WO2020107922A1 - 一种基于3d指纹图像的性别识别方法及系统 - Google Patents

一种基于3d指纹图像的性别识别方法及系统 Download PDF

Info

Publication number
WO2020107922A1
WO2020107922A1 PCT/CN2019/098195 CN2019098195W WO2020107922A1 WO 2020107922 A1 WO2020107922 A1 WO 2020107922A1 CN 2019098195 W CN2019098195 W CN 2019098195W WO 2020107922 A1 WO2020107922 A1 WO 2020107922A1
Authority
WO
WIPO (PCT)
Prior art keywords
gender
fingerprint image
recognition method
fingerprint
training
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.)
Ceased
Application number
PCT/CN2019/098195
Other languages
English (en)
French (fr)
Inventor
刘凤
刘浩哲
张文天
苟文越
齐勇
沈琳琳
赖志辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Publication of WO2020107922A1 publication Critical patent/WO2020107922A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1365—Matching; Classification
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional [3D] objects

Definitions

  • the invention relates to a gender recognition method, in particular to a gender recognition method based on a 3D fingerprint image, and to a gender recognition system adopting the gender recognition method based on a 3D fingerprint image.
  • Fingerprints as the main biological characteristics of fingers, have several ideal characteristics such as universality, distinctiveness, permanence, collectability, acceptability and anti-circumvention, and are widely used in personal identification systems.
  • the existing biometric recognition based on traditional 2D fingerprint images makes it impossible or extremely difficult to form a traditional reference 2D fingerprint pattern when the fingertip surface is severely damaged or there is oil stains or sweat.
  • OCT Optical Coherence Tomography
  • the technical problem to be solved by the present invention is to provide a gender recognition method that can effectively improve accuracy and precision, and further provide a gender recognition system adopting the gender recognition method.
  • the present invention provides a gender recognition method based on a 3D fingerprint image, including the following steps:
  • Step S1 Collect fingerprint image data of the user based on optical coherence tomography
  • Step S2 extract and train the gender characteristics in the fingerprint image data according to the fingerprint image data
  • Step S3 build a gender feature extraction classification model
  • Step S4 perform gender recognition and prediction through the extracted gender classification model.
  • step S2 includes the following sub-steps:
  • Step S201 build a gender feature extraction classification network for the fingerprint image data
  • Step S202 extracting gender characteristics in the fingerprint image data
  • Step S203 training a classification network under the gender characteristics of the fingerprint image.
  • a further improvement of the present invention is that, in step S202, extracting the gender features in the fingerprint image data includes extracting the shape of the ridge-valley line in the epidermal layer and the nipple layer, and then performing training to achieve gender classification.
  • a further improvement of the present invention is that in step S203, two or more graphics cards are used for parallel calculation and training, wherein the gender feature extraction and classification network is first placed in each graphics card for training, and the result is the current number of trainings.
  • Gradient value, loss function value and discriminant correct rate of single graphics card then return the three results of the above gradient value, loss function value and discriminant correct rate in each graphics card to the CPU and average them respectively, and then get the graphics card to calculate The average value of the gradient, the average value of the loss function, and the average value of the discrimination accuracy rate; finally, in the CPU, the gradient average value calculated by the graphics card is back propagated, and the partial derivative of the parameters in the gender discrimination model is calculated layer by layer, and the parameters are calculated.
  • Update, and the gender network after the parameter update is extracted and put into the graphics card to continue training to achieve the iteration iteration of the network parameters.
  • step S3 includes the following sub-steps:
  • Step S301 storing the classification network under the gender characteristics of the fingerprint image after training as a gender feature extraction classification model
  • Step S302 Construct a gender classifier under the gender feature extraction classification model.
  • a further improvement of the present invention is that, in the step S302, after the collected fingerprint image is input to the gender feature extraction classification model, the gender feature extraction matrix of the classification model output is obtained by taking the position where the maximum probability value is located As a gender label, the fingerprint image information is further converted into a gender label indicating gender characteristics to realize gender classification expression, thereby establishing the gender classifier.
  • step S4 includes the following sub-steps:
  • Step S401 Randomly collect user fingerprint image test data
  • Step S402 inputting the fingerprint image test data into the gender classifier
  • Step S403 output the gender recognition result.
  • a further improvement of the present invention is that the amount of fingerprint image test data collected in step S401 is less than the amount of fingerprint image data collected in step S1.
