CN112069482A - An Identity Authentication System Based on Footprint Comparison Algorithm - Google Patents

Abstract

The invention discloses an identity authentication system based on a footprint comparison algorithm. The invention relates to the technical field of identity authentication, in particular to an identity authentication system based on a footprint comparison algorithm, which can help warehouses, banks and prisons in specific places to achieve Automation of admission identity authentication; footprints are a type of biometrics, which can represent a person's identity information; footprints do not need the deliberate cooperation of the collected personnel during the collection process, and the collection device is installed at the entrance and exit of the relevant place. , disguised as normal ground to collect hidden footprints; using the identity authentication system based on the footprint comparison algorithm proposed by the present invention, the insiders of the relevant places only need to collect footprint data once when registering information, and the work of identity authentication will be completed. It is completed by the footprint identity authentication system laid at the entrance and exit; it solves the problem that the current security systems of warehouses, banks, and prisons have insufficient security protection capabilities.

Description

An Identity Authentication System Based on Footprint Comparison Algorithm

technical field

The invention relates to the technical field of identity authentication, in particular to an identity authentication system based on a footprint comparison algorithm.

Background technique

In principle, non-internal personnel are not allowed to enter the warehouses, bank office areas, and prisons where goods are stored. Therefore, security monitoring of these areas is required. The traditional security method requires cameras or IC card-based access control systems at the entrances and exits, which will increase the workload of security staff in the actual operation process, and also bring trouble to those who need to frequent these places. And these methods are not very concealed, and it is easy to be fooled by criminals.

The invention proposes an identity authentication system based on a footprint comparison algorithm, which can construct an identity authentication system and can help warehouses, banks, and prisons to realize the automation of entry and exit identity authentication; Indicates a person's identity information; footprints do not need the deliberate cooperation of the collectors during the collection process. Install the collection device at the entrance and exit of the relevant place, and disguise it as a normal ground to collect hidden footprints.

By adopting the identity authentication system based on the footprint comparison algorithm proposed by the present invention, the internal personnel of the relevant places only need to collect the footprint data once when registering the information, and then they can enter and exit freely. The authentication system is complete.

SUMMARY OF THE INVENTION

(1) Technical problems solved

Aiming at the deficiencies of the prior art, the present invention provides an identity authentication system based on a footprint comparison algorithm, which solves the problems that the current security systems of warehouses, banks, and prisons have insufficient security protection capability and troublesome operation processes.

(2) Technical solutions

In order to achieve the above purpose, the present invention provides the following technical solutions: an identity authentication system based on a footprint comparison algorithm, including a supervision area, a gate channel, a footprint collector and a footprint authentication system, and entering the channel and going out in the supervision area. Each channel is provided with a gate channel, and a footprint collector is arranged at the entry end of the gate channel; the gate channel and the footprint collector are both connected in communication with the footprint authentication system.

As an optimization, the backbone network of the footprint authentication system is VGG11; first, the two footprint images to be compared are spliced in the depth direction, input into VGG11 to extract comparison features, and the 1000-dimensional output of the last layer of VGG11 is used. The vector is used as a comparison feature vector, and the similarity between the two footprints can be calculated by being input into the similarity score calculation network (Similarity ScoreComputing Network, SSCN) designed by the present invention; if the similarity is greater than the set threshold TH It is considered that two footprints belong to the same person, and vice versa, that the two footprints do not belong to the same person. In the footprint comparison network, the similarity score calculation network has two fully connected layers. The first fully connected layer receives the 1000-dimensional comparison feature vector output by VGG11, reduces the dimension to 30 dimensions, and is followed by a BN (Batch Normalization) layer and a ReLU activation function layer, the second fully connected layer accepts the 30-dimensional feature vector output by the first layer, and outputs a one-dimensional vector as the predicted similarity value. This similarity value cannot be directly used as the final The similarity prediction result also needs to be converted into a value between 0 and 1 using the Sigmoid activation function as the final prediction result.

A footprint alignment algorithm includes the following steps:

Step 1: Identity collection: Collect the information of the staff. The three numbers 1, 2, and 3 represent the identity information of the three people. When registering the information, several footprint images and ID cards of the three people will be collected. number, job number and other information, among which, each person's footprint image corresponds to their personal information;

Step 2: Footprint image collection preprocessing, information archiving: After preprocessing the registrant's footprint image, irrelevant information on the image will be removed, and the image resolution will be reduced to a large scale storage without affecting identity authentication The degree of accuracy, after preprocessing, the images collected during registration will be archived and stored in the database;

Step 3: Information comparison: When people enter the venue, their footprint images will be collected, and then their images will be compared with the images in the database to form a pair of images, which will be sent to the footprint authentication system. , the similarity score between the images will be output; assuming that the threshold set by the system TH=0.7, the similarity between the identity authentication person’s footprint and the footprint data of No. 1 is 0.8>TH, indicating that this person is 1 in the database. If the similarity between any footprint image in the database and the image to be authenticated is greater than the threshold, the footprint authentication system will consider that the person has not registered information and will not be released.

