CN111310514A

CN111310514A - Method for reconstructing biological characteristics of coded mask and storage medium

Info

Publication number: CN111310514A
Application number: CN201811509942.7A
Authority: CN
Inventors: 张李亚迪
Original assignee: Shanghai Harvest Intelligence Tech Co Ltd
Current assignee: Shanghai Harvest Intelligence Tech Co Ltd
Priority date: 2018-12-11
Filing date: 2018-12-11
Publication date: 2020-06-19

Abstract

A coding mask biological characteristic reconstruction method and a storage medium are provided, wherein the method comprises the following steps of establishing a neural network, training the neural network through a training sample, inputting a biological characteristic photosensitive image filtered by a coding mask through the training sample, and outputting a biological characteristic original image corresponding to the biological characteristic photosensitive image to obtain the trained neural network. And acquiring the light-sensitive image of the biological characteristics filtered by the coding mask, inputting the light-sensitive image into the trained neural network, and acquiring a predicted value of the biological characteristic image output by the neural network. Compared with the prior art, the scheme of the invention is designed by the method, so that the beneficial effect of restoring the original image of the filtered photosensitive image by means of the neural network is achieved, the requirement on the anti-resolving power resource of hardware equipment is reduced, the universality of the fingerprint identification method is improved, and the safety of the fingerprint identification method is also improved.

Description

Method for reconstructing biological characteristics of coded mask and storage medium

Technical Field

The invention relates to the technical field of biological feature identification, in particular to a reconstruction method of biological feature optical imaging with a coding mask.

Background

With the development of information technology, biometric identification technology plays an increasingly important role in ensuring information security, and fingerprint identification has become one of the key technical means for identity identification and equipment unlocking widely applied in the field of mobile internet. Under the trend that the screen of equipment accounts for more and more, traditional capacitive fingerprint identification can not meet the requirements, and ultrasonic fingerprint identification has the problems in the aspects of technical maturity, cost and the like, and optical fingerprint identification is a mainstream technical scheme expected to become the screen fingerprint identification.

The existing optical fingerprint identification scheme is that light rays passing through an OLED are directly received by an optical sensor after being totally reflected by a screen, and received pictures are analyzed and processed, so that the problem of safety can be brought firstly, and the sensed whole fingerprint image or part of the sensed fingerprint image directly carries information of fingerprints, so that the fingerprint identification scheme is easy to steal by people with different values. Secondly, the imaging mode by the lens also has the problem of the existing thinner and thinner handheld equipment and the irreconcilable adaptation on the focal length and the design thickness of the equipment.

From the optical principle perspective, if a piece of white paper is placed in a natural environment, there will not be any image on it, or there are countless multiple images on it, and the light reflected by all external objects can strike the white paper from all angles, and it is a blank piece when viewed, and if an image that can be viewed is to be formed, some incident light needs to be intensified. One of the methods is to use a lens, light rays at a specific distance can be converged to one point at another specific distance by the lens, light rays at other distances cannot be focused and cannot influence observation when being diffused in space, and a piece of white paper at the specific distance can naturally bear the light shadow of the candle. This is the content of physical experiments in junior middle schools, and is widely used in optical devices such as cameras and video cameras at present. The other method is pinhole imaging, and white paper in a dark room can present a clear inverted image because pinholes on the chamber wall act as a filtering device, and countless multiple influences of original scattering are filtered into only one picture, so that the imaging becomes clear. If the number of the small holes is large, imaging still becomes superposition of a plurality of pictures which are still blurred to the naked eye, but if the holes are regular small holes, the superposition of the plurality of pictures is regularly circulated in the principle of a signal and system, and an original picture can be obtained through inverse solution calculation in a certain mode. This method is disclosed in the academic paper "FlatCam: Thin, Bare-Sensor, Camera using Coded Aperture and computing", and we are working on applying this method to the field of under-screen biometric imaging for improved security.

Disclosure of Invention

Therefore, it is necessary to provide a technical solution that can perform the relevant encoding of the light wave before the incident light enters the sensor, and analyze the optical image of the encoding mask through the neural network.