  • a further improvement of the present invention is that the number of fingerprint image data collected in step S1 is at least 400 fingerprint images of the same finger of the same user.
  • the invention also provides a gender recognition system based on a 3D fingerprint image, which adopts the gender recognition method based on the 3D fingerprint image as described above.
  • the beneficial effects of the present invention are: the use of 3D fingerprint images to analyze and identify gender characteristics, confirms the existence of gender characteristics within the fingerprint, avoids the influence of the surface environment of the fingerprint image on its recognition, and provides The higher accuracy and accuracy of the gender discrimination scheme improves the anti-noise performance, which is of great significance for gender classification, tracking and identification of unknown people in forensic anthropology, and census of population data.
  • FIG. 1 is a schematic diagram of a workflow of an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a fingerprint gender feature extraction training process according to an embodiment of the present invention.
  • this example provides a gender recognition method based on a 3D fingerprint image, including the following steps:
  • Step S1 Collect fingerprint image data of the user based on optical coherence tomography
  • Step S2 extract and train the gender characteristics in the fingerprint image data according to the fingerprint image data
  • Step S3 build a gender feature extraction classification model
  • Step S4 perform gender recognition and prediction through the extracted gender classification model.
  • fingerprint images of OCT optical coherence tomography
  • OCT optical coherence tomography
  • Step S2 in this example includes the following sub-steps:
  • Step S201 build a gender feature extraction classification network for the fingerprint image data
  • Step S202 extracting gender characteristics in the fingerprint image data
  • Step S203 training a classification network under the gender characteristics of the fingerprint image.
  • step S201 uses an existing machine learning algorithm or deep learning algorithm as a model to build a gender feature extraction classification model.
  • extracting the gender features in the fingerprint image data includes extracting the shape of the ridge line in the epidermal layer and the nipple layer, that is, extracting the shape of the ridge line in the epidermal layer and the nipple layer, And extract the transformation shape between the epidermis layer and the nipple layer, and then achieve gender classification through training.
  • the ridge density on the surface of the fingerprint can be used to identify gender.
  • the ridge density is usually male, and the ridge density is female.
  • the amino acid content of the sweat on the finger surface can also be used for gender discrimination; however, if the existing 2D fingerprint images are used to achieve gender classification, the accuracy rate is limited; and the gender classification of fingerprints is for the gender classification, tracking and tracking of forensic anthropology. Identifying unknown people and population data surveys are of great significance.
  • this example is based on optical coherence tomography to collect the user's fingerprint image data, which is preferably more than 400 for a single person.
  • the cross-sectional view of multiple longitudinal fingertip images of the OCT 3D fingerprint image constitutes the fingerprint image data.
  • the 3D scanned image of the fingertip is used for multiple verification of gender discrimination, which effectively improves the noise resistance and accuracy of gender recognition.
  • step S201 in machine learning, such as support vector machine (SVM), K-nearest neighbor algorithm (KNN) and other algorithms, the mapping relationship between the input OCT3D fingerprint image and gender information is searched for .
  • machine learning such as support vector machine (SVM), K-nearest neighbor algorithm (KNN) and other algorithms
  • a network framework such as AlexNet, VGG-Net, or ResNet is used to achieve gender feature extraction, that is, to build a gender feature extraction network through computation layers such as layer-by-layer convolutional layers, pooling layers, and batch normalization. Extract the gender features in the input OCT3D fingerprint image, that is, the different shapes of the ridge-valley lines in the epidermal layer and the nipple layer, so as to achieve gender classification.
  • the 17-layer ResNet network is used as an example to build the gender feature extraction classification network, and the network model is modified to adapt the network model to the input dimension of the current image, and the Google-netInception network is used
  • the structure using 1*A and A*1 instead of A*A's traditional convolution structure, reduces the storage space occupied by the network structure.
  • step S203 in this example the classification network under the gender characteristics of the fingerprint image is trained by a deep learning algorithm; that is, step S203 is extracted by the machine learning and deep learning algorithm (gender classification network) for the step S202
  • the training process of the gender features in the fingerprint image data is updated and trained through repeated iterations.
  • the fingerprint image is extracted with a gender label.
  • This process is implemented by defining the fingerprint information field as M, the gender discriminator as G, and the fingerprint feature extractor as F.
  • M the fingerprint information field
  • M the internal image of the fingerprint
  • the sex discriminator G can be obtained.
