WO2018023884A1 - 一种身份识别装置及方法 - Google Patents

一种身份识别装置及方法 Download PDF

Info

Publication number
WO2018023884A1
WO2018023884A1 PCT/CN2016/103024 CN2016103024W WO2018023884A1 WO 2018023884 A1 WO2018023884 A1 WO 2018023884A1 CN 2016103024 W CN2016103024 W CN 2016103024W WO 2018023884 A1 WO2018023884 A1 WO 2018023884A1
Authority
WO
WIPO (PCT)
Prior art keywords
group delay
fingerprint
identity recognition
finger
biometric
Prior art date
Application number
PCT/CN2016/103024
Other languages
English (en)
French (fr)
Inventor
李景振
聂泽东
刘宇航
王磊
Original Assignee
深圳先进技术研究院
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 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2018023884A1 publication Critical patent/WO2018023884A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V40/1347—Preprocessing; Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V40/13—Sensors therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/70—Multimodal biometrics, e.g. combining information from different biometric modalities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the present invention relates to the field of biometrics, and in particular, to an identity recognition apparatus and method.
  • Biometric technology refers to a recognition technology that uses the inherent biometric characteristics of the human body to perform identity authentication.
  • the biometric features that have been used for biometric identification include fingerprints, palm prints, face shapes, irises, pulses, sounds, and brains. Electricity, ECG, genes, etc., this technology has the advantages of being easy to forget, not easy to forge or stolen, "carrying" with you and being available anytime and anywhere, and is more secure, confidential and convenient than traditional identification methods.
  • biometric technologies are based on a single biometric feature, such as fingerprints, irises, sounds, faces, etc., but each biometric has more or less defects, especially the vulnerability.
  • the influence of the external environment and the damage of biological characteristics has led to a decrease in the recognition rate.
  • the fingerprint is easy to wear, it is difficult to identify after being wet; the iris recognition result is closely related to the illumination and angle; face recognition can not accurately identify the face and the face after plastic surgery.
  • researchers have proposed a multi-modal biometric identification method.
  • Chinese patent CN105117697A proposes a fingerprint recognition method, a fingerprint recognition device and a terminal device thereof, which emit red light, infrared light and green light through a reflection to an object to be detected.
  • the change of the light determines whether the object to be detected is a living body, and after determining the living body, the identity of the user is identified by collecting the fingerprint image. Therefore, strictly speaking, the patent uses a single biometric for identification.
  • Chinese patent CN102542263A proposes a multimodal identity authentication device and method based on finger biometrics.
  • the patent mainly collects biometric information of a user by collecting a finger vein image, a fingerprint image and a fingerprint image.
  • the three kinds of biometric information are collected and identified in isolation, and the three modes are effectively merged, and the recognition rate needs to be further improved.
  • Chinese patent CN1758263 proposes a multimodal identification method based on score difference weighted fusion.
  • the patent proposes a multi-modal identity recognition using a score difference weighted fusion algorithm.
  • the algorithm does not consider the influence of other environmental factors on identity recognition.
  • the present invention provides an identification device and method, which aim to solve at least one of the above technical problems in the prior art.
  • the present invention provides the following technical solutions:
  • An identification device includes a biometric signal acquisition module, a biometrics extraction module, and a biometric identification module;
  • the biometric signal acquisition module is configured to collect a fingerprint image and a finger group delay curve;
  • the biometric extraction module is used for And extracting a fingerprint feature value and a group delay feature value according to the fingerprint image and the finger group delay curve respectively;
  • the biometric recognition module is configured to construct an identity recognition model by training the sample set, and using the convolutional neural network algorithm to extract the extracted The fingerprint feature value and the group delay feature value are cross-verified with the identity recognition model to implement user identification.
  • the biometric signal acquisition module includes a fingerprint collection unit and a dielectric spectrum acquisition unit, and the fingerprint collection unit is a fingerprint sensor, configured to collect a fingerprint image of the finger;
  • the electric spectrum acquisition unit includes a signal transmitting electrode, a signal receiving electrode, and a signal a source, a receiver, the signal transmitting electrode and a signal receiving electrode are respectively located at two ends of the fingerprint sensor; when the finger presses the signal transmitting electrode and the signal receiving electrode, the signal source generates a sine wave, and the signal transmitting electrode is sinusoidal The wave is coupled to the finger of the user, the signal receiving electrode receiving a signal group delay profile after the sine wave passes through the finger and is stored in the receiver.
  • the technical solution adopted by the embodiment of the present invention further includes a signal pre-processing module, where the signal pre-processing module includes:
  • the fingerprint preprocessing unit is configured to perform a Fourier transform filtering process on the fingerprint image, analyze the sharpness of the fingerprint image by using a gradient algorithm, and perform a binarization process on the fingerprint image by using a dynamic binarization algorithm;
  • the group delay curve preprocessing unit is configured to convert the group delay curve into a group delay image, and filter the group delay image by using a non-sand particle filtering algorithm.
  • the technical solution adopted by the embodiment of the present invention further includes an environmental parameter monitoring module, where the environmental parameter monitoring module is configured to collect user environment parameter information; the environmental parameter monitoring module includes a humidity sensing unit and a temperature sensing unit, and the humidity transmission
  • the sensing unit includes a humidity sensor and a grease sensor.
  • the humidity sensor and the grease sensor are respectively used for collecting moisture distribution and greasy degree information of the user's finger.
  • the temperature sensing unit is a body temperature sensor for collecting current body temperature information of the user.
  • the biometric feature extraction module includes:
  • Fingerprint feature extraction unit for segmenting the fingerprint image by using a log-Gabor filter, extracting a fingerprint grain direction feature, and extracting a point feature and a line feature of the fingerprint according to the grain direction;
  • Group delay feature extraction unit used to extract the group delay curvature radius in different frequency bands, the group delay average value in different frequency bands, and extract the spectral characteristics of the group delay curve.
  • the biometric identification module includes a training unit and a testing unit
  • the training method of the training unit includes: collecting fingerprints of multiple volunteers in different time periods Image, finger group delay curve and environmental parameter information, the collected fingerprint image, finger group delay curve and environmental parameter information are used as training sample sets, and the training sample set is iteratively calculated by convolutional neural network algorithm, and output and An identification model related to the environmental parameter information; the identification model is stored in the biometric database.
  • the test mode of the test unit includes: cross-validating the fingerprint feature value extracted by the biometric feature extraction module, the group delay feature value, and the environment parameter information collected by the environmental parameter monitoring module with the identity recognition model in the biometric database to determine The extracted fingerprint feature value, the group delay feature value and the correlation between the environment parameter information and the fingerprint image, the finger group delay curve and the environmental parameter information in the identity recognition model, and output the identity recognition result to realize the user identity recognition.
  • an identity recognition method including the following steps:
  • Step a collecting a fingerprint image and a finger group delay curve
  • Step b extracting fingerprint feature values and group delay feature values according to the fingerprint image and the finger group delay curve respectively;
  • Step c construct an identity recognition model by training the sample set, and cross-verify the extracted fingerprint feature value and the group delay feature value with the identity recognition model by using a convolutional neural network algorithm to implement user identification.
  • step a further includes:
  • Step a1 collecting environmental parameter information;
  • the environmental parameter information includes moisture moisture, greasy degree, and body temperature information;
  • Step a2 After performing Fourier transform filtering on the fingerprint image, the gradient algorithm is used to analyze the sharpness of the fingerprint image, and the dynamic binarization algorithm is used to binarize the fingerprint image;
  • Step a3 Converting the group delay curve into a group delay image, and filtering the group delay image by using the unscented particle filtering algorithm.
  • the technical solution adopted by the embodiment of the present invention further includes: in the step b, the extracting biot
  • the locating information specifically includes: using a log-Gabor filter to segment the fingerprint image, extracting the fingerprint grain direction feature, extracting the point feature and the line feature of the fingerprint according to the grain direction; extracting the group delay curvature radius and the group delay average value in different frequency bands And extract the spectral characteristics of the group delay curve.
  • the identity recognition includes:
  • Step c1 collecting fingerprint images, finger group delay curves and environmental parameter information of multiple volunteers in different time periods, using the collected fingerprint image, finger group delay curve and environmental parameter information as a training sample set, using convolutional nerves
  • the network algorithm performs iterative calculation on the training sample set, and outputs an identity recognition model related to the environment parameter information, and stores the identity recognition model in the biometric database;
  • Step c2 cross-validating the extracted fingerprint feature value, the group delay feature value, and the collected environmental parameter information with the identity recognition model in the biometric database, and determining the extracted fingerprint feature value, group delay feature value, and environment parameter information. The degree of correlation between the fingerprint image, the finger group delay curve and the environmental parameter information in the identification model, and output the identification result to realize the user identification.
  • the embodiment of the present invention has the beneficial effects that the identity recognition apparatus and method of the embodiment of the present invention collects the fingerprint feature of the user and the characteristic information of the finger dielectric spectrum for identification, and at the same time, in order to reduce the external The interference between the environment and the internal environment on the identification, while collecting the fingerprint characteristics, collecting the environmental parameter information such as the water distribution, greasy degree and body temperature of the user's finger, and correcting the identification through the environmental parameter information, effectively improving the accuracy of the identification. Sex and reliability.
  • the multi-modal fusion identification algorithm based on convolutional neural network is used for identification, which further improves the accuracy and reliability of identity recognition.
  • FIG. 1 is a schematic structural diagram of an identity recognition apparatus according to an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of a biometric signal acquisition module according to an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of identity recognition based on a convolutional neural network algorithm according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a process of identifying a biometric identification module according to an embodiment of the present invention.
  • FIG. 5 is a flowchart of an identity recognition method according to an embodiment of the present invention.
  • the identity recognition device of the embodiment of the invention comprises a biometric signal acquisition module, an environmental parameter monitoring module, a signal preprocessing module, a biometrics extraction module, a biometric identification module and a biometric database;
  • the biometric signal acquisition module is used for collecting the finger The fingerprint image and the signal group delay curve after the sine wave passes through the finger;
  • the environmental parameter monitoring module is used to collect environmental parameter information such as moisture moisture, greasy degree and body temperature of the user finger;
  • the signal preprocessing module is used for the fingerprint image and the finger part
  • the group delay curve is preprocessed to obtain a binarized fingerprint image and a finger group delay image;
  • the biometric extraction module is configured to respectively extract fingerprint feature values and group delay feature values according to the binarized fingerprint image and the finger group delayed image;
  • the feature recognition module is used to construct an identity recognition model, and the identity recognition model is stored in the biometric database; and the fingerprint feature value, the group delay feature value and the environmental parameter
  • FIG. 2 is a schematic structural diagram of a biometric signal acquisition module according to an embodiment of the present invention.
  • the biometric signal acquisition module includes a fingerprint collection unit and a dielectric spectrum acquisition unit;
  • the fingerprint collection unit is configured to collect the fingerprint image of the finger; wherein the fingerprint collection unit F is a fingerprint sensor capable of continuous and rapid acquisition.
  • the fingerprint sensor starts to continuously collect the user at a speed of 4 frames/second.
  • the fingerprint of the fingerprint image is automatically ended after the acquisition of the 20-frame fingerprint image. It can be understood that in other embodiments of the present invention, the collection speed and the number of the fingerprint image may also be set according to actual applications.