As an optimization, the step 2: footprint image collection preprocessing, information archiving, during footprint collection: the person to be collected walks through the footprint collector, and the image collected by the footprint collector is an image with 256 gray levels. Grayscale image;

Footprint image preprocessing: The main purpose of image preprocessing is to remove the noise on the image and collect the scale information that comes with the instrument to facilitate image comparison, and to reduce the size of the image for archiving.

As an optimization, the step 2: footprint image acquisition preprocessing, information archiving, includes the following steps:

a. Remove the ruler information on the footprint image. The fixed area of the image collected by the footprint collector will have a ruler information for calculating the size of the footprint. This information is not useful for the comparison of the footprint, so it needs to be removed. The specific operation is The gray value of the fixed area of the image is directly assigned to 255;

b. Based on the Otsu algorithm for denoising and footprint positioning, since the footprint area and the background area of the collected footprint image are easy to distinguish, a distinction threshold th is determined based on the Otsu algorithm; the pixels with a grayscale less than th in the image are considered to be The points in the footprint area, and the rest of the points are considered as the background. After determining th, you can scan the footprint image from four directions, up, down, left, and right, and detect the four boundaries of the footprint area, thereby completing the footprint positioning; the footprint positioning is completed. After that, since most of the noise is outside the footprint area, the denoising work is actually completed at this time;

c. Adjust the size of the footprint image and perform grayscale inversion. The location of the footprint area has been determined in b. Generally, the distance between the upper and lower boundaries of the footprint is greater than the distance between the left and right boundaries, so the system considers that the distance between the upper and lower boundaries is smaller than the left and right boundaries. The footprint image is an abnormal footprint image, and an alarm should be issued in the specific implementation of the system; after determining the distance between the upper and lower boundaries, define a square area with a pixel value of 255, and the side length is the distance between the upper and lower boundaries; Copy directly to the center of the square area, and subtract the pixel value of each point in the square area from 255 to perform grayscale inversion at the pixel level; finally, downsample the obtained square area, adjust the resolution, and save it.

As optimization, the described step 3: the information comparison includes the following steps:

The first step, network training: Divide the collected footprint samples into two parts according to the collected objects, one part of the object's footprint samples are used as training set samples, and the other part is used as validation set samples;

If the two footprint images input to the network belong to the same person, the two images form a positive sample X, and the label Y is 1; otherwise, the two images form a negative sample X, and the label Y is 0; each time from the training set Random sampling, obtain a batch of positive samples or negative samples, input them into the network for similarity calculation, and make an MSE loss on the output value and label. The expression of the loss is:

In the formula, B represents the batch size during training, and FCN represents the footprint comparison network. Input any sample X into the network to obtain the predicted label of the network;

During training, the loss L of each batch of samples is back-propagated, and then the model is optimized by gradient descent; when there are enough training batches, the network is equivalent to optimizing parameters on all possible positive and negative sample spaces; Repeatedly randomly select positive and negative samples for model optimization, and the final model will gradually converge;

The second step, model selection: The model to be saved by the neural network generally needs to be selected from the intermediate model trained by a validation set. When selecting the model, it is necessary to randomly select enough samples from the validation set and input them into the network for Similarity calculation, if the similarity is greater than 0.5, it is considered to be the footprint of the same person, and if the similarity is less than 0.5, it is not the footprint of the same person, and the intermediate model with the highest footprint comparison accuracy is taken as the final saved model.

(3) Beneficial effects

The invention provides an identity authentication system based on a footprint comparison algorithm. Has the following beneficial effects:

The invention provides an identity authentication system based on a footprint comparison algorithm, which can help warehouses, banks, and prisons realize the automation of identity authentication for entry and exit; footprints are a kind of biological feature, which can represent a person's identity information The footprints do not need the deliberate cooperation of the collected personnel during the collection process, and the collection device is installed at the entrance and exit of the relevant place, and the hidden footprints can be collected by disguising the normal ground; Algorithmic identity authentication system, insiders of relevant places only need to collect footprint data once when registering information, and then they can freely enter and exit. The work of identity authentication is completed by the footprint identity authentication system laid at the entrance and exit; The security systems of banks and prisons have insufficient security capabilities, and the operation process is more troublesome.