In order to achieve the above object, the inventor provides a method for reconstructing biological characteristics of a coding mask, further comprising the following steps of establishing a neural network, training the neural network through a training sample, inputting the training sample into a biological characteristic photosensitive image filtered by the coding mask, and outputting the training sample into a biological characteristic original image corresponding to the biological characteristic photosensitive image, so as to obtain the trained neural network.

And acquiring the light-sensitive image of the biological characteristics filtered by the coding mask, inputting the light-sensitive image into the trained neural network, and acquiring a predicted value of the biological characteristic image output by the neural network.

Specifically, the neural network includes a convolutional layer and a deconvolution layer connected in sequence.

Specifically, the neural network comprises two convolutional layers and two deconvolution layers.

Optionally, the convolution kernel size of the convolution layer is 5 × 5, the step size is 2, and the number of feature maps is 64.

A storage medium for reconstructing biological characteristics of a coding mask stores a computer program, the computer program executes the following steps when being executed, a neural network is established, the neural network is trained through a training sample, the input of the training sample is a biological characteristic photosensitive image filtered by the coding mask, the output of the training sample is a biological characteristic original image corresponding to the biological characteristic photosensitive image, and the trained neural network is obtained.

Compared with the prior art, the scheme of the invention is designed by the method, so that the beneficial effect of restoring the original image of the filtered photosensitive image by means of the neural network is achieved, the requirement on the anti-resolving power resource of hardware equipment is reduced, the universality of the fingerprint identification method is improved, and the safety of the fingerprint identification method is also improved.

Drawings

FIG. 1 is a schematic diagram of an architecture of a biometric scanning unit of an off-screen biometric scanning unit according to an embodiment;

FIG. 2 is a diagram of an embodiment of a coding mask design pattern;

FIG. 3 is a diagram of an embodiment of an encoding mask pattern;

FIG. 4 is a diagram illustrating a sensor light-sensitive image and an original image according to an embodiment;

FIG. 5 is a flowchart of an exemplary method for on-screen biometric identification;

FIG. 6 is a flowchart of an exemplary method for on-screen biometric identification;

FIG. 7 is a diagram illustrating a method for reconstructing biometric features of a coded mask according to an embodiment;

FIG. 8 is a diagram of a coded mask biometric reconstruction neural network according to an embodiment;

FIG. 9 is a diagram illustrating a method for coding mask biometric analysis according to an embodiment;

FIG. 10 is a diagram of a coding mask biometric analysis neural network according to an embodiment.

Reference numerals:

2. encoding a mask;

21. and a coding unit.

Detailed Description

To explain technical contents, structural features, and objects and effects of the technical solutions in detail, the following detailed description is given with reference to the accompanying drawings in conjunction with the embodiments.

The purpose of the technical invention is realized by the following technical scheme:

referring to fig. 1, fig. 1 is a schematic diagram illustrating an architecture of an under-screen biometric scanning unit according to some embodiments of the present invention, where the biometric scanning unit may include a light-emitting device and a light-sensing device. The light emitting component is disposed in the illustrated light emitting layer and the light sensing component can be disposed in the sensor layer. Specific arrangements can be found in our other related patent schemes. The light emitted by the light-emitting component can leave an image in the photosensitive component in the sensor layer under the screen through total reflection. The touch screen under the screen is referred to as a touch screen which is used conventionally, for example, a common touch screen of a mobile phone may be a 4.7 inch screen, a 5.1 inch screen, a 5.4 inch screen, and the like, and by arranging a plurality of biological characteristic scanning units, it is intended that any range of the whole screen can acquire scanning data through the biological characteristic scanning units, but excessive energy is not consumed. The biometric features may be fingerprints, but it is clear enough to replace fingerprints with biometric patterns such as palm prints, foot prints, toe prints, retina, iris, etc. As a further scheme, in order to improve the security of the fingerprint identification intermediate image, we propose a biological feature identification structure under a coding mask type screen, which includes a display panel, a light source, and a sensor, and can be similar to the touch screen, the light emitting component, and the light sensing component, but different from the above. The light source and the sensor can be in a one-to-one form, follow the design concept of partial fingerprint imaging, illuminate the whole fingerprint by multiple light sources and then perform imaging, and also perform imaging after light of the light sources is totally reflected by the display panel. In our embodiment, the sensor is located at the extreme end of the optical path, and in front of the sensor, there is further designed a coding mask 2, the coding mask is disposed on the optical path in front of the sensor, and the coding mask is represented as a first matrix in a matrixing manner, and the first matrix satisfies the dirac function after the autocorrelation operation. In an embodiment, as shown in fig. 2, which shows a design pattern of a coding mask 2, the matrixing of the coding mask may be understood as projecting a two-dimensional structure of the coding mask to divide a plurality of blocks with equal horizontal and vertical directions. The encoding mask is a two-dimensional plane with a mixture of light shading and light transmitting parts, and then the matrixing of the encoding mask can be understood as minimizing the encoding mask into encoding units 21, the encoding units have two attributes of light shading parts and light transmitting parts, such as 0 and 1 respectively in the matrixing in the embodiment shown in fig. 2, and the encoding mask can be matrixed, so that the light shading parts and the light transmitting parts on the encoding mask meet the relation of row alignment and column alignment, each minimized encoding unit is usually a regular rectangle, but can also be stretched in the length or width direction, the encoding units forming the light transmitting parts can be all light transmitting or partially hollow, in the example shown in fig. 2, the encoding units can only partially transmit light, each mask unit in the encoding mask on the right side can only have a circular part to transmit light, and the circle can be changed into any shape in the encoding unit, as long as it conforms to the alignment of rows and columns in the entire code mask, the partial transmission does not affect the representation of the information on the small scale of the code mask, which in the example shown in fig. 2 can be matrixed

After the whole coding mask is matrixed by taking a coding unit as a unit, the matrix of the coding mask should meet the autocorrelation operation result of a Dirac function (delta function), the coding mask meeting the conditions can be reversely analyzed, and a specific analysis method can refer to the prior art. Fig. 3 shows an example of a specific available coding mask pattern, and fig. 4 shows an example of a corresponding relationship between a photosensitive image and an original image after passing through the coding mask, and it can be seen that a sensed image of a sensor is very blurred, and information in the sensed image cannot be distinguished by naked eyes. Only by reversely solving the first matrix information corresponding to the coding mask, the original image of the biological characteristics can be obtained. Through the scheme, the fingerprint information encryption method and the fingerprint information encryption device have the advantages that the effect of encrypting the fingerprint information is achieved, the security of inputting the fingerprint information is improved, and the technical effect of preventing the fingerprint information from being leaked is achieved.

In some other specific embodiments, the ratio of blocking to transmission of the coded mask is 1; the number ratio of the encoding units classified into the light shielding portions to the light transmitting portions in the encoding mask is 1: 1. In our example, the ratio of light blocking to light transmission is between 35: the image resolution and the signal-to-noise ratio between 65 and 65:35 are good, in the preferred embodiment, the ratio of the shading to the transmission of the coding mask is 50%, respectively, which is advantageous in that the final imaging has the best signal-to-noise ratio. In other preferred embodiments, the distance between the encoding mask and the sensor is set to be 30-100 sensor pixel widths, which has the advantage of relatively good information retention, and the imaging information on the sensor at too far distance becomes blurred and easily interfered, and the signal-to-noise ratio is reduced, and the imaging information on the sensor at too close distance is interfered by the diffraction image and also reduced.

In order to better improve the security of the whole scheme of the biological feature recognition, a coding mask type under-screen biological feature recognition system is also provided, and the system comprises the recognition structure, a storage unit and a processing unit; the storage unit is used for obtaining the photosensitive image received by the sensor, a first matrix corresponding to the coding mask is prestored, and the processing unit is used for reversely resolving the biological feature image according to the photosensitive image and the first matrix. Based on the scheme, the biological feature recognition structure can be integrated in the terminal equipment with the display panel, further, the storage unit and the processing unit can be integrated in the same terminal equipment, the storage unit and the processing unit can also be designed on a cloud server, the cloud server is used for reversely solving the photosensitive image and reversely solving the photosensitive image, the verification step can be further carried out, the biological feature recognition can be better completed, and the safety performance of the scheme can be further improved through cloud verification. Since the first matrix, which functions as a cryptographic function, is not stored on the local device terminal.