  • the dielectric spectrum acquisition unit is configured to acquire a signal group delay curve after the sine wave passes through the finger; specifically, the dielectric spectrum acquisition unit includes a signal transmitting electrode A, a signal receiving electrode B, a signal source (not shown), and a receiver ( The signal transmitting electrode A and the signal receiving electrode B are respectively located at both ends of the fingerprint sensor.
  • the signal source is spaced at a frequency of 1 MHz, sequentially generates a sine wave of 1 MHz to 200 MHz in 5 seconds, and couples the sine wave to the user through the signal transmitting electrode A. Fingers.
  • the dielectric constant of each person's finger is also different, causing the group delay of the signal to change when the sine wave passes through the finger.
  • the signal receiving electrode B is delayed by receiving the finger signal group at different frequencies and stored in the receiver, thereby collecting the dielectric spectrum information of the user's finger. It can be understood that in other embodiments of the present invention, the signal frequency generated by the dielectric spectrum acquisition unit can also be set according to an actual application.
  • the fingerprint image acquired by the fingerprint acquisition unit and the finger group delay curve obtained by the dielectric spectrum acquisition unit are usually accompanied by various noises.
  • the invention preprocesses the fingerprint image and the group delay curve by the signal preprocessing module, so that the low quality fingerprint image and the group delay curve become clearer and fuller, weakening or eliminating the influence of various noise interference factors, so as to be able to extract or Accurate feature information is identified.
  • the signal pre-processing module includes a fingerprint pre-processing unit and a group delay curve pre-processing unit;
  • the fingerprint preprocessing unit After the fingerprint preprocessing unit is used to perform Fourier transform filtering on the fingerprint image, the gradient algorithm is used to analyze the sharpness of the fingerprint image, and the fingerprint image with the highest definition is selected, and the fingerprint image is performed by using a dynamic binarization algorithm. Binary processing.
  • the processing manner of the fingerprint preprocessing unit for performing Fourier transform filtering processing on the fingerprint image specifically includes:
  • N 20, 1 ⁇ n ⁇ N, and c n represents the nth fingerprint image.
  • Analysis method of fingerprint image preprocessing unit using gradient algorithm to analyze the sharpness of fingerprint image Specifically include:
  • the gradient of the fingerprint image c 1 is acquired by a gradient algorithm, and the gradient of the fingerprint image c 1 can be expressed as:
  • f(x, y) is a pixel point located at the (x, y) position.
  • the processing method of the fingerprint preprocessing unit using the dynamic binarization algorithm to perform binarization processing on the fingerprint image specifically includes:
  • f(x, y) is the gray value of the pixel at the (x, y) position
  • T t is a fixed reference threshold, the size of which is determined according to the gray scale map of the image
  • ⁇ is the misjudgment correction factor
  • the fingerprint image is binarized according to formula (8), and the binarized fingerprint image is collected and saved.
  • the group delay curve preprocessing unit is configured to convert the group delay curve into a corresponding group delay image, and filter the group delay image by using the unscented particle filtering algorithm;
  • the group delay curve preprocessing unit converts the group delay curve into a corresponding group delay image by specifically: establishing an M ⁇ N blank matrix; and scanning the group delay curve from top to bottom according to the linear mapping method.
  • a group delay curve of 1 MHz to 200 MHz is converted into a group delay image of size M ⁇ N.
  • the processing method of filtering the group delay image by the group delay curve preprocessing unit is specifically:
  • the biometric extraction module is configured to extract a fingerprint feature value of the user and a group delay feature value according to the binarized fingerprint image and the group delay image; specifically, the biometric feature extraction module includes a fingerprint feature extraction unit and a group delay feature extraction unit;
  • the fingerprint feature extraction unit is configured to segment the fingerprint image by using a log-Gabor filter, extract a fingerprint grain direction feature, and extract a point feature and a line feature of the fingerprint according to the grain direction;
  • the segmentation manner of the fingerprint feature extraction unit for segmenting the fingerprint image specifically includes: dividing the fingerprint image into a plurality of non-overlapping small ones according to the local directionality of the fingerprint and the statistical characteristics of the pattern, and using a log-Gabor filter. Block; calculate the feature vector of each small block, and determine whether a small block meets the feature extraction requirement according to the feature vector, and if not, discard the small block.
  • the fingerprint feature extraction unit extracts the fingerprint grain direction feature by specifically extracting the grain direction O(i,j) of the square region with W as the side centered on the pixel (x, y) in a small block. And analyze the texture direction of each pixel separately:
  • V y 2G x (x,y)-G y (x,y) (11)
  • G x (x, y), G y (x, y) are gradients at the pixel (x, y), respectively.
  • the method for extracting the point feature and the line feature of the fingerprint feature extraction unit according to the direction of the grain is specifically: obtaining the point feature and the line feature of the fingerprint by tracking the grain direction, including the coordinates, direction, type, length, and maximum of the start point and the end point. Information such as curvature, starting point and ending point order; establishing the adjacent topological relationship between the lines by tracking the fingerprint nodes, and establishing the association order relationship between the nodes through the topological relationship of the lines.
  • the group delay feature extraction unit is used to extract group delay eigenvalues of group delay curvature radii in different frequency bands and group delay average values in different frequency bands, and extract the spectral characteristics of the group delay curve.
  • the group delay delay feature extraction unit collects the group delay curvature radius by:
  • the collection method of the group delay average of the group delay feature extraction unit includes:
  • the group delay feature extraction unit extracts the spectral characteristics of the group delay curve by adopting the Fourier transform principle and performing Fourier transform on the group delay curve to obtain the distribution of the group delay curve in the frequency domain.
  • the acquisition formula is:
  • the environmental parameter monitoring module is used to collect environmental parameter information such as moisture moisture, greasy degree and user body temperature of the user's finger, and the collected environmental parameter information is used to correct the biometric identification module to improve the reliability of the identification.
  • environmental parameter information such as moisture moisture, greasy degree and user body temperature of the user's finger
  • biometric identification module is used to improve the reliability of the identification.
  • fingerprint-based identification is greatly reduced when the user's finger is wet or greasy.
  • the dielectric constant of the user's finger may change, resulting in a change in the dielectric spectrum of the finger, resulting in a change in the group delay characteristics of the finger, affecting the identification.
  • the embodiment of the present invention can effectively improve the accuracy of the identity recognition by introducing the environmental parameter monitoring module.
  • the environmental parameter monitoring module includes a humidity sensing unit and a temperature sensing unit
  • the humidity sensing unit includes a humidity sensor and a grease sensor for collecting information such as moisture distribution and greasy degree of the user's finger.
  • the temperature sensing unit is a body temperature sensor for collecting the current body temperature information of the user, and the detection range of the body temperature sensor is 34 ° C - 39 ° C.
  • the biometric identification module is used to iteratively calculate the training sample set (including multiple fingerprint images and finger group delay curves and environmental parameter information) by using a convolutional neural network algorithm to obtain an identification model, and extract the extracted biometric information with And the environmental parameter information and the identification model are cross-validated, and the correlation between the extracted biometric information and the fingerprint image and the finger group delay curve in the identification model is determined, and whether the two biometric information are from the same user is determined. To achieve user identification.
  • FIG. 3 is a schematic diagram of identity recognition based on a convolutional neural network algorithm according to an embodiment of the present invention.
  • the convolutional neural network algorithm structure includes an input layer, a convolution layer, a sub-sampling layer, a fully connected layer, and an output layer.
  • the layers are set as follows:
  • the number of input layer nodes is set to 16 ⁇ 16, a total of 256 nodes:
  • a 01n represents the direction of the grain
  • a 02n represents the length of the line
  • a 03n represents the point feature in the texture
  • a 04n represents the line feature in the texture
  • a 05n represents the degree of association between the texture nodes
  • a 06n Representing the adjacent topological relationship between the lines
  • a 07n represents the group delay curvature radius
  • a 08n represents the group delay curvature direction
  • a 09n represents the group delay average
  • a 10n represents the group delay of different frequencies in the time domain.
  • a 11n represents the spectral characteristics of the group delay
  • a 12n represents the weight of the group delay of different frequencies in the frequency domain
  • a 13n represents the humidity of the external environment
  • a 14n represents the moisture content of the surface of the user's finger
  • a 15n represents the surface of the finger of the user.
  • a 16n represents the body temperature.
  • the values of ⁇ a 01n , a 02n , a 03n ,..., a 12n ⁇ are derived from the calculation results of the biometric extraction module
  • the values of ⁇ a 13n , a 14n , a 15n , a 16n ⁇ are derived from Monitoring results of the environmental parameter monitoring module.
  • the number of nodes in the output layer is set to 2, which respectively represent two recognition results: (1) biometric matching, the user is a legitimate user; (2) the biometrics do not match, and the user is an illegal user.
  • each cell in the layer receives a set of cells in a small neighborhood in the previous layer as input, multiplies a trainable convolution kernel, adds an offset, and then outputs through the activation function. .
  • the convolutional layer is calculated as follows:
  • w is the convolution kernel of size J ⁇ I
  • function f is the activation function
  • b is the bias the amount.
  • the size of the convolution kernel is 5 ⁇ 5
  • the activation function always uses the tanh function
  • the offset is determined based on the empirical value.
  • the purpose of the sub-sampling layer is to perform sampling operations on each feature map obtained in the previous layer, so that the size of the feature map is reduced, and the network can be made to be invariant to the translation and scaling of the object, so that the network is more robust.
  • sample as follows:
  • x is a two-dimensional input quantity
  • y is an output obtained after sampling
  • S 1 ⁇ S 2 is a size of a sampling template.
  • the size of the sampling template is set to 2 ⁇ 2.
  • FIG. 4 is a schematic diagram of a biometric identification module identification process according to an embodiment of the present invention.
  • the biometric identification module of the embodiment of the invention comprises a training unit and a testing unit.
  • the training process of the training unit specifically includes: firstly collecting fingerprint images and finger group delay curves of a plurality of volunteers in different time periods (the embodiment of the present invention includes 200 fingerprint images and 200 finger group delay curves, and the specific collection quantity is Not limited to this).
  • the humidity parameter, the grease sensor and the body temperature sensor are used to collect the environmental parameter information such as the moisture, greasy degree and body temperature data of the finger surface of the volunteer (the surface moisture, greasy degree and body temperature data of the finger of the volunteer are collected 200 times in the embodiment of the invention).
  • the specific number of collections is not limited to this).
  • the collected fingerprint image, the finger group delay curve and the environmental parameter information are used as the training sample set, and the training sample set is preprocessed to divide the training sample set into fingerprint, group delay, moisture humidity, greasy degree and body temperature.
  • a small sample set and distortion processing of the sample set After the distortion processing is completed, the training sample set is subjected to multiple iteration calculations by using a convolutional neural network algorithm.
  • the iterative operation is stopped, and a The identification model related to the environmental parameter information (ie, the surface moisture moisture, greasyness, and user's body temperature of the user's finger) is stored in the biometric database to complete the training process of the identity recognition.
  • a The identification model related to the environmental parameter information ie, the surface moisture moisture, greasyness, and user's body temperature of the user's finger
  • the test process of the test unit specifically includes: the fingerprint feature value extracted by the biometric extraction module, The group delay eigenvalue and the environmental parameter information collected by the environmental parameter monitoring module are cross-validated with the identity recognition model to determine the extracted fingerprint feature value, the group delay feature value, and the environmental parameter information and the identity recognition model stored in the biometric database.
  • the degree of correlation between the fingerprint image, the finger group delay curve and the environmental parameter information determines whether the two biological features are from the same user, thereby realizing the user's identification.
  • FIG. 5 is a flowchart of an identity recognition method according to an embodiment of the present invention.
  • the identity recognition method of the embodiment of the present invention includes the following steps:
  • Step 100 collecting a fingerprint image of the finger and a signal group delay curve after the sine wave passes through the finger;
  • the fingerprint sensor collects the fingerprint image of the finger through the fingerprint sensor; when the user's finger presses the fingerprint sensor, the fingerprint sensor starts to continuously collect the fingerprint of the user at a speed of 4 frames/second, and collects 20 frames. After the fingerprint image, the fingerprint image is automatically ended.