  • T M is a sample of the test set of fingerprint image data
  • sgn is a symbol function
  • G(M) is a sex discriminator G under the fingerprint information field M
  • L M is a gender tag of the test set of fingerprint image data, namely
  • step S203 described in this example two or more graphics cards are used for parallel calculation and training.
  • machine learning and training are performed on each graphics card to implement an update iteration, and the average value of more than two graphics cards is calculated.
  • the updated variable parameters are returned to the CPU to be input into each graphics card to achieve repeated iterations.
  • step S203 in this example is shown in FIG. 2, and more than two graphics cards are used for parallel calculation and training, for example, using three graphics cards for calculation.
  • the gender classification network is placed on each graphics card for training, and the results obtained are the gradient value, loss function value and discrimination accuracy rate of the single graphics card under the current training training times, and the three results in each graphics card are returned to the CPU
  • the average value is calculated separately in order to obtain the gradient value, loss function value and discrimination accuracy calculated by the graphics card.
  • the CPU performs back propagation according to the gradient value calculated by the graphics card.
  • the parameters in the gender discrimination model include weight and (Offset, etc.) Find the partial derivative layer by layer, update the parameters, and put the classification network with the updated parameters into the graphics card to continue training to achieve the iteration iteration of the network parameters.
  • step S203 uses multi-graphics parallel computing to speed up the network training speed. It implements the update iteration of the relevant parameters in the specific algorithm in the CPU and inputs it to the GPU, and implements machine learning and deep learning algorithms (gender classification network) in three GPUs.
  • the training process is to update the variables and return the calculated parameters to the CPU and iterate repeatedly, which can effectively improve the accuracy and precision of the training.
  • Step S3 in this example includes the following sub-steps:
  • Step S301 storing the classification network under the gender characteristics of the fingerprint image after training as a gender feature extraction classification model
  • Step S302 Construct a gender classifier under the gender feature extraction classification model.
  • step S302 in this example after inputting the collected fingerprint image to the gender feature extraction classification model, the matrix output by the gender feature extraction classification model is used as the gender label by taking the position where the maximum probability value is located, and The fingerprint image information is converted into gender tags representing gender characteristics to achieve gender classification expression, thereby establishing the gender classifier.
  • step S302 in this example builds a gender classifier on the basis of the gender feature extraction classification model saved in step S301, inputs the input OCT3D fingerprint image into the classification model, and performs specific operations on the output classification results, that is,
  • the sparse matrix output by the classification model can be converted into an intelligible label representing gender by taking the position (corresponding to a gender label) where the numerical value (that is, probability value) in the sparse matrix is the largest, to realize the categorical expression of gender.
  • Step S4 in this example includes the following sub-steps:
  • Step S401 Randomly collect user fingerprint image test data
  • Step S402 inputting the fingerprint image test data into the gender classifier
  • Step S403 output the gender recognition result.
  • the amount of fingerprint image test data collected in step S401 is less than the amount of fingerprint image data collected in step S1, that is, less than the number of 400 individual images during training.
  • this gender classifier can finally achieve a recognition accuracy rate of 99.3%; In the case of inputting more than three OCT 3D fingerprint images, the recognition accuracy of this gender classifier can approach 100% indefinitely. Therefore, this example can effectively improve the accuracy and precision of gender recognition.
  • step S203 using the multi-graphics parallel training method in step S203 in this example can greatly shorten the training time; in step S201, other existing feature extraction methods (including deep learning methods and machine learning algorithms) can also be used instead.
  • This example also provides a gender recognition system based on a 3D fingerprint image, which uses the gender recognition method based on the 3D fingerprint image as described above.
  • this example uses 3D fingerprint images to analyze and identify gender characteristics, confirming the existence of gender characteristics within the fingerprint, avoiding the impact of the surface environment of the fingerprint image on its recognition, and providing higher accuracy and accuracy
  • the gender discrimination scheme improves the anti-noise performance, which is of great significance for gender classification, tracking and identification of unknown people in forensic anthropology, and population data census.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Collating Specific Patterns (AREA)
  • Image Analysis (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