In a specific embodiment, as shown in the flow chart of FIG. 5, our off-screen biometric identification system performs the following method steps:

s500, the sensor under the screen receives the light-sensitive image which is reflected by the display panel and filtered by the coding mask,

s502 transmits the exposed image to a processing unit,

and S504, the processing unit calls the first matrix information recorded in the storage unit to reversely decode the photosensitive image.

And finally obtaining the biological characteristic image information.

In other embodiments, where the biometric identification structure is relatively small and is designed to identify only a portion of the area in the biometric image, our off-screen biometric identification system performs the following method steps:

s502 transmits the exposed image to a processing unit,

And finally obtaining partial biological characteristic image information.

Meanwhile, because the biological characteristic recognition structure is relatively small, and the plurality of structures act together, the coding masks in the plurality of structures are different, so that the second biological characteristic recognition structure also performs the following method steps:

s500 the second under-screen sensor receives the light-sensitive image after the light source is reflected by the display panel and filtered by the coding mask,

s502 transmits the exposed image to a processing unit,

and S504, the processing unit calls the second matrix information recorded in the storage unit to reversely decode the photosensitive image.

And finally, obtaining the second part of biological characteristic image information.

In other embodiments, as shown in fig. 6, the storage unit and the processing unit are located in the cloud, and the biometric device terminal further includes an off-screen biometric recognition system for performing the following method steps:

s600, the sensor under the screen receives the light-sensitive image which is reflected by the display panel and filtered by the coding mask,

s602, the photosensitive image is transmitted to the cloud server through the communication unit, and the device code information is uploaded at the same time,

s604 calls the first matrix information corresponding to the device code recorded in the storage unit of the cloud server, and reversely decodes the light-sensitive image.

And finally S606, obtaining the biological characteristic image information and returning the biological characteristic image information to the equipment.

In the embodiment shown in fig. 7, which is a schematic diagram of a coding mask biological feature reconstruction method, the scheme may begin with the steps of establishing a neural network, and training the neural network through a training sample, where the input of the training sample is a biological feature photosensitive image filtered by a coding mask, and the output is a biological feature original image corresponding to a biological feature image, so as to obtain a trained neural network.

S704, the photosensitive image of the biological characteristics filtered by the coding mask is obtained and input into the trained neural network, and the predicted value of the biological characteristic image output by the neural network is obtained.

In this embodiment, the neural network is dedicated to process the photosensitive image received by the sensor filtered by the coding mask in the above structure, and the photosensitive image filtered by the coding mask may correspond to the entire fingerprint or may be a partial fingerprint image obtained by a specific device. In a specific embodiment, the photosensitive image filtered by the encoding mask and the corresponding original fingerprint image are stored in the fingerprint database in advance, and the original image is obtained by performing inverse solution according to the matrix information and the signal and system principle by the method described above. In the process of the method, the matrix information between the plurality of groups of photosensitive images stored in the fingerprint database and the original image may be inconsistent, which is determined by the learning characteristics of the neural network, so the method may proceed in advance, and step S700 is to establish a fingerprint database in which the photosensitive images filtered by the encoding mask and the corresponding fingerprint original images are stored. When the number of the pre-stored fingerprint original images is enough, the steps can be conveniently carried out, a neural network framework for image processing is established, the corresponding fingerprint photosensitive images filtered by the coding mask are used as input, the corresponding fingerprint original images are used as output for training, and after the training result is stable, the neural network framework is specialized for processing the fingerprint photosensitive images filtered by the coding mask to calculate the neural network of the fingerprint original images. By applying the neural network image to the analysis processing of the fingerprint photosensitive image filtered by the coding mask, the calculation steps can be simplified, and the neural network image is converted into the neuron parameters of the neural network image in the encoding mask, such as de-duplication, normalization and splicing. And secondly, by utilizing the learning characteristic of the neural network, even if the fingerprint lacks partial information, the predicted value of the fingerprint original image of the corresponding complete fingerprint can be obtained in the neural network, and the more training samples, the more accurate the predicted result is. And the complete fingerprint image is restored through the neural network again, so that the information leakage is avoided, and the safety of the existing fingerprint analysis technology is improved.