  • the signal group delay curve after the sine wave passes through the finger is collected by the dielectric spectrum acquisition unit; the dielectric spectrum acquisition unit includes a signal transmitting electrode A, a signal receiving electrode B, a signal source, a receiver, a signal transmitting electrode A, and a signal receiving electrode. B is located at both ends of the fingerprint sensor.
  • the signal source When the finger is pressed to the signal transmitting electrode A and the signal receiving electrode B, the signal source is spaced at a frequency of 1 MHz, and a sine wave of 1 MHz-200 MHz is sequentially generated in 5 seconds, and is transmitted by signal.
  • Electrode A couples a sine wave to the user's finger. Due to the human body's differences, the dielectric constant of each person's finger is also different, causing the group delay of the signal to change when the sine wave passes through the finger.
  • the signal receiving electrode B is delayed by receiving the finger signal group at different frequencies and stored in the receiver, thereby collecting the dielectric spectrum information of the user's finger.
  • Step 200 collecting environmental parameter information such as moisture moisture, greasy degree, and body temperature of the user's finger;
  • step 200 when the user's finger is wet or has a greasy distribution, the fingerprint-based identification effect is greatly reduced.
  • the dielectric constant of the user's finger may change, resulting in a change in the dielectric spectrum of the finger, resulting in a change in the group delay characteristics of the finger. Affects identity.
  • the embodiment of the present invention introduces an environmental parameter monitoring module, which includes a humidity sensing unit and a temperature sensing unit.
  • the humidity sensing unit includes a humidity sensor and a grease sensor, respectively. It is used to collect information such as the moisture distribution and greasy degree of the user's finger.
  • the temperature sensing unit is a body temperature sensor for collecting the current body temperature information of the user, and the detection range of the body temperature sensor is 34 ° C - 39 ° C.
  • Step 300 After performing Fourier transform filtering on the fingerprint image, the gradient algorithm is used to analyze the sharpness of the fingerprint image, select the fingerprint image with the highest definition, and use the dynamic binarization algorithm to binarize the fingerprint image. deal with;
  • the processing manner of the Fourier transform filtering processing on the fingerprint image specifically includes:
  • Step 310 Using the fingerprint images collected by the fingerprint sensor to establish a fingerprint image processing database:
  • N 20, 1 ⁇ n ⁇ N, and c n represents the nth fingerprint image.
  • Step 311 Divide the fingerprint image c 1 into sub-blocks of 16 ⁇ 16 size, generate a matrix A+Bj for each sub-block, and perform Fourier transform on the matrix:
  • Step 312 When the frequency band of X jk +Y jk j is greater than ten times the center frequency band, set it to 0; similarly, when the frequency band of X jk +Y jk j is less than one tenth of the center frequency band, set it Is 0; then it is divided into linear transformations:
  • Step 313 Perform an inverse Fourier transform to inversely convert the enhanced frequency domain information into spatial domain information, and implement Fourier transform filtering on the fingerprint image:
  • Step 314 Repeat the above steps on the other fingerprint images c 2 , . . . , c n , . . . , c N of the fingerprint image processing database, and perform Fourier transform filtering processing on each fingerprint image.
  • the analysis method for analyzing the sharpness of the fingerprint image specifically includes:
  • Step 320 The gradient of the fingerprint image c 1 is acquired by a gradient algorithm, and the gradient of the fingerprint image c 1 can be expressed as:
  • f(x, y) is a pixel point located at the (x, y) position.
  • Step 321 When the size of the fingerprint image c 1 is M ⁇ N, the sharpness of the fingerprint image c 1 can be expressed as:
  • Step 322 Repeat the above steps on the other fingerprint images c 2 , . . . , c n , . . . , c N of the fingerprint image processing database, and analyze the sharpness of each fingerprint image;
  • Step 323 Sort the fingerprint image sharpness in the fingerprint image processing database, and select the fingerprint image with the highest definition.
  • the processing method for performing binarization processing on the fingerprint image specifically includes:
  • Step 330 Determine the window size of the dynamic binarization of the fingerprint image, and the window size used in the embodiment of the present invention is 8 ⁇ 8;
  • Step 331 Determine a threshold T(x, y) of the dynamic binarization algorithm:
  • f(x, y) is the gray value of the pixel at the (x, y) position
  • T t is a fixed reference threshold, the size of which is determined according to the gray scale map of the image
  • ⁇ is the misjudgment correction factor
  • Step 332 Let the binarized fingerprint image be I(x, y), then I(x, y) can be expressed as:
  • Step 333 Perform binarization operation on the fingerprint image according to formula (8), collect the binarized fingerprint image, and save it.
  • Step 400 Convert the group delay curve into a corresponding group delay image, and filter the group delay image by using the unscented particle filtering algorithm;
  • step 400 the conversion mode of converting the group delay curve into the corresponding group delay image is specifically: creating an M ⁇ N blank matrix; scanning the group delay curve from top to bottom according to the linear mapping method, and 1 MHz- The 200 MHz group delay curve is converted into a group delay image of size M ⁇ N.
  • the processing method of filtering the group delay image specifically includes:
  • Step 411 Update the state of the image pixels ⁇ x i k-1 , p i k-1 ⁇ according to the density function, thereby obtaining a new pixel set.
  • Step 412 Calculating a pixel set Mean And variance
  • Step 413 Obtain a density function using the mean and variance described above Sampling new pixels from
  • Step 414 Calculate the weight of each pixel and normalize according to the latest predicted result:
  • Step 415 By continuously updating the position of the pixel, the weight of each pixel is recalculated after each iteration, until all the iterations are completed, and finally the filtering process of the signal is completed.
  • Step 500 After using the log-Gabor filter to segment the fingerprint image, extract the fingerprint grain direction feature, and extract the point feature and the line feature of the fingerprint according to the grain direction;
  • the segmentation manner of dividing the fingerprint image specifically includes: dividing the fingerprint image into a plurality of non-overlapping small blocks by using a log-Gabor filter according to characteristics such as local directionality of the fingerprint and statistical features of the pattern. Calculate the feature vector of each small block, and judge whether a small block meets the feature extraction requirement according to the feature vector. If it is not suitable, discard the small block.
  • the extraction method of extracting the fingerprint direction feature is as follows: in a small block, taking the pixel (x, y) as the center, calculating the grain direction O(i, j) of the square region with W as the side length, and respectively The direction of the grain of each pixel is analyzed:
  • V y 2G x (x,y)-G y (x,y) (11)
  • G x (x, y), G y (x, y) are gradients at the pixel (x, y), respectively.
  • the point feature and the line feature extraction method for extracting the fingerprint according to the direction of the grain are specifically: obtaining the point feature and the line feature of the fingerprint by tracking the grain direction, including the coordinates, direction, type, length, maximum curvature, starting point of the start point and the end point. End point sequence and other information; by tracking the fingerprint nodes between the lines Adjacent topological relationship establishes the association order relationship between nodes through the topological relationship of the lines.
  • Step 600 extract group delay eigenvalues of group delay curvature radii in different frequency bands and group delay average values in different frequency bands, and extract spectral characteristics of the group delay curve;
  • step 600 the acquisition mode of the group delay radius of curvature is:
  • the first-order partial guide of y Is the second-order partial derivative of x with respect to y.
  • the collection method of the average value of the acquisition group delay includes:
  • Step 622 Calculate the weight of the t n segment group delay average value in the group delay curve And use this weight as one of the characteristic values of the group delay.
  • the spectrum characteristic of the extracted group delay curve is obtained by performing Fourier transform on the group delay curve according to the principle of Fourier transform, and obtaining the distribution of the group delay curve in the frequency domain, and the collection formula is:
  • Step 700 Collect multiple fingerprint images, finger group delay curves, and environmental parameter information as training sample sets, perform iterative calculation on the training sample set by using a convolutional neural network algorithm, obtain an identity recognition model, and store the identity recognition model in the biological In the feature database;
  • the convolutional neural network algorithm structure includes an input layer, a convolutional layer, a sub-sampling layer, a fully connected layer, and an output layer.
  • the layers are set as follows:
  • the number of input layer nodes is set to 16 ⁇ 16, a total of 256 nodes:
  • a 01n represents the direction of the grain
  • a 02n represents the length of the line
  • a 03n represents the point feature in the texture
  • a 04n represents the line feature in the texture
  • a 05n represents the degree of association between the texture nodes
  • a 06n Representing the adjacent topological relationship between the lines
  • a 07n represents the group delay curvature radius
  • a 08n represents the group delay curvature direction
  • a 09n represents the group delay average
  • a 10n represents the group delay of different frequencies in the time domain.
  • a 11n represents the spectral characteristics of the group delay
  • a 12n represents the weight of the group delay of different frequencies in the frequency domain
  • a 13n represents the humidity of the external environment
  • a 14n represents the moisture content of the surface of the user's finger
  • a 15n represents the surface of the finger of the user.
  • a 16n represents the body temperature.
  • the values of ⁇ a 01n , a 02n , a 03n ,..., a 12n ⁇ are derived from the calculation results of the biometric extraction module
  • the values of ⁇ a 13n , a 14n , a 15n , a 16n ⁇ are derived from Monitoring results of the environmental parameter monitoring module.
  • the number of nodes in the output layer is set to 2, which respectively represent two recognition results: (1) biometric matching, the user is a legitimate user; (2) the biometrics do not match, and the user is an illegal user.
  • each cell in the layer receives a set of cells in a small neighborhood in the previous layer as input, multiplies a trainable convolution kernel, adds an offset, and then outputs through the activation function.
  • the convolutional layer is calculated as follows:
  • w is a convolution kernel of size J ⁇ I
  • a function f is an activation function
  • b is an offset amount.
  • the size of the convolution kernel is 5 ⁇ 5
  • the activation function always uses the tanh function
  • the offset is determined based on the empirical value.
  • the purpose of the sub-sampling layer is to perform sampling operations on each feature map obtained in the previous layer, so that the size of the feature map is reduced, and the network can be made to be invariant to the translation and scaling of the object, so that the network is more robust.
  • sample as follows:
  • x is a two-dimensional input quantity
  • y is an output obtained after sampling
  • S 1 ⁇ S 2 is a size of a sampling template.
  • the size of the sampling template is set to 2 ⁇ 2.
  • Step 800 Extract the extracted fingerprint feature value, group delay feature value, and environmental parameter information with the biological
  • the identity recognition model stored in the feature database is cross-validated, and the degree of correlation between the extracted fingerprint feature value, the group delay feature value, and the environmental parameter information and the fingerprint image and the finger group delay curve in the identity recognition model is determined, and the two types are determined. Whether the biometric information comes from the same user, thereby realizing the user's identification.
  • the identity recognition mode of the embodiment of the present invention includes a training process of identity recognition and a test process of identity recognition; the training process of the identity recognition specifically includes: firstly collecting fingerprint images and fingers of multiple volunteers in different time periods. Group delay curve. At the same time, the humidity parameter, the grease sensor and the body temperature sensor are used to collect the environmental parameter information such as the surface moisture, greasy degree and body temperature data of the volunteer finger, and the collected fingerprint image, the finger group delay curve and the environmental parameter information are used as the training sample set. By preprocessing the training sample set, the training sample set is divided into five small sample sets of fingerprint, group delay, moisture humidity, greasy degree and body temperature, and the sample set is distorted.
  • the training sample set is iteratively calculated by using the convolutional neural network algorithm.
  • the iterative operation is stopped, and an identification model related to the environmental parameter information is output, and the identification is recognized.
  • the model is stored in the biometric database to complete the training process of identification.
  • the testing process of the identification specifically includes: cross-validating the extracted fingerprint feature value, the group delay feature value, and the collected environmental parameter information with the identity recognition model in the biometric database, and determining the extracted fingerprint feature value and the group delay feature value. And the degree of correlation between the environmental parameter information and the fingerprint image, the finger group delay curve and the environmental parameter information in the identification model, thereby realizing the user's identification.
  • the identity recognition apparatus and method of the embodiment of the present invention collects the fingerprint feature of the user and the characteristic information of the finger dielectric spectrum for identification, and at the same time, in order to reduce the interference of the external environment and the internal environment on the identity recognition, while collecting the fingerprint feature Collecting environmental parameter information such as water distribution, greasy degree and body temperature of the user's finger, and correcting the identification by environmental parameter information, effectively improving the identity Other accuracy and reliability.
  • the multi-modal fusion identification algorithm based on convolutional neural network is used for identification, which further improves the accuracy and reliability of identity recognition.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Collating Specific Patterns (AREA)