本发明提供一种基于3D指纹图像的性别识别方法及系统，所述基于3D指纹图像的性别识别方法包括以下步骤：步骤S1，基于光学相干断层扫描采集用户的指纹图像数据；步骤S2，根据所述指纹图像数据，提取并训练所述指纹图像数据中的性别特征；步骤S3，搭建性别特征提取分类模型；步骤S4，通过所性别特征提取分类模型执行性别的识别和预测。本发明利用3D指纹图像对性别特征进行分析和识别，证实了指纹内部性别特征的存在，避免了指纹图像的表面环境对其识别的影响，提供了更高正确率和精确度的性别判别方案，提高了抗噪性能，对于法医人类学的性别分类、跟踪与标识未知人群以及人口数据普查等具有非常重要的意义。

Description

一种基于3D指纹图像的性别识别方法及系统 技术领域
本发明涉及一种性别识别方法，尤其涉及一种基于3D指纹图像的性别识别方法，并涉及采用了该基于3D指纹图像的性别识别方法的性别识别系统。
背景技术
指纹作为手指的主要生物特征，具有普遍性、显著性、永久性、可收集性、可接受性和抗规避性等几个理想特性，在个人识别系统中广泛应用。现有基于传统2D指纹图像的生物特征识别，在指尖表面被严重损坏或存在油污、汗液的情况下，使得不可能或极难形成传统的参考2D指纹图案。
随着OCT(Optical coherence tomography，光学相干断层扫描)成像技术的成熟，实现了指纹的3D-图像分析。而OCT 3D指纹图像一方面克服了2D图像中指纹破损、油污以及汗液干扰等缺陷，另一方面还提供了更多、更完整的指纹特征。那么，如何利用OCT 3D指纹图像分析指纹内部信息，进而尽量避免2D图像中的此类问题，将是一个有利的方向。
发明内容
本发明所要解决的技术问题是需要提供一种能够有效提高准确性和精准度的性别识别方法，并进一步提供采用了该性别识别方法的性别识别系统。
对此，本发明提供一种基于3D指纹图像的性别识别方法，包括以下步骤：
步骤S1，基于光学相干断层扫描采集用户的指纹图像数据；
步骤S2，根据所述指纹图像数据，提取并训练所述指纹图像数据中的性别特征；
步骤S3，搭建性别特征提取分类模型；
步骤S4，通过所性别特征提取分类模型执行性别的识别和预测。
本发明的进一步改进在于，所述步骤S2包括以下子步骤：
步骤S201，针对所述指纹图像数据搭建性别特征提取分类网络；
步骤S202，提取所述指纹图像数据中的性别特征；
步骤S203，训练所述指纹图像性别特征下的分类网络。
本发明的进一步改进在于，所述步骤S202中，提取所述指纹图像数据中的性别特征包括提取脊谷线在表皮层和乳头层中变换区域的形状，进而通过训练来实现性别分类。
本发明的进一步改进在于，所述步骤S203中，采用两块以上的显卡并行计算和训练，其中，先将性别特征提取分类网络放至每张显卡中进行训练，得到的结果为当前训练次数下单显卡的梯度值、损失函数值和判别正确率；然后将上述每张显卡中的梯度值、损失函数值和判别正确率这3个结果返回到CPU中并分别求均值，进而得到显卡计算出的梯度平均值、损失函数平均值和判别正确率平均值；最后，在CPU中根据显卡的计算出的梯度平均值进行反向传播，对性别判别模型中的参数逐层求偏导数，进行参数更新，并将参数更新后的性别特征提取分类网络再放入显卡中继续训练，以实现网络参数的更新迭代。
本发明的进一步改进在于，所述步骤S3包括以下子步骤：
步骤S301，保存训练好之后的指纹图像性别特征下的分类网络作为性别特征提取分类模型；
步骤S302，搭建所述性别特征提取分类模型下的性别分类器。
本发明的进一步改进在于，所述步骤S302中，将所采集的指纹图像输入至所述性别特征提取分类模型后，将所述性别特征提取分类模型输出的矩阵,通过取最大概率值所在的位置作为性别标签，进而将指纹图像信息转化成表示性别特征的性别标签以实现性别的分类表达，从而建立所述性别分类器。
本发明的进一步改进在于，所述步骤S4包括以下子步骤：
步骤S401，随机采集用户的指纹图像测试数据；
步骤S402，将所述指纹图像测试数据输入所述性别分类器中；
步骤，S403，输出性别识别结果。
本发明的进一步改进在于，所述步骤S401中所采集的指纹图像测试数据的量小于所述步骤S1中所采集的指纹图像数据的数量。
本发明的进一步改进在于，所述步骤S1中所采集的指纹图像数据的数量为采集同一用户的同一个手指至少400张指纹图像。