In some embodiments, we can tailor the structure and corresponding parameters in the neural network. In the embodiment, the input of the neural network is a plurality of fingerprint photosensitive images filtered by a coding mask, convolution result data is obtained through the convolution layers, the number of layers can be set by the convolution layers according to needs, and then the convolution result data is input into the deconvolution layer, and finally the prediction result of the fingerprint original image corresponding to the fingerprint photosensitive image filtered by the coding mask is output by the neural network. Through the neural network architecture design, the analysis of the fingerprint photosensitive image filtered by the coding mask can be carried out more quickly, different results expressed by various different coding masks are better represented, and the universality is high.

In some embodiments as shown in fig. 8, our neural network comprises a first convolutional layer, a second convolutional layer, a first deconvolution layer, and a second deconvolution layer connected in sequence. Wherein the convolution kernel size of the first convolution layer is 5 × 5; the step length of the convolution kernel, namely the displacement of each kernel is 2, and the number of the feature maps is 64; the second deconvolution layer is arranged corresponding to the first convolution layer; the convolution kernel size, step length and feature map number of the second convolution layer can be set to be the same as those of the first convolution layer, and different parameters can be selected according to needs. In the embodiment shown in fig. 2, the size of the input coded mask filtered fingerprint sensing image is 80 × 80 pixels, and 40 × 40 data results are obtained after the first convolution layer; and inputting the result into the second convolution layer to obtain a secondary convolution result of 20 × 20. And performing deconvolution operation on the secondary convolution result through two deconvolution layers, and adjusting parameters to obtain a complete fingerprint image with the size of 160 × 160 and outputting the complete fingerprint image. Through the design, the fingerprint reconstruction step can be better carried out. From a practical perspective, the smaller the size of the convolution kernel, the finer the features extracted by the convolution algorithm, but the easier overfitting occurs, and the higher the computation power requirement, the larger the convolution kernel, the coarser the feature extraction, so that the matching result is not accurate enough. The step size is also selected to have the same characteristics, and the reader can adjust the size and the step size of the convolution kernel according to the requirement.

In other embodiments, the hierarchy of the neural network architecture may be further optimized, and a connection layer may be disposed in the second convolutional layer and the first deconvolution layer, where the connection layer is configured to process the convolution operation result of the second convolutional layer and input the result to the first deconvolution layer. A multilayer structure can be arranged in the connecting layer, each layer of structure is provided with a plurality of neurons, the more the connecting layer is, the more abundant the number of the neurons in each layer is, the more accurate the processing result of the neural network is, and the more the calculation power is occupied. In some embodiments, the device further comprises a first connecting layer and a second connecting layer, and the number of the neurons in each layer is set to be 400. By designing a multi-level connection layer, the processing capacity of the neural network is further improved, the processing efficiency is optimized, and the accuracy of the reconstructed fingerprint analysis is improved. The scheme of the invention is designed by the method, so that the beneficial effect of restoring the original image of the filtered photosensitive image by means of the neural network is achieved, the requirement on the anti-resolving power resource of hardware equipment is reduced, the universality of the fingerprint identification method is improved, and the safety of the fingerprint identification method is also improved.

In the embodiment shown in fig. 9, we also provide a method for analyzing a biometric characteristic, including the steps of, S900, establishing a neural network, the neural network including a convolutional layer, a connection layer, a characteristic value layer, and a deconvolution layer;

