all finger vein images in the image database;
s2: introducing a dense network densenert to construct a fully-connected finger vein image recognition deep neural network, and performing transfer learning on the finger vein image recognition deep neural network by using Imagenet initialization weight values;
s3: respectively obtaining a training set, a verification set and a test set from the image database established in the step S1, loading data in the training set to train the finger vein image deep neural network, and establishing and using the verification set to correct the finger vein image characteristics to extract a learning model;
s4: after the model is tested by the test set, measuring the performance of the finger vein image feature extraction learning model by using the classification recognition evaluation index until the test result reaches the preset recognition effect, ending the training to obtain the final finger vein image feature extraction learning model;
s5: and inputting the actually obtained finger vein image into a final finger vein image feature extraction learning model to obtain a feature matrix of the finger vein image, so that the recognition of the finger vein is completed by utilizing the feature matrix.
2. The method for finger vein recognition based on the denet convolutional neural network of claim 1, wherein: in step S1, the preprocessing includes ROI interest region extraction and image equalization and normalization processing.
3. The method for finger vein recognition based on the denet convolutional neural network of claim 2, wherein: the ROI extraction processing adopts a method of combining sobel operator edge detection and centerline calculation.
4. The method for finger vein recognition based on the denet convolutional neural network of claim 2, wherein: the contrast limit adaptive histogram equalization method adopted by the image equalization processing is adopted.
5. The method for finger vein recognition based on the denet convolutional neural network of claim 1, wherein: in step S2, the improvement point of the finger vein image recognition deep neural network framework is that the layers of the finger vein image recognition deep neural network framework adopt a full connection mode, the dense connection blocks all receive the outputs of all the previous layers to establish a connection relationship between different layers, the activation function selects RELU to reduce the situation of gradient disappearance, the SGD random gradient descent method is used to quickly optimize parameters, the cross entropy method is used to calculate loss values, and softmax is used as the final classification layer output.
6. The method for finger vein recognition based on the denet convolutional neural network of claim 1, wherein: in step S3, migration learning is performed on the finger vein image recognition deep neural network by using the Imagenet initialization weight, then the finger vein image deep neural network is trained by using the training set, and finally the loss value is calculated and the finger vein image feature is optimized by using the verification set to extract the learning model parameters.
7. The method for finger vein recognition based on the denet convolutional neural network of claim 1, wherein: in step S4, the finger vein image feature extraction learning model is tested by using a test set, the evaluation indexes used include an FAR-FRR curve, an FPR-TPR curve, and a similarity distance distribution map, the smaller the EER value of the horizontal and vertical coordinate intersection of the FAR-FRR curve is, the better the performance of the finger vein image feature extraction learning model is, the better the ROC curve composed of the FPR-TPR curve is, the larger the area under the curve is, the better the performance is, and the more dispersed the positive and negative samples of the similarity distance distribution map are, the stronger the separability of the finger vein image feature extraction learning model is.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110541779.8A CN113435249A (en) | 2021-05-18 | 2021-05-18 | Densenet-based convolutional neural network finger vein identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110541779.8A CN113435249A (en) | 2021-05-18 | 2021-05-18 | Densenet-based convolutional neural network finger vein identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113435249A true CN113435249A (en) | 2021-09-24 |
Family
ID=77802652
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110541779.8A Pending CN113435249A (en) | 2021-05-18 | 2021-05-18 | Densenet-based convolutional neural network finger vein identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113435249A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114141A (en) * | 2023-10-20 | 2023-11-24 | 安徽蔚来智驾科技有限公司 | Model training method, evaluation method, computer device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555382A (en) * | 2019-07-31 | 2019-12-10 | 浙江理工大学 | Finger vein identification method based on deep learning and Wasserstein distance measurement |