Description of drawings

FIG. 1 is a schematic structural diagram of the present invention.

FIG. 2 is a schematic diagram of a footprint image preprocessing process of the present invention.

FIG. 3 is a network diagram of footprint comparison of the present invention.

FIG. 4 is a flowchart of the footprint identity authentication of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Example 1

As shown in Figures 1-4, the present invention provides a technical solution: an identity authentication system based on a footprint comparison algorithm, including a supervision area, a gate channel, a footprint collector and a footprint authentication system. The entry channel and the exit channel are each provided with a gate channel, and a footprint collector is arranged at the entry end of the gate channel; the gate channel and the footprint collector are both connected in communication with the footprint authentication system.

In this embodiment, the backbone network of the footprint authentication system is VGG11; first, the two footprint images to be compared are spliced in the depth direction, input into VGG11 to extract comparison features, and the last layer of VGG11 is output. The 1000-dimensional vector is used as the comparison feature vector, and the similarity between the two footprints can be calculated by being input into the similarity score calculation network (SimilarityScore Computing Network, SSCN) designed by the present invention; if the similarity is greater than the set threshold TH can think that the two footprints belong to the same person, and vice versa, that the two footprints do not belong to the same person. In the footprint comparison network, the similarity score calculation network has two fully connected layers. The first fully connected layer receives the 1000-dimensional comparison feature vector output by VGG11, reduces the dimension to 30 dimensions, and is followed by a BN (Batch Normalization) layer and a ReLU activation function layer, the second fully connected layer accepts the 30-dimensional feature vector output by the first layer, and outputs a one-dimensional vector as the predicted similarity value. This similarity value cannot be directly used as the final The similarity prediction result also needs to be converted into a value between 0 and 1 using the Sigmoid activation function as the final prediction result.

A footprint alignment algorithm includes the following steps:

Step 1: Identity collection: Collect the information of the staff. The three numbers 1, 2, and 3 represent the identity information of the three people. When registering the information, several footprint images and ID cards of the three people will be collected. number, job number and other information, among which, each person's footprint image corresponds to their personal information;

Step 2: Footprint image collection preprocessing, information archiving: After preprocessing the registrant's footprint image, irrelevant information on the image will be removed, and the image resolution will be reduced to a large scale storage without affecting identity authentication The degree of accuracy, after preprocessing, the images collected during registration will be archived and stored in the database;

Step 3: Information comparison: When people enter the venue, their footprint images will be collected, and then their images will be compared with the images in the database to form a pair of images, which will be sent to the footprint authentication system. , the similarity score between the images will be output; assuming that the threshold set by the system TH=0.7, the similarity between the identity authentication person’s footprint and the footprint data of No. 1 is 0.8>TH, indicating that this person is 1 in the database. If the similarity between any footprint image in the database and the image to be authenticated is greater than the threshold, the footprint authentication system will consider that the person has not registered information and will not be released.

In this embodiment, the step 2: footprint image collection preprocessing, information archiving, during footprint collection: the person to be collected walks through the footprint collector, and the image collected by the footprint collector is a piece of 256 gray degree-level grayscale image;

Footprint image preprocessing: The main purpose of image preprocessing is to remove the noise on the image and collect the scale information that comes with the instrument to facilitate image comparison, and to reduce the size of the image for archiving.

In this embodiment, the step 2: footprint image acquisition preprocessing, information archiving, includes the following steps:

a. Remove the ruler information on the footprint image. The fixed area of the image collected by the footprint collector will have a ruler information for calculating the size of the footprint. This information is not useful for the comparison of the footprint, so it needs to be removed. The specific operation is The gray value of the fixed area of the image is directly assigned to 255;

b. Based on the Otsu algorithm for denoising and footprint positioning, since the footprint area and the background area of the collected footprint image are easy to distinguish, a distinction threshold th is determined based on the Otsu algorithm; the pixels with a grayscale less than th in the image are considered to be The points in the footprint area, and the rest of the points are considered as the background. After determining th, you can scan the footprint image from four directions, up, down, left, and right, and detect the four boundaries of the footprint area, thereby completing the footprint positioning; the footprint positioning is completed. After that, since most of the noise is outside the footprint area, the denoising work is actually completed at this time;

c. Adjust the size of the footprint image and perform grayscale inversion. The location of the footprint area has been determined in b. Generally, the distance between the upper and lower boundaries of the footprint is greater than the distance between the left and right boundaries, so the system considers that the distance between the upper and lower boundaries is smaller than the left and right boundaries. The footprint image is an abnormal footprint image, and an alarm should be issued in the specific implementation of the system; after determining the distance between the upper and lower boundaries, define a square area with a pixel value of 255, and the side length is the distance between the upper and lower boundaries; Copy directly to the center of the square area, and subtract the pixel value of each point in the square area from 255 to perform grayscale inversion at the pixel level; finally, downsample the obtained square area, adjust the resolution, and save it.