Abstract

本发明涉及生物识别技术领域，特别涉及一种身份识别装置及方法。所述身份识别装置包括生物特征信号采集模块、生物特征提取模块和生物特征识别模块；所述生物特征信号采集模块用于采集指纹图像以及指部群延迟曲线；所述生物特征提取模块用于根据所述指纹图像以及指部群延迟曲线分别提取指纹特征值和群延迟特征值；所述生物特征识别模块用于通过训练样本集构建身份识别模型，并利用卷积神经网络算法将所提取的指纹特征值和群延迟特征值与所述身份识别模型进行交叉验证，实现用户的身份识别。本发明通过采集用户的指纹特征和指部介电谱特征信息进行身份识别，并采用基于卷积神经网络的多模态融合身份识别算法进行身份识别，提高身份识别的准确性。

Description

一种身份识别装置及方法 技术领域
本发明涉及生物识别技术领域，特别涉及一种身份识别装置及方法。
背景技术
随着科技的迅猛发展，整个社会的信息化程度日益提高，信息泄密的概率与日递增。这些信息一旦被不法分子采集并冒用，将会给被冒用者带来重大的经济利益损失，严重时甚至会危害到他人及整个社会的安全。信息保护的一个重要手段是采用身份识别技术。传统的身份识别技术普遍是基于IC射频卡、静态密码或动态密码等，存在易丢失、易遗忘、易替代等安全漏洞问题。
生物识别技术是指通过计算机，利用人体固有的生物特征来进行身份认证的一种识别技术，目前已被用于生物识别的人体生物特征包括指纹、掌纹、脸形、虹膜、脉搏、声音、脑电、心电、基因等，该技术具有不易遗忘、不易伪造或被盗、随身“携带”和随时随地可用等优点，比传统的身份识别方法更具安全、保密和便捷性。
然而，目前大部分的生物识别技术都是基于单一的生物特征，如仅仅采用指纹、虹膜、声音、脸部等，但由于每一种生物特征都有或多或少的缺陷，尤其是易受外界环境以及生物特征受损等因数的影响，导致识别率下降。如指纹易磨损，沾水后难以识别；虹膜识别结果与光照，角度等都密切相关；人脸识别不能对脸部饰物和整形手术后的脸部进行精确识别等。针对上述问题，有研究人员提出了基于多模态的生物特征识别方法。
例如，中国专利CN105117697A提出了一种指纹识别方法、指纹识别装置及其终端设备，该专利通过向待检测对象发射红光、红外光和绿光，通过反射 光的变化来判断待检测对象是否为活体，确定为活体后再通过采集指纹图像对用户的身份进行识别。因此，严格来说，该专利还是采用单一的生物特征进行身份识别。
中国专利CN102542263A提出了一种基于指部生物特征的多模态身份认证装置及方法。该专利主要是通过采集指静脉图像、指节纹图像和指纹图像来采集用户的生物特征信息。然而，在该专利中，只是孤立地采集这三种生物特征信息并进行身份识别，缺乏将这三种模态进行有效的融合，识别率还需进一步提高。
中国专利CN1758263提出了一种基于得分差加权融合的多模态身份识别方法。该专利提出了采用得分差加权融合算法实现多模态的身份识别，然而，该算法并没有考虑其他环境因素对身份识别的影响。
发明内容
本发明提供了一种身份识别装置及方法，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本发明提供了如下技术方案：
一种身份识别装置，包括生物特征信号采集模块、生物特征提取模块和生物特征识别模块；所述生物特征信号采集模块用于采集指纹图像以及指部群延迟曲线；所述生物特征提取模块用于根据所述指纹图像以及指部群延迟曲线分别提取指纹特征值和群延迟特征值；所述生物特征识别模块用于通过训练样本集构建身份识别模型，并利用卷积神经网络算法将所提取的指纹特征值和群延迟特征值与所述身份识别模型进行交叉验证，实现用户的身份识别。
本发明实施例采取的技术方案还包括：所述生物特征信号采集模块包括指纹采集单元和介电谱采集单元，所述指纹采集单元为指纹传感器，用于采集指部的指纹图像；所述介电谱采集单元包括信号发送电极、信号接收电极、信号 源、接收器，所述信号发送电极和信号接收电极分别位于指纹传感器的两端；当手指按压到信号发送电极和信号接收电极时，所述信号源产生正弦波，并通过信号发送电极将正弦波耦合到用户的指部，所述信号接收电极接收正弦波经过指部后的信号群延迟曲线，并存储在接收器中。
本发明实施例采取的技术方案还包括信号预处理模块，所述信号预处理模块包括：
指纹预处理单元：用于对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，并采用动态二值化算法对指纹图像进行二值化处理；
群延迟曲线预处理单元：用于将所述群延迟曲线转换为群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理。
本发明实施例采取的技术方案还包括环境参数监测模块，所述环境参数监测模块用于采集用户环境参数信息；所述环境参数监测模块包括湿度传感单元和温度传感单元，所述湿度传感单元包括湿度传感器和油脂传感器，所述湿度传感器和油脂传感器分别用于采集用户手指的水分分布和油腻程度信息；所述温度传感单元为体温传感器，用于采集用户当前的体温信息。
本发明实施例采取的技术方案还包括：所述生物特征提取模块包括：
指纹特征提取单元：用于利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，并根据纹路方向提取指纹的点特征和线特征；
群延迟特征提取单元：用于提取不同频带下的群延迟曲率半径、不同频带下的群延迟平均值，并提取群延迟曲线的频谱特性。
本发明实施例采取的技术方案还包括：所述生物特征识别模块包括训练单元和测试单元；
所述训练单元的训练方式包括：采集多个志愿者在不同时间段的指纹图 像、指部群延迟曲线和环境参数信息，将采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样本集，利用卷积神经网络算法对训练样本集进行迭代计算，并输出与所述环境参数信息相关的身份识别模型；将该身份识别模型存储在生物特征数据库中。
所述测试单元的测试方式包括：将所述生物特征提取模块提取的指纹特征值、群延迟特征值以及环境参数监测模块采集的环境参数信息与生物特征数据库中的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线以及环境参数信息的相关程度，并输出身份识别结果，实现用户的身份识别。
本发明实施例采取的另一技术方案为：一种身份识别方法，包括以下步骤：
步骤a：采集指纹图像以及指部群延迟曲线；
步骤b：根据指纹图像以及指部群延迟曲线分别提取指纹特征值和群延迟特征值；
步骤c：通过训练样本集构建身份识别模型，并利用卷积神经网络算法将所提取的指纹特征值和群延迟特征值与所述身份识别模型进行交叉验证，实现用户的身份识别。
本发明实施例采取的技术方案还包括：所述步骤a还包括：
步骤a1：采集环境参数信息；所述环境参数信息包括水分湿度、油腻程度、体温信息；
步骤a2：对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，采用动态二值化算法对指纹图像进行二值化处理；
步骤a3：将群延迟曲线转换为群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理。
本发明实施例采取的技术方案还包括：在所述步骤b中，所述提取生物特 征信息具体包括：利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，根据纹路方向提取指纹的点特征和线特征；提取不同频带下的群延迟曲率半径、群延迟平均值，并提取群延迟曲线的频谱特性。
本发明实施例采取的技术方案还包括：在所述步骤c中，所述的身份识别包括：
步骤c1：采集多个志愿者在不同时间段的指纹图像、指部群延迟曲线和环境参数信息，将采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样本集，利用卷积神经网络算法对训练样本集进行迭代计算，并输出与所述环境参数信息相关的身份识别模型，将该身份识别模型存储在生物特征数据库中；
步骤c2：将提取的指纹特征值、群延迟特征值以及采集的环境参数信息与生物特征数据库中的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线以及环境参数信息的相关程度，并输出身份识别结果，实现用户的身份识别。
相对于现有技术，本发明实施例产生的有益效果在于：本发明实施例的身份识别装置及方法通过采集用户的指纹特征和指部介电谱特征信息进行身份识别，同时，为了减小外部环境和内部环境对身份识别的干扰，在采集指纹特征的同时采集用户手指的水分分布、油腻程度及体温等环境参数信息，通过环境参数信息对身份识别进行修正，有效地提高了身份识别的准确性和可靠性。并采用基于卷积神经网络的多模态融合身份识别算法进行身份识别，进一步提高身份识别的准确性和可靠性。
附图说明
图1是本发明实施例的身份识别装置的结构示意图；
图2是本发明实施例的生物特征信号采集模块的结构示意图；
图3是本发明实施例的基于卷积神经网络算法的身份识别示意图；
图4是本发明实施例的生物特征识别模块身份识别过程示意图；
图5是本发明实施例的身份识别方法的流程图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。
请参阅图1，是本发明实施例的身份识别装置的结构示意图。本发明实施例的身份识别装置包括生物特征信号采集模块、环境参数监测模块、信号预处理模块、生物特征提取模块、生物特征识别模块和生物特征数据库；生物特征信号采集模块用于采集指部的指纹图像以及正弦波经过指部后的信号群延迟曲线；环境参数监测模块用于采集用户指部的水分湿度、油腻程度、体温等环境参数信息；信号预处理模块用于对指纹图像和指部群延迟曲线进行预处理，得到二值化指纹图像及指部群延迟图像；生物特征提取模块用于根据二值化指纹图像以及指部群延迟图像分别提取指纹特征值和群延迟特征值；生物特征识别模块用于构建身份识别模型，将身份识别模型存储在生物特征数据库中；并通过卷积神经网络算法将生物特征提取模块提取的指纹特征值、群延迟特征值以及环境参数监测模块采集的环境参数信息进行融合后，与生物特征数据库中存储的身份识别模型进行交叉验证，判断这两种生物特征是否来自于同一个用户，从而实现用户的身份识别。
具体地，请一并参阅图2，是本发明实施例的生物特征信号采集模块的结构示意图；生物特征信号采集模块包括指纹采集单元和介电谱采集单元；
指纹采集单元用于采集指部的指纹图像；其中指纹采集单元F为一个能够连续快速采集的指纹传感器，当用户的手指按压指纹传感器时，指纹传感器则以4帧/秒的速度开始连续采集用户的指纹，当采集到20帧指纹图像后，自动结束指纹图像的采集；可以理解，在本发明其他实施例中，指纹图像的采集速度和数量还可以根据实际应用进行设定。
介电谱采集单元用于采集正弦波经过指部后的信号群延迟曲线；具体地，介电谱采集单元包括信号发送电极A、信号接收电极B、信号源(图未示)、接收器(图未示)，信号发送电极A和信号接收电极B分别位于指纹传感器的两端。当手指按压到信号发送电极A和信号接收电极B时，信号源以1MHz为频率间隔，在5秒内依序产生1MHz-200MHz的正弦波，并通过信号发送电极A将正弦波耦合到用户的指部。由于人体的差异性，每个人的手指的介电常数也不一样，导致正弦波在经过指部时信号的群延迟会发生改变。信号接收电极B通过接收不同频率下的指部信号群延迟，并存储在接收器中，从而采集用户指部的介电谱信息。可以理解，在本发明其他实施例中，介电谱采集单元产生的信号频率还可以根据实际应用进行设定。
通过指纹采集单元采集获得的指纹图像以及通过介电谱采集单元获得的指部群延迟曲线通常都伴有各种各样的噪声。本发明通过信号预处理模块对指纹图像和群延迟曲线进行预处理，使得低质量的指纹图像和群延迟曲线变得更加清晰、饱满，削弱或消除各种噪声干扰因素的影响，以便能提取或识别到精准的特征信息。
具体地，信号预处理模块包括指纹预处理单元和群延迟曲线预处理单元；
指纹预处理单元用于对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，选择出清晰度最高的指纹图像，并采用动态二值化算法对指纹图像进行二值化处理。
其中，指纹预处理单元用于对指纹图像进行傅里叶变换滤波处理的处理方式具体包括：
1.1：利用指纹传感器所采集的20幅指纹图像，建立指纹图像处理数据库：
C＝{c1,c2,...,cn,...,cN}   (1)
在公式(1)中，N＝20，1≤n≤N，cn表示第n幅指纹图像。
1.2：将指纹图像c1分成16×16大小的子块，对每一个子块生成矩阵A+Bj，并对矩阵进行傅里叶变换：
Figure PCTCN2016103024-appb-000001
1.3：当Xjk+Yjkj的频带大于十倍的中心频带时，将其设置为0；同样，当Xjk+Yjkj的频带小于中心频带的十分之一时，将其设置为0；然后对其进行分线性变换：
Cjk+Djkj＝(Xjk 2+Yjk 2)pow(Xjk+Yjkj)  (3)
1.4：进行逆傅里叶变换，将增强后的频率域信息反转换为空间域信息，实现对指纹图像的傅里叶变换滤波：
Figure PCTCN2016103024-appb-000002
1.5：对指纹图像处理数据库的其他指纹图像c2,...,cn,...,cN，重复执行上述步骤，对每一幅指纹图像进行傅里叶变换滤波处理。
指纹预处理单元利用梯度算法对指纹图像的清晰度进行分析的分析方式 具体包括：
