A method and system for authenticating a management cabinet
Through facial live detection and image steganography encryption technology, the lack of security and user information leakage of traditional management cabinet authentication methods are solved, and efficient authentication security and data protection are achieved.
Patent Information
- Application Number
- CN202410467641.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-18
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-04-18
AI Technical Summary
The traditional management cabinet authentication method has the problem of key copying and facial recognition being easily attacked by non-living bodies, resulting in insufficient security and risk of user information leakage.
Using the authentication method of live facial detection combined with image steganography encryption, dynamic feature extraction and static descriptor comparison are obtained by obtaining continuous face images, combined with SURF algorithm and image convolution fusion technology to achieve authenticity and privacy of faces Protect.
Effectively prevent in vivo attacks, improve authentication security and accuracy, and at the same time protect user personal information, enhancing the security and data confidentiality of the management cabinet.
Smart Images

Figure CN118570848B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of security authentication technology, and in particular to an authentication method and system for a management cabinet. Background Art
[0002] The traditional management cabinet authentication method is to use a key or a key card to open the cabinet. The key and the key card are not unique and may be copied or stolen, allowing unauthorized personnel to enter the management cabinet. Especially when using traditional physical keys, the security risk is higher because the key can be copied without leaving any traces. The use of face recognition technology for management cabinet authentication improves security, but face recognition is vulnerable to non-living attacks, using static photos, videos or masks and other non-living methods to deceive the management cabinet to obtain unauthorized management cabinet access rights. The photo library used for face recognition may also be invaded by criminals, resulting in user information leakage.
[0003] Therefore, we urgently need to provide an authentication method and system for management cabinets with face liveness detection and encrypted storage of face images. Summary of the invention
[0004] The present invention aims to provide an authentication method and system for a management cabinet with face liveness detection and encrypted storage information, which improves the security of management cabinet authentication and effectively protects user personal information.
[0005] An authentication method for a management cabinet, comprising:
[0006] S1. Obtain m frames of continuous face images to be authenticated o i , i = 1, 2, ..., m; m frames of continuous face images to be authenticated o i Combine into a face image set to be authenticated; obtain the authentication time at the current moment and obtain the timestamp to be authenticated;
[0007] S2. Based on the face liveness detection model, feature extraction is performed on the face image set to be authenticated to obtain dynamic features of the face image to be authenticated; liveness detection is performed according to the dynamic features of the face image to be authenticated to obtain the liveness detection result to be authenticated; if the liveness detection result to be authenticated is liveness, proceed to step S3; otherwise, the authentication is failed;
[0008] S3. Extract key features from the face image set to be authenticated, obtain static descriptors of the face images to be authenticated, compare the static descriptors of the face images to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtain the comparison results of the faces to be authenticated; if the comparison result of the face to be authenticated is successful, the authentication is passed and the door of the management cabinet is opened; otherwise, the authentication is failed;
[0009] Obtain the authenticated user's face image, the authenticated user's user code and corresponding personal information;
[0010] The public database of the management cabinet face authentication comparison library extracts features from the authenticated user's face image to obtain the authenticated user's face feature descriptor; all authenticated user's face feature descriptors and their corresponding authenticated user codes are combined to obtain;
[0011] The private database of the face authentication comparison database of the management cabinet is encrypted by image steganography of the authenticated user's face image and its corresponding personal information to obtain the encrypted authenticated user's face image; the encrypted authenticated user's face images of all authenticated users and their corresponding authenticated user codes are combined to obtain;
[0012] The public database of the management cabinet face authentication comparison database and the private database of the management cabinet face authentication comparison database are combined to obtain the management cabinet face authentication comparison database.