本发明还提供一种基于3D指纹图像的性别识别系统，采用了如上所述的基于3D指纹图像的性别识别方法。
与现有技术相比，本发明的有益效果在于：利用3D指纹图像对性别特征进行分析和识别，证实了指纹内部性别特征的存在，避免了指纹图像的表面环境对其识别的影响，提供了更高正确率和精确度的性别判别方案，提高了抗噪性能，对于法医人类学的性别分类、跟踪与标识未知人群以及人口数据普查等具有非常重要的意义。
附图说明
图1是本发明一种实施例的工作流程示意图；
图2是本发明一种实施例的指纹性别特征的提取训练过程示意图。
具体实施方式
下面结合附图，对本发明的较优的实施例作进一步的详细说明。
如图1所示，本例提供一种基于3D指纹图像的性别识别方法，包括以下步骤：
步骤S1，基于光学相干断层扫描采集用户的指纹图像数据；
步骤S2，根据所述指纹图像数据，提取并训练所述指纹图像数据中的性别特征；
步骤S3，搭建性别特征提取分类模型；
步骤S4，通过所性别特征提取分类模型执行性别的识别和预测。
本例所述步骤S1中，采集OCT(光学相干断层扫描)的指纹图像作为训练数据，优选为集同一用户的同一个手指至少400张的指纹图像作为指纹图像数据，以保证训练数据的准确性。
本例所述步骤S2包括以下子步骤：
步骤S201，针对所述指纹图像数据搭建性别特征提取分类网络；
步骤S202，提取所述指纹图像数据中的性别特征；
步骤S203，训练所述指纹图像性别特征下的分类网络。
本例所述步骤S201使用现有机器学习算法或深度学习算法作为模型进行性别特征提取分类模型的搭建。本例所述步骤S202中，提取所述指纹图像数据中的性别特征包括提取脊谷线在表皮层和乳头层中变换区域的形状，即提取脊谷线在表皮层和乳头层中的形状，以及提取在表皮层和乳头层之间的变换形状，进而通过训练来实现性别分类。
事实上，手指存在许多性别特征，比如可以使用指纹表面脊线密度识别性别，脊线密度较低的通常为男性，而具有较高脊线密度的多为女性。另外，手指表面的汗液的氨基酸含量同样可以用于性别的判别；不过如果利用现有2D指纹图像来实现性别分 类，其正确率有限；而指纹的性别分类对于法医人类学的性别分类、跟踪与标识未知人群和人口数据普查等具有重要意义。
因此，本例基于光学相干断层扫描采集用户的指纹图像数据，优选为单人400张以上，通过OCT 3D指纹图像的多张纵向指尖图像的截面图构成指纹图像数据，通过这种多张单指尖的3D扫描图像进行性别判别的多次验证，有效提高了性别识别的抗噪性和准确性。
更为具体的，所述步骤S201中，在机器学习中，如支持向量机(SVM)，K-最近邻算法(KNN)等算法，寻找所输入的OCT3D指纹图像与性别信息之间的映射关系。在深度学习中，使用如AlexNet、VGG-Net或ResNet等网络框架实现性别特征的提取功能，即通过逐层的卷积层、池化层以及批量归一化等计算层搭建性别特征提取网络，提取所输入的OCT3D指纹图像中的性别特征，即脊谷线在表皮层和乳头层中变换区域的不同形状，从而实现性别分类。
本例在此以使用17层ResNet网络为例进行性别特征提取分类网络的搭建，并对网络模型进行了一定的修改，使网络模型适应当前图像的输入维度，并借鉴了Google-net Inception的网络结构，利用1*A和A*1代替了A*A的传统卷积结构，缩小了网络结构的存储空间占用。
本例所述步骤S203中，通过深度学习算法训练所述指纹图像性别特征下的分类网络；即所述步骤S203为通过机器学习和深度学习算法(性别分类网络)对所述步骤S202所提取的指纹图像数据中的性别特征进行训练的过程，通过反复迭代实现更新和训练。
关于所述指纹图像性别特征提取模型，本例针对指纹图像提取具有性别标签，这一过程通过定义指纹信息场为M、性别判别器为G以及指纹特征提取器为F来实现，则搭建的性别特征提取分类网络的模型可定义为：min Remp(G)s.t.G＝F(M)，其中，Remp为经验风险泛函，专利以使用cross-entropy为例作为Remp，以ResNet深度学习算法为例作为指纹特征提取器F，在实验中，我们将指纹信息场M近似为指纹内部图像，即，M～Image。其中，所述性别特征提取模型公式min Remp(G)s.t.G＝F(M)的意思为：在指纹特征提取器F(M)中提取出指纹信息场M(也称指纹生物特征信息场M)中的某种特征满足性别判别器G的约束条件下，使得性别判别器G的经验风险泛函Remp最小；s.t.G指的是(subject to性别判别器G)译为受约束于离散化特征判别器G。