s902, training a neural network through a training sample, inputting a plurality of coded mask filtered fingerprint photosensitive images belonging to a complete fingerprint and outputting a plurality of coded mask filtered fingerprint original images, obtaining a neural network with neurons solidified after training is completed, and then inputting the obtained coded mask filtered fingerprint photosensitive images into the neural network in step S904 to obtain a characteristic value output by a characteristic value layer in the neural network. FIG. 10 is a neural network for biometric analysis of the present invention, which in our example comprises a first convolutional layer, a second convolutional layer, a first fully-connected layer, a eigenvalue layer, a second fully-connected layer, a first deconvolution layer, a second deconvolution layer, connected in sequence; the number of the specific convolution layer, the connection layer and the deconvolution layer can be set according to actual needs. By training the neural network in the above manner, after the training is completed, the neurons in the characteristic value layer and the neural network layers in front of the neurons are solidified to form a specific connection coefficient, and the fingerprint photosensitive image filtered by a specific coding mask can be output corresponding to a specific overall image, and similarly, the output in the characteristic value layer of the neural network can be specified. In our embodiment, the number of neurons in the eigenvalue layer is preferably 128, but may be any number such as 32, 64, 256, etc. When the number of the neurons is 128, the output of the characteristic value layer is a 128-dimensional vector, and the vector can be used for characterizing the output of the corresponding fingerprint original image, and can also characterize a plurality of fingerprint photosensitive images filtered by the corresponding encoding masks, and the more the number of the neurons in the characteristic value layer is, the more accurate the characterization is. When the trained neural network is actually applied, the final original fingerprint image does not need to be output, and only the feature value layer is calculated to call the feature value for representing the original fingerprint image. By designing the neural network for image processing with the characteristic value layer, the fingerprint photosensitive images filtered by the plurality of coding masks are converted into one characteristic value on the technical level, and the calculation of synthesizing the fingerprint photosensitive images filtered by the plurality of coding masks by analysis is avoided on the application level.

In order to better satisfy the requirement of the biometric analysis, the correspondence between the partial image of the biometric pattern obtained by the above analysis method and the feature value can be stored as a database, and the method further includes a step of storing the correspondence between the fingerprint exposed image filtered by the encoding mask and the feature value in the fingerprint database S806. For example, in an embodiment, the encoded mask-filtered fingerprint photosensitive images a1, a2, A3, and a4 all belong to the same fingerprint a1, and a1, a2, A3, and a4 are input into the trained neural network to obtain an output d1 of a feature value layer thereof, and a corresponding relationship between a fingerprint original image and a feature value [ a1, d1 ] can be stored in a database; the correspondence between the fingerprint-sensitive image filtered by the encoding mask and the feature value [ a1, a2, a3, a4, d1 ] can also be stored; it is even possible to store only the valid characteristic value d1 in the library. A large number of fingerprint feature values d1, d2, d3 … … d100, etc. constitute a fingerprint database. When the comparison is needed, the step of comparing the first fingerprint with the second fingerprint and obtaining the output of the characteristic value layer through the neural network obtained by training through the method can be carried out. And S808, matching the fingerprint image to be compared with the characteristic value of the fingerprint database. And if the difference value between the characteristic value obtained after the fingerprint photosensitive images filtered by the plurality of the coding masks to be compared are processed by the neural network and the existing characteristic value in the database is smaller than a preset threshold value, the fingerprint to be compared is considered to be matched with the fingerprint in the database. The calculation method of the difference value may refer to the calculation method of the vector difference value, and the preset threshold may be adjusted according to different actual situations, and is preferably 1. By establishing the fingerprint database through the scheme, the characteristic value of the existing fingerprint is stored, and the new fingerprint pattern is compared with the database when the fingerprint is subsequently verified, so that the comparison operation among different fingerprints can be realized. The efficiency of fingerprint identification is further enhanced.

In the preferred embodiment shown in fig. 10, the kernel size of the convolution is 5 x 5 with a displacement of 2 per kernel and the number of feature maps is 64. The number of neurons in the first fully-connected layer was 1600, and the number of neurons in the second layer was 6400. The feature code layer has 128 neurons, the coefficients of which are feature codes, which can be expressed as 128-dimensional vectors, and we perform fingerprint comparison by calculating the distance between feature codes obtained from different input images. Let the input image be x, its corresponding complete fingerprint image be y, and the output of the neural network be

The training objective is to minimize the reconstruction error function

The size and step length of the convolution kernel, the number of the feature maps and the number of the neurons can be changed according to specific needs. The preferred arrangement described above enables the present neural network to perform eigenvalue layer calculations better and faster and to increase robustness.