2.1：通过梯度算法采集指纹图像c1的梯度，指纹图像c1的梯度可表示为：
Figure PCTCN2016103024-appb-000003
在公式(5)中，f(x,y)是位于(x,y)位置处的像素点。
2.2：当指纹图像c1的大小为M×N时，指纹图像c1的清晰度可表示为：
Figure PCTCN2016103024-appb-000004
2.3：对指纹图像处理数据库的其他指纹图像c2,...,cn,...,cN，重复执行上述步骤，对每一幅指纹图像的清晰度进行分析；
2.4：对指纹图像处理数据库中的20幅指纹图像清晰度进行排序，选择出清晰度最高的指纹图像。
指纹预处理单元采用动态二值化算法对指纹图像进行二值化处理的处理方式具体包括：
3.1：确定指纹图像动态二值化的窗口大小，本发明实施例中所采用的窗口大小为8×8；
3.2：确定动态二值化算法的阈值T(x,y)：
Figure PCTCN2016103024-appb-000005
在公式(7)中，
Figure PCTCN2016103024-appb-000006
f(x,y)是位于(x,y)位置处像素点的灰度值，Tt为固定的参考阈值，其大小根据图像的灰度分布图确定，ε为误判修正因子。
3.3：令二值化后的指纹图像为I(x,y)，则I(x,y)可以表示为：
Figure PCTCN2016103024-appb-000007
3.4：根据公式(8)对指纹图像进行二值化运算，采集经过二值化后的指纹图像，并保存。
群延迟曲线预处理单元用于将群延迟曲线转换为相应的群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理；
其中，群延迟曲线预处理单元将群延迟曲线转换为相应的群延迟图像的转换方式具体为：建立一个M×N的空白矩阵；按照线性映射的方法对群延迟曲线从上至下进行扫描，将1MHz-200MHz的群延迟曲线转换为大小为M×N的群延迟图像。
群延迟曲线预处理单元对群延迟图像进行滤波处理的处理方式具体为：
4.1：从群延迟图像中采集N个像素{xi 0,i＝1,2,...,N}，并令每个样本的初始权值为wi 0＝1/N,i＝1,2,...,N；
4.2：根据密度函数，更新图像像素{xi k-1,pi k-1}的状态，由此得到新的像素集
Figure PCTCN2016103024-appb-000008
4.3：计算像素集
Figure PCTCN2016103024-appb-000009
的均值
Figure PCTCN2016103024-appb-000010
和方差
Figure PCTCN2016103024-appb-000011
4.4：利用上述的均值和方差得到密度函数
Figure PCTCN2016103024-appb-000012
从中抽样得到新的像素
Figure PCTCN2016103024-appb-000013
4.5：根据最新预测所得的结果，计算各个像素的权值并进行归一化：
Figure PCTCN2016103024-appb-000014
4.6：通过不断更新像素的位置，每一次迭代结束后重新计算每个像素的权值，直至所有的迭代完成，最终完成信号的滤波处理。
生物特征提取模块用于根据二值化指纹图像和群延迟图像提取用户的指纹特征值以及群延迟特征值；具体地，生物特征提取模块包括指纹特征提取单元和群延迟特征提取单元；
指纹特征提取单元用于利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，并根据纹路方向提取指纹的点特征和线特征；
其中，指纹特征提取单元对指纹图像进行分割的分割方式具体包括：根据指纹的局部方向性、方向图的统计特征等特性，利用log-Gabor滤波器，将指纹图像分割成许多互不重叠的小块；计算每一小块的特征向量，根据特征向量来判断某一小块是否符合特征提取要求，如不合适，则舍弃该一小块。
指纹特征提取单元提取指纹纹路方向特征的提取方式具体为：在某一小块内，以像素(x,y)为中心，计算以W为边长的方形区域的纹路方向O(i,j)，并分别对每一像素的纹路方向进行分析：
Figure PCTCN2016103024-appb-000015
Vy＝2Gx(x,y)-Gy(x,y)   (11)
Figure PCTCN2016103024-appb-000016
在上述公式中，Gx(x,y)、Gy(x,y)分别为像素(x,y)处的梯度。
指纹特征提取单元根据纹路方向提取指纹的点特征和线特征的提取方式具体为：通过对纹路方向的跟踪获得指纹的点特征和线特征，包括起点和终点的坐标、方向、类型，长度、最大曲率、起点与终点顺序等信息；通过跟踪指纹节点建立纹线之间的相邻拓扑关系，通过纹线的拓扑关系建立节点之间的关联排序关系。
群延迟特征提取单元用于提取不同频带下的群延迟曲率半径以及不同频带下的群延迟平均值等群延迟特征值，并提取群延迟曲线的频谱特性。
其中，群延迟特征提取单元采集群延迟曲率半径的采集方式为：
5.1：根据群延迟曲线变化规律，将群延迟曲线划分为多段曲线G＝{g1,g2,...,gn,...,gN}，其中gn表示为第n段曲线；
5.2：采用多项式y＝f(x)＝Ax2+Bx+C对G＝{g1,g2,...,gn,...,gN}中每一段曲线进行多项式拟合，确定参数A，B，C的大小；
5.3：根据曲率计算公式
Figure PCTCN2016103024-appb-000017
计算G＝{g1,g2,...,gn,...,gN}中每一段曲线的曲率半径大小，并将曲线的曲率半径大小作为群延迟的特征值之一，其中
Figure PCTCN2016103024-appb-000018
为x关于y的一阶偏导，
Figure PCTCN2016103024-appb-000019
为x关于y的二阶偏导。
群延迟特征提取单元采集群延迟平均值的采集方式包括：
6.1：将1MHz-200MHz的群延迟曲线按照频率的大小，以5MHz为间隔，将群延迟曲线均匀划分为40段，T＝{t1,t2,...,tn,...,t40}，即其中t1表示1MHz-5MHz的群延迟，t2表示6MHz-10MHz的群延迟，依此类推；
6.2：计算T＝{t1,t2,...,tn,...,t40}每一段tn群延迟的平均值，计算方式为
Figure PCTCN2016103024-appb-000020
6.3：计算tn段群延迟平均值在群延迟曲线中的权重
Figure PCTCN2016103024-appb-000021
并将该权重作为群延迟的特征值之一。
群延迟特征提取单元提取群延迟曲线的频谱特性的采集方式为：根据傅里叶变换原理，对群延迟曲线进行傅里叶变换，获得群延迟曲线在频域的分布情 况，采集公式为：
Figure PCTCN2016103024-appb-000022
并计算每一个频带的权重
Figure PCTCN2016103024-appb-000023
将该权重作为群延迟的特征值之一。
环境参数监测模块用于采集用户手指的水分湿度、油腻程度、用户体温等环境参数信息，所采集的环境参数信息用于对生物特征识别模块进行修正，提高身份识别的可靠性。研究表明，当用户的手指沾水或者有油腻分布时，基于指纹的身份识别效果大幅降低。此外，由于用户体温的变化，可能会引起用户指部介电常数改变，从而导致指部的介电谱也发生改变，造成指部的群延迟特性发生变化，影响身份识别。为了减小外部因素对身份识别的干扰，本发明实施例通过引入环境参数监测模块，可有效提高身份识别的准确性。
具体地，环境参数监测模块包括湿度传感单元和温度传感单元，湿度传感单元包括一个湿度传感器和油脂传感器，分别用于采集用户手指的水分分布和油腻程度等信息。温度传感单元为一个体温传感器，用于采集用户当前的体温信息，该体温传感器的检测范围为34℃-39℃。
生物特征识别模块用于利用卷积神经网络算法对训练样本集(包括多幅指纹图像和指部群延迟曲线以及环境参数信息)进行迭代计算，得到身份识别模型，并将提取的生物特征信息与以及环境参数信息与身份识别模型进行交叉验证，判断所提取的生物特征信息与身份识别模型中的指纹图像、指部群延迟曲线的相关程度，判断这两种生物特征信息是否来自于同一个用户，从而实现用户的身份识别。
请参阅图3，是本发明实施例的基于卷积神经网络算法的身份识别示意图。卷积神经网络算法结构包括输入层，卷积层，子采样层，全连接层和输出层。各层设置如下：
1、输入层设置：
在基于卷积神经网络的多模态融合身份识别算法中，输入层节点数设置为16×16，共256个节点数：
Figure PCTCN2016103024-appb-000024
在公式(14)中，a01n代表纹路方向，a02n代表纹路长度，a03n代表纹路中的点特征，a04n代表纹路中的线特征，a05n代表纹路节点之间的关联度，a06n代表纹线之间的相邻拓扑关系，a07n代表群延迟曲率半径大小，a08n代表群延迟曲率变化方向，a09n代表群延迟的平均值，a10n代表不同频率的群延迟在时域内的权重，a11n代表群延迟的频谱特征，a12n代表不同频率的群延迟在频域内的权重，a13n代表外部环境的湿度，a14n代表用户手指表面的水分含量，a15n代表用户手指表面的油腻程度，a16n代表人体体温大小。此外，{a01n,a02n,a03n,...,a12n}的值均来自于生物特征提取模块的计算结果，{a13n,a14n,a15n,a16n}的值则来自于环境参数监测模块的监测结果。
2、输出层设置
输出层的节点数设置为2，分别代表两种识别结果：(1)生物特征匹配，用户为合法用户；(2)生物特征不匹配，用户为非法用户。
3、卷积层设置
在卷积层，该层中的每个单元接收前一层中的一个小邻域内的一组单元作为输入，乘以一个可训练的卷积核，然后加一个偏置，之后通过激活函数输出。
该卷积层按如下方式计算：
Figure PCTCN2016103024-appb-000025
在公式(15)中，w为尺寸J×I的卷积核，函数f为激活函数，b为偏置 量。在本次设置中，卷积核的大小为5×5，激活函数一律采用tanh函数，偏置量则根据经验值确定。
4、子采样层设置
子采样层的目的是对上一层得到的每一个特征图进行采样操作，使得特征图的尺寸减小，并且可以使得网络对物体平移、缩放有一定的不变性，使得网络更加鲁棒。在子采样层中，按如下方式进行采样：
Figure PCTCN2016103024-appb-000026
在公式(16)中，x为二维输入量，y为采样后得到的输出，S1×S2为采样模板的尺寸大小，在此算法中，该采样模板的大小设置为2×2。
请参阅图4，是本发明实施例的生物特征识别模块身份识别过程示意图。本发明实施例的生物特征识别模块包括训练单元和测试单元。
训练单元的训练过程具体包括：首先采集多个志愿者在不同时间段的指纹图像、指部群延迟曲线(本发明实施例包括200幅指纹图像、200幅指部群延迟曲线，具体采集数量并不仅限于此)。同时，利用湿度传感器、油脂传感器和体温传感器分别采集志愿者的手指表面水分、油腻程度和体温数据等环境参数信息(本发明实施例分别采集200次志愿者的手指表面水分、油腻程度和体温数据，具体采集次数并不仅限于此)。将所采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样本集，通过对训练样本集进行预处理，将训练样本集划分为指纹、群延迟、水分湿度、油腻程度、体温五个小样本集，并对样本集进行畸变处理。畸变处理完成后，利用卷积神经网络算法对训练样本集进行多次迭代计算，当迭代次数到达一定次数(本发明实施例仅以迭代100次为例)后，停止迭代运算，并输出一个与环境参数信息(即用户手指表面水分湿度、油腻程度、用户体温)相关的身份识别模型，将该身份识别模型存储在生物特征数据库中，完成身份识别的训练过程。
测试单元的测试过程具体包括：将生物特征提取模块提取的指纹特征值、 群延迟特征值以及环境参数监测模块采集的环境参数信息与身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与生物特征数据库中存储的身份识别模型中的指纹图像、指部群延迟曲线以及环境参数信息的相关程度，判断这两种生物特征是否来自于同一个用户，从而实现用户的身份识别。
请参阅图5，是本发明实施例的身份识别方法的流程图。本发明实施例的身份识别方法包括以下步骤：
步骤100：采集指部的指纹图像以及正弦波经过指部后的信号群延迟曲线；
在步骤100中，本发明实施例通过指纹传感器采集指部的指纹图像；当用户的手指按压指纹传感器时，指纹传感器则以4帧/秒的速度开始连续采集用户的指纹，当采集到20帧指纹图像后，自动结束指纹图像的采集。并通过介电谱采集单元采集正弦波经过指部后的信号群延迟曲线；介电谱采集单元包括信号发送电极A、信号接收电极B、信号源、接收器，信号发送电极A和信号接收电极B分别位于指纹传感器的两端，当手指按压到信号发送电极A和信号接收电极B时，信号源以1MHz为频率间隔，在5秒内依序产生1MHz-200MHz的正弦波，并通过信号发送电极A将正弦波耦合到用户的指部。由于人体的差异性，每个人的手指的介电常数也不一样，导致正弦波在经过指部时信号的群延迟会发生改变。信号接收电极B通过接收不同频率下的指部信号群延迟，并存储在接收器中，从而采集用户指部的介电谱信息。
步骤200：采集用户指部的水分湿度、油腻程度、体温等环境参数信息；