通过求解该近似优化的过程，即可得到性别判别器G。定义性别判别器G的正确率为
Figure PCTCN2019098195-appb-000001
T M为指纹图像数据的测试集的样本，sgn为符号函数，G(M)为在指纹信息场M下的性别判别器G，L M为指纹图像数据的测试集的性别标签，即
Figure PCTCN2019098195-appb-000002
如图2所示，本例所述步骤S203中，采用两块以上的显卡并行计算和训练，先通过对每一块显卡进行机器学习和训练，实现更新迭代，并计算两块以上的显卡的均值以实现变量的更新；然后将更新后的变量参数返回到至CPU中，以输入至每一块显卡中，实现反复迭代。
更为具体的，本例所述步骤S203训练过程如图2所示，采用两块以上的显卡并行计算和训练，比如使用三块显卡进行计算为例。其中将性别分类网络放至每张显卡中进行训练，得到的结果为当前训练训练次数下单显卡的梯度值、损失函数值和判别正确率，将上述每张显卡中的3个结果返回到CPU中并分别求均值，得到显卡计算出的梯度值、损失函数值和判别正确率，在CPU中根据显卡的计算出的梯度值进行反向传播，对性别判别模型中的参数(参数包括权重和偏置等)逐层求偏导数，进行参数更新，并将参数更新后的分类网络再放入显卡中继续训练，实现网络参数的更新迭代。
本例所述步骤S203采用多显卡并行计算加快网络训练速度，在CPU中实现具体算法中相关参数的更新迭代并输入GPU中，在三块GPU中实现机器学习和深度学习算法(性别分类网络)的训练过程，即实现变量的更新，并将计算的参数返回到CPU中，反复迭代，进而能够有效提高训练的准确性和精确度。
本例所述步骤S3包括以下子步骤：
步骤S301，保存训练好之后的指纹图像性别特征下的分类网络作为性别特征提取分类模型；
步骤S302，搭建所述性别特征提取分类模型下的性别分类器。
本例所述步骤S302中，将所采集的指纹图像输入至所述性别特征提取分类模型后，将所述性别特征提取分类模型输出的矩阵,通过取最大概率值所在的位置作为性别标签，进而将指纹图像信息转化成表示性别特征的性别标签以实现性别的分类表达，从而建立所述性别分类器。
即，本例所述步骤S302在步骤S301所保存的性别特征提取分类模型的基础上， 构建性别分类器，将所输入的OCT3D指纹图像输入分类模型后，对输出的分类结果进行具体运算，即将分类模型输出的稀疏矩阵,通过取稀疏矩阵中数值(即概率值)最大所在的位置(对应一种性别标签)，即可转化成表示性别的可理解的标签，实现性别的分类表达。
本例所述步骤S4包括以下子步骤：
步骤S401，随机采集用户的指纹图像测试数据；
步骤S402，将所述指纹图像测试数据输入所述性别分类器中；
步骤，S403，输出性别识别结果。
本例所述步骤S401中所采集的指纹图像测试数据的量小于所述步骤S1中所采集的指纹图像数据的数量，即小于训练时的单人400张的数量。
通过实验证明，本例所述步骤S4中，经过多人多组验证，在单人单张OCT 3D指纹图像的输入情况下，此性别分类器最终可以达到99.3％的识别准确率；在单人三张以上OCT 3D指纹图像的输入情况下，此性别分类器识别准确率可以无限逼近100％。因此，本例能够有效提高性别识别的准确性和精准度。
此外，本例所述步骤S203中使用多显卡并行训练方法可大大缩短训练时间；所述步骤S201也可以使用其他现有特征提方法(包括深度学习方法、机器学习算法)代替。
本例还提供一种基于3D指纹图像的性别识别系统，采用了如上所述的基于3D指纹图像的性别识别方法。
综上所述，本例利用3D指纹图像对性别特征进行分析和识别，证实了指纹内部性别特征的存在，避免了指纹图像的表面环境对其识别的影响，提供了更高正确率和精确度的性别判别方案，提高了抗噪性能，对于法医人类学的性别分类、跟踪与标识未知人群以及人口数据普查等具有非常重要的意义。
以上内容是结合具体的优选实施方式对本发明所作的进一步详细说明，不能认定本发明的具体实施只局限于这些说明。对于本发明所属技术领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干简单推演或替换，都应当视为属于本发明的保护范围。