According to the technical scheme, the neural network for image processing with the characteristic value layer is designed, the fingerprint image after the coding mask filtering is converted into the characteristic value on the technical level, the calculation of inverse solution into the original fingerprint image is skipped on the application level, and as the integral fingerprint image is not finally synthesized, the data leakage and embezzlement are avoided, and the safety of the biological characteristic analysis method is improved.

In a specific embodiment, the method further includes a code mask biometric analysis storage medium storing a computer program, where the computer program, when executed, performs the following steps to establish a neural network, where the neural network includes a convolutional layer, a connection layer, a eigenvalue layer, and a deconvolution layer;

training a neural network through a training sample, wherein the training sample is a biological characteristic photosensitive image filtered by a coding mask, outputting a biological characteristic original image corresponding to the biological characteristic photosensitive image to obtain the trained neural network, and after the training is finished, acquiring a characteristic value output by a characteristic value layer in the neural network from the biological characteristic photosensitive image filtered by the coding mask.

Specifically, the computer program further executes a step when executed, wherein the biological characteristic print library stores characteristic values corresponding to the biological characteristic print.

Optionally, the computer program when executed further performs a step of comparing the feature value output by the neural network with a feature value in the biometric fingerprint library.

Preferably, the neural network includes a first convolutional layer, a second convolutional layer, a first fully-connected layer, a characteristic value layer, a second fully-connected layer, a first deconvolution layer, and a second deconvolution layer, which are connected in sequence.

In a specific embodiment, the system further comprises a biological characteristic analysis neural network, wherein the neural network comprises a convolution layer, a connection layer and a characteristic value layer which are sequentially connected; the neuron connection weight curing relation among the convolutional layers, the connection layers and the characteristic value layers is formed by the following steps:

establishing a neural network, wherein the neural network comprises a convolutional layer, a connection layer, a characteristic value layer and a deconvolution layer;

and training the neural network through a training sample, wherein the input of the training sample is a biological characteristic photosensitive image filtered by a coding mask, and the output of the training sample is a biological characteristic original image corresponding to the biological characteristic photosensitive image until the training is finished.

It should be noted that, although the above embodiments have been described herein, the invention is not limited thereto. Therefore, based on the innovative concepts of the present invention, the technical solutions of the present invention can be directly or indirectly applied to other related technical fields by making changes and modifications to the embodiments described herein, or by using equivalent structures or equivalent processes performed in the content of the present specification and the attached drawings, which are included in the scope of the present invention.

Claims

1. A coding mask biological characteristic reconstruction method is characterized by comprising the following steps of establishing a neural network, training the neural network through a training sample, inputting a biological characteristic photosensitive image filtered by a coding mask into the training sample, and outputting a biological characteristic original image corresponding to the biological characteristic photosensitive image to obtain the trained neural network;

2. The method of coded matte biometric reconstruction according to claim 1, wherein the neural network comprises a convolutional layer and an anti-convolutional layer connected in sequence.

3. The method of coded matte biometric reconstruction according to claim 2, wherein the neural network comprises two convolutional layers, two anti-convolutional layers.

4. The method of coded matte biometric reconstruction according to claim 2, wherein the convolution kernel size of the convolution layer is 5 x 5, the step size is 2, and the number of feature maps is 64.

5. A coding mask biological characteristic reconstruction storage medium is characterized in that a computer program is stored, and when the computer program is operated, the computer program executes the following steps of establishing a neural network, training the neural network through a training sample, inputting a biological characteristic photosensitive image filtered by a coding mask, and outputting a biological characteristic original image corresponding to the biological characteristic photosensitive image to obtain the trained neural network;

6. The coded matte biometric reconstruction storage medium of claim 5, wherein the neural network comprises sequentially connected convolutional layers and anti-convolutional layers.

7. The coded matte biometric reconstruction storage medium of claim 6, wherein the neural network comprises two convolutional layers, two anti-convolutional layers.

8. The storage medium for coded matte biometric reconstruction according to claim 6, wherein the convolution kernel size of the convolution layer is 5 x 5, the step size is 2, and the number of feature maps is 64.