在步骤200中，当用户的手指沾水或者有油腻分布时，基于指纹的身份识别效果大幅降低。此外，由于用户体温的变化，可能会引起用户指部介电常数改变，从而导致指部的介电谱也发生改变，造成指部的群延迟特性发生变化， 影响身份识别。为了减小外部因素对身份识别的干扰，本发明实施例通过引入环境参数监测模块，环境参数监测模块包括湿度传感单元和温度传感单元，湿度传感单元包括一个湿度传感器和油脂传感器，分别用于采集用户手指的水分分布和油腻程度等信息。温度传感单元为一个体温传感器，用于采集用户当前的体温信息，该体温传感器的检测范围为34℃-39℃。
步骤300：对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，选择出清晰度最高的指纹图像，并采用动态二值化算法对指纹图像进行二值化处理；
在步骤300中，对指纹图像进行傅里叶变换滤波处理的处理方式具体包括：
步骤310：利用指纹传感器所采集的20幅指纹图像，建立指纹图像处理数据库：
C＝{c1,c2,...,cn,...,cN}    (1)
在公式(1)中，N＝20，1≤n≤N，cn表示第n幅指纹图像。
步骤311：将指纹图像c1分成16×16大小的子块，对每一个子块生成矩阵A+Bj，并对矩阵进行傅里叶变换：
Figure PCTCN2016103024-appb-000027
步骤312：当Xjk+Yjkj的频带大于十倍的中心频带时，将其设置为0；同样，当Xjk+Yjkj的频带小于中心频带的十分之一时，将其设置为0；然后对其进行分线性变换：
Cjk+Djkj＝(Xjk 2+Yjk 2)pow(Xjk+Yjkj)   (3)
步骤313：进行逆傅里叶变换，将增强后的频率域信息反转换为空间域信息，实现对指纹图像的傅里叶变换滤波：
Figure PCTCN2016103024-appb-000028
步骤314：对指纹图像处理数据库的其他指纹图像c2,...,cn,...,cN，重复执行上述步骤，对每一幅指纹图像进行傅里叶变换滤波处理。
对指纹图像的清晰度进行分析的分析方式具体包括：
步骤320：通过梯度算法采集指纹图像c1的梯度，指纹图像c1的梯度可表示为：
Figure PCTCN2016103024-appb-000029
在公式(5)中，f(x,y)是位于(x,y)位置处的像素点。
步骤321：当指纹图像c1的大小为M×N时，指纹图像c1的清晰度可表示为：
Figure PCTCN2016103024-appb-000030
步骤322：对指纹图像处理数据库的其他指纹图像c2,...,cn,...,cN，重复执行上述步骤，对每一幅指纹图像的清晰度进行分析；
步骤323：对指纹图像处理数据库中的20幅指纹图像清晰度进行排序，选择出清晰度最高的指纹图像。
对指纹图像进行二值化处理的处理方式具体包括：
步骤330：确定指纹图像动态二值化的窗口大小，本发明实施例中所采用的窗口大小为8×8；
步骤331：确定动态二值化算法的阈值T(x,y)：
Figure PCTCN2016103024-appb-000031
在公式(7)中，
Figure PCTCN2016103024-appb-000032
f(x,y)是位于(x,y)位置处像素点的灰度值，Tt为固定的参考阈值，其大小根据图像的灰度分布图确定，ε为误判修正因子。
步骤332：令二值化后的指纹图像为I(x,y)，则I(x,y)可以表示为：
Figure PCTCN2016103024-appb-000033
步骤333：根据公式(8)对指纹图像进行二值化运算，采集经过二值化后的指纹图像，并保存。
步骤400：将群延迟曲线转换为相应的群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理；
在步骤400中，将群延迟曲线转换为相应的群延迟图像的转换方式具体为：建立一个M×N的空白矩阵；按照线性映射的方法对群延迟曲线从上至下进行扫描，将1MHz-200MHz的群延迟曲线转换为大小为M×N的群延迟图像。
上述中，对群延迟图像进行滤波处理的处理方式具体包括：
步骤410：从群延迟图像中采集N个像素{xi 0,i＝1,2,...,N}，并令每个样本的初始权值为wi 0＝1/N,i＝1,2,...,N；
步骤411：根据密度函数，更新图像像素{xi k-1,pi k-1}的状态，由此得到新的像素集
Figure PCTCN2016103024-appb-000034
步骤412：计算像素集
Figure PCTCN2016103024-appb-000035
的均值
Figure PCTCN2016103024-appb-000036
和方差
Figure PCTCN2016103024-appb-000037
步骤413：利用上述的均值和方差得到密度函数
Figure PCTCN2016103024-appb-000038
从中抽样得到新的像素
Figure PCTCN2016103024-appb-000039
步骤414：根据最新预测所得的结果，计算各个像素的权值并进行归一化：
Figure PCTCN2016103024-appb-000040
步骤415：通过不断更新像素的位置，每一次迭代结束后重新计算每个像素的权值，直至所有的迭代完成，最终完成信号的滤波处理。
步骤500：利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，并根据纹路方向提取指纹的点特征和线特征；
在步骤500中，对指纹图像进行分割的分割方式具体包括：根据指纹的局部方向性、方向图的统计特征等特性，利用log-Gabor滤波器，将指纹图像分割成许多互不重叠的小块；计算每一小块的特征向量，根据特征向量来判断某一小块是否符合特征提取要求，如不合适，则舍弃该一小块。
提取指纹纹路方向特征的提取方式具体为：在某一小块内，以像素(x,y)为中心，计算以W为边长的方形区域的纹路方向O(i,j)，并分别对每一像素的纹路方向进行分析：
Figure PCTCN2016103024-appb-000041
Vy＝2Gx(x,y)-Gy(x,y)   (11)
Figure PCTCN2016103024-appb-000042
在上述公式中，Gx(x,y)、Gy(x,y)分别为像素(x,y)处的梯度。
根据纹路方向提取指纹的点特征和线特征的提取方式具体为：通过对纹路方向的跟踪获得指纹的点特征和线特征，包括起点和终点的坐标、方向、类型，长度、最大曲率、起点与终点顺序等信息；通过跟踪指纹节点建立纹线之间的 相邻拓扑关系，通过纹线的拓扑关系建立节点之间的关联排序关系。
步骤600：提取不同频带下的群延迟曲率半径以及不同频带下的群延迟平均值等群延迟特征值，并提取群延迟曲线的频谱特性；
在步骤600中，采集群延迟曲率半径的采集方式为：
步骤610：根据群延迟曲线变化规律，将群延迟曲线划分为多段曲线G＝{g1,g2,...,gn,...,gN}，其中gn表示为第n段曲线；
步骤611：采用多项式y＝f(x)＝Ax2+Bx+C对G＝{g1,g2,...,gn,...,gN}中每一段曲线进行多项式拟合，确定参数A，B，C的大小；
步骤612：根据曲率计算公式
Figure PCTCN2016103024-appb-000043
计算G＝{g1,g2,...,gn,...,gN}中每一段曲线的曲率半径大小，并将曲线的曲率半径大小作为群延迟的特征值之一，其中
Figure PCTCN2016103024-appb-000044
为x关于y的一阶偏导，
Figure PCTCN2016103024-appb-000045
为x关于y的二阶偏导。
上述中，采集群延迟平均值的采集方式包括：
步骤620：将1MHz-200MHz的群延迟曲线按照频率的大小，以5MHz为间隔，将群延迟曲线均匀划分为40段，T＝{t1,t2,...,tn,...,t40}，即其中t1表示1MHz-5MHz的群延迟，t2表示6MHz-10MHz的群延迟，依此类推；
步骤621：计算T＝{t1,t2,...,tn,...,t40}每一段tn群延迟的平均值，计算方式为
Figure PCTCN2016103024-appb-000046
步骤622：计算tn段群延迟平均值在群延迟曲线中的权重
Figure PCTCN2016103024-appb-000047
并将该权重作为群延迟的特征值之一。
上述中，提取群延迟曲线的频谱特性的采集方式为：根据傅里叶变换原理，对群延迟曲线进行傅里叶变换，获得群延迟曲线在频域的分布情况，采集公式为：
Figure PCTCN2016103024-appb-000048
并计算每一个频带的权重
Figure PCTCN2016103024-appb-000049
将该权重作为群延迟的特征值之一。
步骤700：采集多幅指纹图像、指部群延迟曲线以及环境参数信息作为训练样本集，利用卷积神经网络算法对训练样本集进行迭代计算，得到身份识别模型，并将身份识别模型存储在生物特征数据库中；
在步骤700中，卷积神经网络算法结构包括输入层，卷积层，子采样层，全连接层和输出层。各层设置如下：
1、输入层设置：
在基于卷积神经网络的多模态融合身份识别算法中，输入层节点数设置为16×16，共256个节点数：
Figure PCTCN2016103024-appb-000050
在公式(14)中，a01n代表纹路方向，a02n代表纹路长度，a03n代表纹路中的点特征，a04n代表纹路中的线特征，a05n代表纹路节点之间的关联度，a06n代表纹线之间的相邻拓扑关系，a07n代表群延迟曲率半径大小，a08n代表群延迟曲率变化方向，a09n代表群延迟的平均值，a10n代表不同频率的群延迟在时域内的权重，a11n代表群延迟的频谱特征，a12n代表不同频率的群延迟在频域内的权重， a13n代表外部环境的湿度，a14n代表用户手指表面的水分含量，a15n代表用户手指表面的油腻程度，a16n代表人体体温大小。此外，{a01n,a02n,a03n,...,a12n}的值均来自于生物特征提取模块的计算结果，{a13n,a14n,a15n,a16n}的值则来自于环境参数监测模块的监测结果。
2、输出层设置
输出层的节点数设置为2，分别代表两种识别结果：(1)生物特征匹配，用户为合法用户；(2)生物特征不匹配，用户为非法用户。
3、卷积层设置
在卷积层，该层中的每个单元接收前一层中的一个小邻域内的一组单元作为输入，乘以一个可训练的卷积核，然后加一个偏置，之后通过激活函数输出。该卷积层按如下方式计算：
Figure PCTCN2016103024-appb-000051
在公式(15)中，w为尺寸J×I的卷积核，函数f为激活函数，b为偏置量。在本次设置中，卷积核的大小为5×5，激活函数一律采用tanh函数，偏置量则根据经验值确定。
4、子采样层设置
子采样层的目的是对上一层得到的每一个特征图进行采样操作，使得特征图的尺寸减小，并且可以使得网络对物体平移、缩放有一定的不变性，使得网络更加鲁棒。在子采样层中，按如下方式进行采样：
Figure PCTCN2016103024-appb-000052
在公式(16)中，x为二维输入量，y为采样后得到的输出，S1×S2为采样模板的尺寸大小，在此算法中，该采样模板的大小设置为2×2。
步骤800：将提取的指纹特征值、群延迟特征值以及环境参数信息与生物 特征数据库中存储的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线的相关程度，判断这两种生物特征信息是否来自于同一个用户，从而实现用户的身份识别。
在步骤800中，本发明实施例的身份识别方式包括身份识别的训练过程和身份识别的测试过程；身份识别的训练过程具体包括：首先采集多个志愿者在不同时间段的指纹图像、指部群延迟曲线。同时，利用湿度传感器、油脂传感器和体温传感器分别采集志愿者的手指表面水分、油腻程度和体温数据等环境参数信息，将所采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样本集，通过对训练样本集进行预处理，将训练样本集划分为指纹、群延迟、水分湿度、油腻程度、体温五个小样本集，并对样本集进行畸变处理。畸变处理完成后，利用卷积神经网络算法对训练样本集进行多次迭代计算，当迭代次数到达一定次数后，停止迭代运算，并输出一个与环境参数信息相关的身份识别模型，将该身份识别模型存储在生物特征数据库中，完成身份识别的训练过程。
身份识别的测试过程具体包括：将提取的指纹特征值、群延迟特征值以及采集的环境参数信息与生物特征数据库中的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线以及环境参数信息的相关程度，从而实现用户的身份识别。
本发明实施例的身份识别装置及方法通过采集用户的指纹特征和指部介电谱特征信息进行身份识别，同时，为了减小外部环境和内部环境对身份识别的干扰，在采集指纹特征的同时采集用户手指的水分分布、油腻程度及体温等环境参数信息，通过环境参数信息对身份识别进行修正，有效地提高了身份识 别的准确性和可靠性。并采用基于卷积神经网络的多模态融合身份识别算法进行身份识别，进一步提高身份识别的准确性和可靠性。
虽然本发明参照当前的较佳实施方式进行了描述，但本领域的技术人员应能理解，上述较佳实施方式仅用来说明本发明，并非用来限定本发明的保护范围，任何在本发明的精神和原则范围之内，所做的任何修饰、等效替换、改进等，均应包含在本发明的权利保护范围之内。