Claims (10)

  1. 一种基于3D指纹图像的性别识别方法，其特征在于，包括以下步骤：
    步骤S1，基于光学相干断层扫描采集用户的指纹图像数据；
    步骤S2，根据所述指纹图像数据，提取并训练所述指纹图像数据中的性别特征；
    步骤S3，搭建性别特征提取分类模型；
    步骤S4，通过所性别特征提取分类模型执行性别的识别和预测。
  2. 根据权利要求1所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S2包括以下子步骤：
    步骤S201，针对所述指纹图像数据搭建性别特征提取分类网络；
    步骤S202，提取所述指纹图像数据中的性别特征；
    步骤S203，训练所述指纹图像性别特征下的分类网络。
  3. 根据权利要求2所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S202中，提取所述指纹图像数据中的性别特征包括提取脊谷线在表皮层和乳头层中变换区域的形状，进而通过训练来实现性别分类。
  4. 根据权利要求3所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S203中，采用两块以上的显卡并行计算和训练，其中，先将性别特征提取分类网络放至每张显卡中进行训练，得到的结果为当前训练次数下单显卡的梯度值、损失函数值和判别正确率；然后将上述每张显卡中的梯度值、损失函数值和判别正确率这3个结果返回到CPU中并分别求均值，进而得到显卡计算出的梯度平均值、损失函数平均值和判别正确率平均值；最后，在CPU中根据显卡的计算出的梯度平均值进行反向传播，对性别判别模型中的参数逐层求偏导数，进行参数更新，并将参数更新后的性别特征提取分类网络再放入显卡中继续训练，以实现网络参数的更新迭代。
  5. 根据权利要求1至4任意一项所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S3包括以下子步骤：
    步骤S301，保存训练好之后的指纹图像性别特征下的分类网络作为性别特征提取分类模型；
    步骤S302，搭建所述性别特征提取分类模型下的性别分类器。
  6. 根据权利要求5所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S302中，将所采集的指纹图像输入至所述性别特征提取分类模型后，将所述性别特征提取分类模型输出的矩阵,通过取最大概率值所在的位置作为性别标签，进而将指 纹图像信息转化成表示性别特征的性别标签以实现性别的分类表达，从而建立所述性别分类器。
  7. 根据权利要求5所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S4包括以下子步骤：
    步骤S401，随机采集用户的指纹图像测试数据；
    步骤S402，将所述指纹图像测试数据输入所述性别分类器中；
    步骤，S403，输出性别识别结果。
  8. 根据权利要求7所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S401中所采集的指纹图像测试数据的量小于所述步骤S1中所采集的指纹图像数据的数量。
  9. 根据权利要求8所述的基于3D指纹图像的性别识别方法，其特征在于，所述步骤S1中所采集的指纹图像数据的数量为采集同一用户的同一个手指至少400张指纹图像。
  10. 一种基于3D指纹图像的性别识别系统，其特征在于，采用了如权利要求1至9任意一项所述的基于3D指纹图像的性别识别方法。
PCT/CN2019/098195 2018-11-30 2019-07-29 一种基于3d指纹图像的性别识别方法及系统 Ceased WO2020107922A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811455190.0A CN109508692B (zh) 2018-11-30 2018-11-30 一种基于3d指纹图像的性别识别方法及系统
CN201811455190.0 2018-11-30

Publications (1)

Publication Number Publication Date
WO2020107922A1 true WO2020107922A1 (zh) 2020-06-04

Family

ID=65750004

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/098195 Ceased WO2020107922A1 (zh) 2018-11-30 2019-07-29 一种基于3d指纹图像的性别识别方法及系统

Country Status (2)

Country Link
CN (1) CN109508692B (zh)
WO (1) WO2020107922A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112115965A (zh) * 2020-08-04 2020-12-22 西安交通大学 一种基于svm的被动操作系统识别方法、存储介质及设备
CN112966610A (zh) * 2021-03-09 2021-06-15 金陵科技学院 一种基于增量学习的电梯智能指纹识别系统
CN114980122A (zh) * 2022-05-05 2022-08-30 南京航空航天大学 一种小样本射频指纹智能识别系统与方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508692B (zh) * 2018-11-30 2020-06-16 深圳大学 一种基于3d指纹图像的性别识别方法及系统
CN110309738B (zh) * 2019-06-17 2022-09-30 深圳大学 一种对oct指纹图像进行标注的方法
CN110909614B (zh) * 2019-10-28 2023-04-07 陕西科技大学 一种将自编码网络用于指纹性别分类的方法
CN116189239A (zh) * 2022-11-17 2023-05-30 深圳大学 指纹性别的分类方法、装置、计算机设备及存储介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102266241A (zh) * 2011-08-05 2011-12-07 上海交通大学 融合人脸和指纹视觉信息的协同性别识别方法
CN102902986A (zh) * 2012-06-13 2013-01-30 上海汇纳网络信息科技有限公司 自动性别识别系统及方法
US20140241596A1 (en) * 2013-02-28 2014-08-28 The Regents Of The University Of California Apparatus and Method for Capturing a Vital Vascular Fingerprint
CN105303179A (zh) * 2015-10-28 2016-02-03 小米科技有限责任公司 指纹识别方法、装置
CN107657239A (zh) * 2017-09-30 2018-02-02 清华大学深圳研究生院 掌纹图像性别分类方法及装置、计算机装置及可读存储介质
CN107958217A (zh) * 2017-11-28 2018-04-24 广州麦仑信息科技有限公司 一种基于深度学习的指纹分类识别系统及方法
CN108446633A (zh) * 2018-03-20 2018-08-24 深圳大学 一种新型指纹自动防伪与活体检测的方法、系统及装置
CN109508692A (zh) * 2018-11-30 2019-03-22 深圳大学 一种基于3d指纹图像的性别识别方法及系统