In this embodiment, the step 3: the information comparison includes the following steps:

The first step, network training: Divide the collected footprint samples into two parts according to the collected objects, one part of the object's footprint samples are used as training set samples, and the other part is used as validation set samples;

If the two footprint images input to the network belong to the same person, the two images form a positive sample X, and the label Y is 1; otherwise, the two images form a negative sample X, and the label Y is 0; each time from the training set Random sampling, obtain a batch of positive samples or negative samples, input them into the network for similarity calculation, and make an MSE loss on the output value and label. The expression of the loss is:

In the formula, B represents the batch size during training, and FCN represents the footprint comparison network. Input any sample X into the network to obtain the predicted label of the network;

During training, the loss L of each batch of samples is back-propagated, and then the model is optimized by gradient descent; when there are enough training batches, the network is equivalent to optimizing parameters on all possible positive and negative sample spaces; Repeatedly randomly select positive and negative samples for model optimization, and the final model will gradually converge;

The second step, model selection: The model to be saved by the neural network generally needs to be selected from the intermediate model trained by a validation set. When selecting the model, it is necessary to randomly select enough samples from the validation set and input them into the network for Similarity calculation, if the similarity is greater than 0.5, it is considered to be the footprint of the same person, and if the similarity is less than 0.5, it is not the footprint of the same person, and the intermediate model with the highest footprint comparison accuracy is taken as the final saved model.

working principle:

When using, first collect information on the staff. The person to be collected walks through the footprint collector. The image collected by the footprint collector is a grayscale image with 256 gray levels. The fixed area of the image collected by the footprint collector will be It comes with a ruler for calculating the size of the footprint. This information is not very useful for the comparison of the footprint, so it needs to be removed. The specific operation is to directly assign the gray value of the fixed area of the image to 255, because the collected footprint image footprint The area and the background area are easy to distinguish, so a distinction threshold th is determined based on the Otsu algorithm; the pixel points in the image whose gray level is less than th are considered as the points of the footprint area, and the remaining points are considered as the background. Based on this, the footprint image is scanned from the four directions of up, down, left, and right, and the four boundaries of the footprint area are detected to complete the footprint location; after the footprint location is completed, since most of the noise is outside the footprint area, it is actually Complete the denoising work; adjust the size of the footprint image and perform grayscale inversion, after determining the distance between the upper and lower boundaries, define a square area with a pixel value of 255, and the side length is the distance between the upper and lower boundaries; Copy directly to the center of the square area, and subtract the pixel value of each point in the square area from 255 to perform grayscale inversion at the pixel level; finally, downsample the obtained square area, adjust the resolution, and save it. In the process of use, when people enter the venue, their footprint images will be collected, and then their images will be compared with the images in the database to form an image pair, which will be sent to the footprint authentication system. , will output the similarity score between the images; assuming that the threshold TH=0.7 set by the system, the similarity between the person's footprint and the No. 1 footprint data for identity authentication is 0.8>TH, indicating that this person is the No. 1 person in the database; If the comparison is passed, it can be released; if the similarity between any footprint image in the database and the image to be authenticated is greater than the threshold, the footprint authentication system will consider that the person has not registered information and will not be released.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, and substitutions can be made in these embodiments without departing from the principle and spirit of the invention and modifications, the scope of the invention is defined by the appended claims and their equivalents.