Claims (10)

  1. 一种身份识别装置，其特征在于，包括生物特征信号采集模块、生物特征提取模块和生物特征识别模块；所述生物特征信号采集模块用于采集指纹图像以及指部群延迟曲线；所述生物特征提取模块用于根据所述指纹图像以及指部群延迟曲线分别提取指纹特征值和群延迟特征值；所述生物特征识别模块用于通过训练样本集构建身份识别模型，并利用卷积神经网络算法将所提取的指纹特征值和群延迟特征值与所述身份识别模型进行交叉验证，实现用户的身份识别。
  2. 根据权利要求1所述的身份识别装置，其特征在于，所述生物特征信号采集模块包括指纹采集单元和介电谱采集单元，所述指纹采集单元为指纹传感器，用于采集指部的指纹图像；所述介电谱采集单元包括信号发送电极、信号接收电极、信号源、接收器，所述信号发送电极和信号接收电极分别位于指纹传感器的两端；当手指按压到信号发送电极和信号接收电极时，所述信号源产生正弦波，并通过信号发送电极将正弦波耦合到用户的指部，所述信号接收电极接收正弦波经过指部后的信号群延迟曲线，并存储在接收器中。
  3. 根据权利要求1所述的身份识别装置，其特征在于，还包括信号预处理模块，所述信号预处理模块包括：
    指纹预处理单元：用于对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，并采用动态二值化算法对指纹图像进行二值化处理；
    群延迟曲线预处理单元：用于将所述群延迟曲线转换为群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理。
  4. 根据权利要求1所述的身份识别装置，其特征在于，还包括环境参数监测模块，所述环境参数监测模块用于采集用户环境参数信息；所述环境参数监测模块包括湿度传感单元和温度传感单元，所述湿度传感单元包括湿度传感器和油脂传感器，所述湿度传感器和油脂传感器分别用于采集用户手指的水分分布和油腻程度信息；所述温度传感单元为体温传感器，用于采集用户当前的体温信息。
  5. 根据权利要求1所述的身份识别装置，其特征在于，所述生物特征提取模块包括：
    指纹特征提取单元：用于利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，并根据纹路方向提取指纹的点特征和线特征；
    群延迟特征提取单元：用于提取不同频带下的群延迟曲率半径、不同频带下的群延迟平均值，并提取群延迟曲线的频谱特性。
  6. 根据权利要求1或4所述的身份识别装置，其特征在于，所述生物特征识别模块包括训练单元和测试单元；
    所述训练单元的训练方式包括：采集多个志愿者在不同时间段的指纹图像、指部群延迟曲线和环境参数信息，将采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样本集，利用卷积神经网络算法对训练样本集进行迭代计算，并输出与所述环境参数信息相关的身份识别模型；将该身份识别模型存储在生物特征数据库中；
    所述测试单元的测试方式包括：将所述生物特征提取模块提取的指纹特征值、群延迟特征值以及环境参数监测模块采集的环境参数信息与生物特征数据库中的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线以及环境参 数信息的相关程度，并输出身份识别结果，实现用户的身份识别。
  7. 一种身份识别方法，其特征在于，包括以下步骤：
    步骤a：采集指纹图像以及指部群延迟曲线；
    步骤b：根据指纹图像以及指部群延迟曲线分别提取指纹特征值和群延迟特征值；
    步骤c：通过训练样本集构建身份识别模型，并利用卷积神经网络算法将所提取的指纹特征值和群延迟特征值与所述身份识别模型进行交叉验证，实现用户的身份识别。
  8. 根据权利要求7所述的身份识别方法，其特征在于，所述步骤a还包括：
    步骤a1：采集环境参数信息；所述环境参数信息包括水分湿度、油腻程度、体温信息；
    步骤a2：对指纹图像进行傅里叶变换滤波处理后，利用梯度算法对指纹图像的清晰度进行分析，采用动态二值化算法对指纹图像进行二值化处理；
    步骤a3：将群延迟曲线转换为群延迟图像，并采用无迹粒子滤波算法对群延迟图像进行滤波处理。
  9. 根据权利要求8所述的身份识别方法，其特征在于，在所述步骤b中，所述提取生物特征信息具体包括：利用log-Gabor滤波器对指纹图像进行分割后，提取指纹纹路方向特征，根据纹路方向提取指纹的点特征和线特征；提取不同频带下的群延迟曲率半径、群延迟平均值，并提取群延迟曲线的频谱特性。
  10. 根据权利要求9所述的身份识别方法，其特征在于，在所述步骤c中，所述的身份识别包括：
    步骤c1：采集多个志愿者在不同时间段的指纹图像、指部群延迟曲线和环境参数信息，将采集的指纹图像、指部群延迟曲线和环境参数信息作为训练样 本集，利用卷积神经网络算法对训练样本集进行迭代计算，并输出与所述环境参数信息相关的身份识别模型，将该身份识别模型存储在生物特征数据库中；
    步骤c2：将提取的指纹特征值、群延迟特征值以及采集的环境参数信息与生物特征数据库中的身份识别模型进行交叉验证，判断所提取的指纹特征值、群延迟特征值以及环境参数信息与身份识别模型中的指纹图像、指部群延迟曲线以及环境参数信息的相关程度，并输出身份识别结果，实现用户的身份识别。
PCT/CN2016/103024 2016-08-04 2016-10-24 一种身份识别装置及方法 WO2018023884A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610632580.5A CN106250857A (zh) 2016-08-04 2016-08-04 一种身份识别装置及方法
CN201610632580.5 2016-08-04