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102266241A (zh) * 2011-08-05 2011-12-07 上海交通大学 融合人脸和指纹视觉信息的协同性别识别方法
CN102902986A (zh) * 2012-06-13 2013-01-30 上海汇纳网络信息科技有限公司 自动性别识别系统及方法
US20140241596A1 (en) * 2013-02-28 2014-08-28 The Regents Of The University Of California Apparatus and Method for Capturing a Vital Vascular Fingerprint
CN105303179A (zh) * 2015-10-28 2016-02-03 小米科技有限责任公司 指纹识别方法、装置
CN107657239A (zh) * 2017-09-30 2018-02-02 清华大学深圳研究生院 掌纹图像性别分类方法及装置、计算机装置及可读存储介质
CN107958217A (zh) * 2017-11-28 2018-04-24 广州麦仑信息科技有限公司 一种基于深度学习的指纹分类识别系统及方法
CN108446633A (zh) * 2018-03-20 2018-08-24 深圳大学 一种新型指纹自动防伪与活体检测的方法、系统及装置
CN109508692A (zh) * 2018-11-30 2019-03-22 深圳大学 一种基于3d指纹图像的性别识别方法及系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
FENG LI ET AL: "Study on Novel Curvature Features for 3D Fingerprint Recognition", NEUROCOMPUTING, vol. 168, 30 November 2015 (2015-11-30), pages 599 - 608, XP055711088, ISSN: 0925-2312, DOI: 10.1016/j.neucom.2015.05.065 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112115965A (zh) * 2020-08-04 2020-12-22 西安交通大学 一种基于svm的被动操作系统识别方法、存储介质及设备
CN112966610A (zh) * 2021-03-09 2021-06-15 金陵科技学院 一种基于增量学习的电梯智能指纹识别系统
CN114980122A (zh) * 2022-05-05 2022-08-30 南京航空航天大学 一种小样本射频指纹智能识别系统与方法

Also Published As

Publication number Publication date
CN109508692A (zh) 2019-03-22
CN109508692B (zh) 2020-06-16

Similar Documents

Publication Publication Date Title
CN109508692B (zh) 一种基于3d指纹图像的性别识别方法及系统
CN107748900B (zh) 基于区分性卷积神经网络的乳腺肿瘤分类装置和存储介质
CN102637251B (zh) 基于参考特征的人脸识别方法
CN109657567B (zh) 一种基于3d指纹图像的弱监督的特征分析方法及系统
CN107392107B (zh) 一种基于异构张量分解的人脸特征提取方法
CN101794372B (zh) 基于频域分析的步态特征表示及识别方法
WO2022001123A1 (zh) 关键点检测方法、装置、电子设备及存储介质
WO2016150240A1 (zh) 身份认证方法和装置
CN108550156A (zh) 用于图像引导治疗计划的结构形状的自动描绘的方法和装置
Cao et al. A novel ant colony optimization algorithm for large-distorted fingerprint matching
CN116645705B (zh) 基于轻量化网络的近红外掌静脉roi提取方法及系统
CN107392187A (zh) 一种基于梯度方向直方图的人脸活体检测方法
CN104700089A (zh) 一种基于Gabor小波和SB2DLPP的人脸识别方法
CN110633634A (zh) 一种关于中医体质的脸型分类方法、系统和计算机可读存储介质
CN116797817A (zh) 基于自监督图卷积模型的自闭症疾病预测技术
CN114913610A (zh) 一种基于指纹和指静脉的多模态识别方法
CN105279508A (zh) 一种基于kap有向图模型的医学图像分类方法
CN118658185A (zh) 结合CNN与Transformer的指掌静脉认证方法
Wu et al. Fusion recognition of palmprint and palm vein based on modal correlation
CN110969101A (zh) 一种基于hog和特征描述子的人脸检测与跟踪方法
CN115862055B (zh) 基于对比学习和对抗训练的行人重识别方法及装置
Ranjan et al. Hierarchical approach for breast cancer histopathology images classification
Yuan et al. Fingerprint liveness detection using an improved CNN with the spatial pyramid pooling structure
Doroz et al. An accurate fingerprint reference point determination method based on curvature estimation of separated ridges
de Almeida et al. Texture classification based on co-occurrence matrix and self-organizing map

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19891630

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 09.09.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19891630

Country of ref document: EP

Kind code of ref document: A1