Claims (5)

1. a kind of identity authentication system based on footprint comparison algorithm, it is characterized in that: this identity authentication system based on footprint comparison algorithm comprises supervision area, gate passage, footprint collector and footprint authentication system, in described supervision The entry channel and the exit channel of the area are each provided with a gate channel, and a footprint collector is set at the entry end of the gate channel; the gate channel and the footprint collector are both connected in communication with the footprint authentication system. 2. a footprint comparison algorithm, is characterized in that: described footprint comparison algorithm comprises the following steps: Step 1: Identity collection: Collect the information of the staff. The three numbers 1, 2, and 3 represent the identity information of the three people. When registering the information, several footprint images and ID cards of the three people will be collected. number, job number and other information, among which, each person's footprint image corresponds to their personal information; Step 2: Footprint image collection preprocessing, information archiving: After preprocessing the registrant's footprint image, irrelevant information on the image will be removed, and the image resolution will be reduced to a large scale storage without affecting identity authentication The degree of accuracy, after preprocessing, the images collected during registration will be archived and stored in the database; Step 3: Information comparison: When people enter the venue, their footprint images will be collected, and then their images will be compared with the images in the database to form a pair of images, which will be sent to the footprint authentication system. , the similarity score between the images will be output; assuming that the threshold set by the system TH=0.7, the similarity between the identity authentication person’s footprint and the footprint data of No. 1 is 0.8>TH, indicating that this person is 1 in the database. If the similarity between any footprint image in the database and the image to be authenticated is greater than the threshold, the footprint authentication system will consider that the person has not registered information and will not be released. 3. The footprint comparison algorithm according to claim 2 is characterized in that: said step 2: footprint image collection preprocessing, information archiving, when footprint collection: the person to be collected walks through the footprint collector, and footprint collection The image collected by the detector is a grayscale image with 256 grayscales; Footprint image preprocessing: The main purpose of image preprocessing is to remove the noise on the image and collect the scale information that comes with the instrument to facilitate image comparison, and to reduce the size of the image for archiving. 4. footprint comparison algorithm according to claim 2 is characterized in that: described step 2: footprint image acquisition preprocessing, information archiving, comprises the following steps: a. Remove the ruler information on the footprint image. The fixed area of the image collected by the footprint collector will have a ruler information for calculating the size of the footprint. This information is not useful for the comparison of the footprint, so it needs to be removed. The specific operation is The gray value of the fixed area of the image is directly assigned to 255; b. Based on the Otsu algorithm for denoising and footprint positioning, since the footprint area and the background area of the collected footprint image are easy to distinguish, a distinction threshold th is determined based on the Otsu algorithm; the pixels with a grayscale less than th in the image are considered to be The points in the footprint area, and the rest of the points are considered as the background. After determining th, you can scan the footprint image from four directions, up, down, left, and right, and detect the four boundaries of the footprint area, thereby completing the footprint positioning; the footprint positioning is completed. After that, since most of the noise is outside the footprint area, the denoising work is actually completed at this time; c. Adjust the size of the footprint image and perform grayscale inversion. The location of the footprint area has been determined in b. Generally, the distance between the upper and lower boundaries of the footprint is greater than the distance between the left and right boundaries, so the system considers that the distance between the upper and lower boundaries is smaller than the left and right boundaries. The footprint image is an abnormal footprint image, and an alarm should be issued in the specific implementation of the system; after determining the distance between the upper and lower boundaries, define a square area with a pixel value of 255, and the side length is the distance between the upper and lower boundaries; Copy directly to the center of the square area, and subtract the pixel value of each point in the square area from 255 to perform grayscale inversion at the pixel level; finally, downsample the obtained square area, adjust the resolution, and save it. 5. footprint comparison algorithm according to claim 2 is characterized in that: described step 3: information comparison comprises the following steps: The first step, network training: Divide the collected footprint samples into two parts according to the collected objects, one part of the object's footprint samples are used as training set samples, and the other part is used as validation set samples; If the two footprint images input to the network belong to the same person, the two images form a positive sample X, and the label Y is 1; otherwise, the two images form a negative sample X, and the label Y is 0; each time from the training set Random sampling, obtain a batch of positive samples or negative samples, input them into the network for similarity calculation, and make an MSE loss on the output value and label. The expression of the loss is:

In the formula, B represents the batch size during training, and FCN represents the footprint comparison network. Input any sample X into the network to obtain the predicted label of the network; During training, the loss L of each batch of samples is back-propagated, and then the model is optimized by gradient descent; when there are enough training batches, the network is equivalent to optimizing parameters on all possible positive and negative sample spaces; Repeatedly randomly select positive and negative samples for model optimization, and the final model will gradually converge; The second step, model selection: The model to be saved by the neural network generally needs to be selected from the intermediate model trained by a validation set. When selecting the model, it is necessary to randomly select enough samples from the validation set and input them into the network for Similarity calculation, if the similarity is greater than 0.5, it is considered to be the footprint of the same person, and if the similarity is less than 0.5, it is not the footprint of the same person, and the intermediate model with the highest footprint comparison accuracy is taken as the final saved model.

Images