Publications (1)

Publication Number Publication Date
WO2018023884A1 true WO2018023884A1 (zh) 2018-02-08

Family

ID=58077467

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/103024 WO2018023884A1 (zh) 2016-08-04 2016-10-24 一种身份识别装置及方法

Country Status (2)

Country Link
CN (1) CN106250857A (zh)
WO (1) WO2018023884A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110049441A (zh) * 2019-05-05 2019-07-23 山东科技大学 基于深度集成学习的WiFi室内定位方法
CN111324880A (zh) * 2020-03-09 2020-06-23 深圳连心电子科技有限公司 一种指纹和心电特征双认证身份识别系统及方法
CN111368598A (zh) * 2018-12-26 2020-07-03 上海银晨智能识别科技有限公司 人体生物特征采集系统
CN112687022A (zh) * 2020-12-18 2021-04-20 山东盛帆蓝海电气有限公司 一种基于视频的智能楼宇巡检方法及系统
CN113468988A (zh) * 2021-06-18 2021-10-01 南京润楠医疗电子研究院有限公司 一种基于ecg信号的多压力状态下身份识别方法
CN113612751A (zh) * 2021-07-28 2021-11-05 深圳供电局有限公司 一种配电网电力线载波通信系统接入安全性检测方法
CN115761820A (zh) * 2022-11-29 2023-03-07 河南职业技术学院 基于计算机生物特征采集识别系统及装置

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108228674B (zh) * 2016-12-22 2020-06-26 北京字节跳动网络技术有限公司 一种基于dkt的信息处理方法及装置
CN106599669B (zh) * 2016-12-31 2018-11-09 上海百芝龙网络科技有限公司 一种基于人体体表导电性的身份识别系统
CN106618570B (zh) * 2017-01-22 2019-08-20 深圳先进技术研究院 一种基于生物介电谱的皮肤生化指标检测方法及系统
CN107145839B (zh) * 2017-04-17 2020-05-05 努比亚技术有限公司 一种指纹图像补全模拟方法及其系统
CN109583387A (zh) * 2018-11-30 2019-04-05 龙马智芯（珠海横琴）科技有限公司 身份认证方法及装置
CN111310514A (zh) * 2018-12-11 2020-06-19 上海耕岩智能科技有限公司 编码遮罩生物特征重建方法及存储介质
CN110008676B (zh) * 2019-04-02 2022-09-16 合肥智查数据科技有限公司 一种人员多维身份查验与真实身份甄别系统及方法
CN110390290A (zh) * 2019-07-17 2019-10-29 北京天正聚合科技有限公司 一种识别目标对象的方法、装置、电子设备及存储介质
CN110415722B (zh) * 2019-07-25 2021-10-08 北京得意音通技术有限责任公司 语音信号处理方法、存储介质、计算机程序和电子设备
CN110516595B (zh) * 2019-08-27 2023-04-07 中国民航大学 基于卷积神经网络的手指多模态特征融合识别方法
CN111461022A (zh) * 2020-04-01 2020-07-28 北京心智计算科技有限公司 一种基于人工智能技术的皮纹识别系统
CN112559332B (zh) 2020-07-13 2023-04-07 支付宝(杭州)信息技术有限公司 一种生物特征识别的攻击测试方法、装置及设备
CN112073375B (zh) * 2020-08-07 2023-09-26 中国电力科学研究院有限公司 一种适用于电力物联网客户侧的隔离装置及隔离方法
CN114742647B (zh) * 2022-03-01 2023-09-29 陕西维一信息技术有限公司 一种基于区块链和大数据的金融交易系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1201541A (zh) * 1995-10-06 1998-12-09 安德拉斯·波德马尼克斯 识别手指活件特征的指纹识别设备中的检测器
CA2230386A1 (en) * 1998-02-24 1999-08-24 Identech Inc. Automated indentification and analysis of fingerprints
US20090016573A1 (en) * 2004-09-14 2009-01-15 Mcafee Ii Kenneth N Bioindex mechanism for increasing the relative speed of biometric indentification against large population samples
CN103605963A (zh) * 2013-03-01 2014-02-26 新乡学院 一种指纹识别方法
CN105184254A (zh) * 2015-09-02 2015-12-23 深圳先进技术研究院 一种身份认证方法及系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101773394B (zh) * 2010-01-06 2011-09-07 中国航天员科研训练中心 身份识别方法及应用该方法的身份识别系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1201541A (zh) * 1995-10-06 1998-12-09 安德拉斯·波德马尼克斯 识别手指活件特征的指纹识别设备中的检测器
CA2230386A1 (en) * 1998-02-24 1999-08-24 Identech Inc. Automated indentification and analysis of fingerprints
US20090016573A1 (en) * 2004-09-14 2009-01-15 Mcafee Ii Kenneth N Bioindex mechanism for increasing the relative speed of biometric indentification against large population samples
CN103605963A (zh) * 2013-03-01 2014-02-26 新乡学院 一种指纹识别方法
CN105184254A (zh) * 2015-09-02 2015-12-23 深圳先进技术研究院 一种身份认证方法及系统

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368598A (zh) * 2018-12-26 2020-07-03 上海银晨智能识别科技有限公司 人体生物特征采集系统
CN110049441A (zh) * 2019-05-05 2019-07-23 山东科技大学 基于深度集成学习的WiFi室内定位方法
CN110049441B (zh) * 2019-05-05 2020-09-22 山东科技大学 基于深度集成学习的WiFi室内定位方法
CN111324880A (zh) * 2020-03-09 2020-06-23 深圳连心电子科技有限公司 一种指纹和心电特征双认证身份识别系统及方法
CN112687022A (zh) * 2020-12-18 2021-04-20 山东盛帆蓝海电气有限公司 一种基于视频的智能楼宇巡检方法及系统
CN113468988A (zh) * 2021-06-18 2021-10-01 南京润楠医疗电子研究院有限公司 一种基于ecg信号的多压力状态下身份识别方法
CN113468988B (zh) * 2021-06-18 2024-04-05 南京润楠医疗电子研究院有限公司 一种基于ecg信号的多压力状态下身份识别方法
CN113612751A (zh) * 2021-07-28 2021-11-05 深圳供电局有限公司 一种配电网电力线载波通信系统接入安全性检测方法
CN113612751B (zh) * 2021-07-28 2023-06-13 深圳供电局有限公司 一种配电网电力线载波通信系统接入安全性检测方法
CN115761820A (zh) * 2022-11-29 2023-03-07 河南职业技术学院 基于计算机生物特征采集识别系统及装置
CN115761820B (zh) * 2022-11-29 2024-03-26 河南职业技术学院 基于计算机生物特征采集识别系统及装置

Also Published As

Publication number Publication date
CN106250857A (zh) 2016-12-21

Similar Documents

Publication Publication Date Title
WO2018023884A1 (zh) 一种身份识别装置及方法
US9633269B2 (en) Image-based liveness detection for ultrasonic fingerprints
Gottschlich Curved-region-based ridge frequency estimation and curved Gabor filters for fingerprint image enhancement
Han et al. Palm vein recognition using adaptive Gabor filter
Tom et al. Fingerprint based gender classification using 2D discrete wavelet transforms and principal component analysis
Li et al. Personal identification using knuckleprint
Zanganeh et al. Partial fingerprint matching through region-based similarity
Sagayam et al. Authentication of biometric system using fingerprint recognition with euclidean distance and neural network classifier
CN104809450B (zh) 基于在线极限学习机的腕部静脉认证系统
Cao et al. Automatic latent value determination
KR101601187B1 (ko) 손금 기반 사용자 인식 정보를 이용한 기기 컨트롤 장치 및 그 방법
Lamia et al. Biometric authentication based on multi-instance fingerprint fusion in degraded context
Xie et al. Fingerprint quality analysis and estimation for fingerprint matching
Yan et al. Non-intrusive fingerprints extraction from hyperspectral imagery
Ezzaki et al. Iris recognition algorithm based on Contourlet Transform and Entropy
Najafi et al. A new iris identification method based on ridgelet transform
Ismail et al. Enhancement of finger vein patterns extracted by maximum curvature method
Ain An efficient algorithm for fingerprint recognition using minutiae extraction
Chakraborty et al. Bio-metric identification using automated iris detection technique
Iwasokun et al. Singular-minutiae points relationship-based approach to fingerprint matching
Han et al. Iris recognition based on local mean decomposition
Mehrotra et al. Multi-algorithmic Iris authentication system
Kalaimathi et al. Extraction and authentication of biometric finger vein using gradient boosted feature algorithm
Halvi et al. Compression Technique based Iris Recognition via Statistical and HOG Measures on DWT
Ajij et al. PIPP: Person Identification from Palm-surface Polygons

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: 16911479

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 04.06.2019)

122 Ep: pct application non-entry in european phase

Ref document number: 16911479

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 16911479

Country of ref document: EP

Kind code of ref document: A1

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

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04/06/2019)