| CN112560710A (en) * | 2020-12-18 | 2021-03-26 | 北京曙光易通技术有限公司 | Method for constructing finger vein recognition system and finger vein recognition system |
| US20210097681A1 (en) * | 2019-09-26 | 2021-04-01 | Wuyi University | Method for vein recognition, and apparatus, device and storage medium thereof |
-
2021
- 2021-05-18 CN CN202110541779.8A patent/CN113435249A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555382A (en) * | 2019-07-31 | 2019-12-10 | 浙江理工大学 | Finger vein identification method based on deep learning and Wasserstein distance measurement |
| US20210097681A1 (en) * | 2019-09-26 | 2021-04-01 | Wuyi University | Method for vein recognition, and apparatus, device and storage medium thereof |
| CN112560710A (en) * | 2020-12-18 | 2021-03-26 | 北京曙光易通技术有限公司 | Method for constructing finger vein recognition system and finger vein recognition system |
Non-Patent Citations (6)
| Title |
|---|
| JONG MIN SONG ET AL.: "Finger-Vein Recognition Based on Deep DenseNet Using Composite Image", 《IEEE ACCESS》 * |
| JONG MIN SONG ET AL.: "Finger-Vein Recognition Based on Deep DenseNet Using Composite Image", 《IEEE ACCESS》, vol. 7, 4 June 2019 (2019-06-04), pages 66845 - 66863, XP011728408, DOI: 10.1109/ACCESS.2019.2918503 * |
| ZHIYONG TAO ET AL.: "Finger-Vein Recognition Using Bidirectional Feature Extraction and Transfer Learning", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
| ZHIYONG TAO ET AL.: "Finger-Vein Recognition Using Bidirectional Feature Extraction and Transfer Learning", 《MATHEMATICAL PROBLEMS IN ENGINEERING》, vol. 2021, 3 May 2021 (2021-05-03), pages 1 - 11 * |
| 唐溯: "基于深度学习的指静脉识别算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
| 唐溯: "基于深度学习的指静脉识别算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》, no. 12, 15 December 2018 (2018-12-15), pages 138 - 1723 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117114141A (en) * | 2023-10-20 | 2023-11-24 | 安徽蔚来智驾科技有限公司 | Model training method, evaluation method, computer device and storage medium |
| CN117114141B (en) * | 2023-10-20 | 2024-02-27 | 安徽蔚来智驾科技有限公司 | Model training method, evaluation method, computer device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103632132B (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| CN105956582B (en) | A kind of face identification system based on three-dimensional data | |
| CN105005765B (en) | A kind of facial expression recognizing method based on Gabor wavelet and gray level co-occurrence matrixes | |
| CN108009520A (en) | A kind of finger vein identification method and system based on convolution variation self-encoding encoder neutral net | |
| CN107066942A (en) | A kind of living body faces recognition methods and system | |
| CN110427832A (en) | A kind of small data set finger vein identification method neural network based | |
| CN107590797A (en) | A kind of CT images pulmonary nodule detection method based on three-dimensional residual error neutral net | |
| CN108596082A (en) | Human face in-vivo detection method based on image diffusion velocity model and color character | |
| CN106529504B (en) | A kind of bimodal video feeling recognition methods of compound space-time characteristic | |
| CN111126240B (en) | Three-channel feature fusion face recognition method | |
| CN112580590A (en) | Finger vein identification method based on multi-semantic feature fusion network | |
| CN106778489A (en) | The method for building up and equipment of face 3D characteristic identity information banks | |
| CN112288010B (en) | Finger vein image quality evaluation method based on network learning | |
| CN110555380A (en) | Finger vein identification method based on Center Loss function | |
| CN109003275A (en) | The dividing method of weld defect image | |
| CN111832405A (en) | Face recognition method based on HOG and depth residual error network | |
| CN113469143A (en) | Finger vein image identification method based on neural network learning | |
| CN105956570A (en) | Lip characteristic and deep learning based smiling face recognition method | |
| CN113537111A (en) | Iris segmentation method based on double-branch deep convolutional network | |
| CN106778491B (en) | The acquisition methods and equipment of face 3D characteristic information | |
| CN113435249A (en) | Densenet-based convolutional neural network finger vein identification method | |
| CN111209850B (en) | Method for generating applicable multi-device identification finger vein image based on improved cGAN network | |
| CN103942545A (en) | Method and device for identifying faces based on bidirectional compressed data space dimension reduction | |
| CN103279960A (en) | Human body hidden thing image segmentation method based on X-ray back scattering image | |
| CN113420793A (en) | Improved convolutional neural network ResNeSt 50-based gastric ring cell carcinoma classification method |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210924 |