[0013] As a preferred technical solution of the present invention, the face liveness detection model includes a data preprocessing layer, m feature extraction layers, a feature cascade layer, a face liveness comparison layer, and a result output layer;
[0014] The data preprocessing layer is used to convert the face image to be authenticated into i Perform preprocessing to obtain the preprocessed face image x to be authenticated i ;
[0015] m feature extraction layers are used to extract any preprocessed face image x to be authenticated. i Do the following:
[0016] Preprocess the face image x to be authenticated i Perform optical flow feature extraction to obtain the optical flow vector field t of the face image to be authenticated of the preprocessed face image xi i ; The optical flow vector field t of the face image to be authenticated i Perform sparse feature extraction to obtain the dynamic optical flow feature w of the face image to be authenticated i ;
[0017] Preprocess the face image x to be authenticated i Perform segmentation to obtain 2*2 segmented face images x to be authenticated i1 , x i2 , x i3 , x i4 ; Segment the face image x to be authenticated i1 , x i2 , x i3 , x i4The LTP operator is used to extract features and obtain the local features y1, y2, y3, y4 of the face image to be authenticated; the local features y1, y2, y3, y4 of the face image to be authenticated are cascaded to obtain the local features Fi of the face image to be authenticated. y = [y1, y2, y3, y4];
[0018] The dynamic optical flow feature w of the face image to be authenticated i and the local features Fi of the face image to be authenticated y Perform feature cascading to obtain the dynamic feature F of the face image to be authenticated xi =[T i ,Fi y ];
[0019] The feature cascade layer is used to dynamically identify all face images to be authenticated. xi Combine to obtain a dynamic feature sequence of the face image to be authenticated, perform maximum pooling on the dynamic feature sequence of the face image to be authenticated, and obtain the dynamic features of the face image to be authenticated;
[0020] The result output layer is used to output the liveness detection result to be authenticated according to the dynamic features of the face image to be authenticated. If the liveness detection result to be authenticated is liveness, go to step S3; otherwise, it prompts that the authentication fails.
[0021] As a preferred technical solution of the present invention, the specific steps of training a face liveness detection model are as follows:
[0022] H1. Collect several live and non-live face image sets, each of which contains m consecutive face images, and the target value of each face image set is live or non-live; combine the face image sets marked with the target values to obtain a live face detection training set;
[0023] H2. Input the face liveness detection training set into the face liveness detection initial model, train it with the target value as the target, and obtain the face liveness detection accuracy. If the face liveness detection accuracy is within the error standard range, output the trained face liveness detection initial model, that is, the face liveness detection model; otherwise, continue model training using the face liveness detection training set.
[0024] As a preferred technical solution of the present invention, the specific steps of step S3 include:
[0025] S31. The face image to be authenticated in the face image set o i Perform SURF algorithm feature extraction to obtain the face image to be authenticated o i The static descriptor z of the face image to be authenticated i ;
[0026] S32. Static descriptors z of all face images to be authenticated i Perform maximum pooling to obtain the static descriptor of the face image to be authenticated;
[0027] S33. Perform a similarity comparison one by one between the static descriptor of the face image to be authenticated and the facial feature descriptors of authenticated users in the public library of the management cabinet face authentication comparison library to obtain the comparison similarity of the face to be authenticated; if the comparison similarity of the face to be authenticated reaches the set threshold, the comparison result of the face to be authenticated is a successful comparison, indicating that the authentication is passed and the door of the management cabinet is opened; otherwise, the comparison result of the face to be authenticated is a failed comparison, indicating that the authentication is failed.
[0028] As a preferred technical solution of the present invention, the specific steps of constructing a management cabinet face recognition comparison library are as follows:
[0029] Obtain the face images of n authenticated users, the user codes of the authenticated users and the corresponding personal information, and obtain the face image p of the authenticated user j , where j = 1, 2, ..., n; n is the number of authenticated users in the management cabinet;
[0030] The authenticated user face image pj and its corresponding personal information are encrypted by image steganography to obtain the encrypted authenticated user face image pm j ;
[0031] The authenticated user's face image p j Perform SURF algorithm feature extraction to obtain the authenticated user's face feature descriptor pt j ;
[0032] PM all encrypted authenticated user face images j and its corresponding authenticated user code storage, obtain the face authentication comparison private library of the management cabinet, and upload the face authentication comparison private library of the management cabinet to the blockchain for storage;
[0033] All authenticated user face feature descriptors pt j and its corresponding authenticated user code storage, and obtain the management cabinet face authentication comparison library public library, and the management cabinet face authentication comparison library public library is stored locally;
[0034] The user code of the same authenticated user in the public database of the management cabinet face authentication comparison library and the private database of the management cabinet face authentication comparison library is the same.
[0035] As a preferred technical solution of the present invention, the specific steps of image steganography encryption include:
[0036] The authenticated user's face image p j Perform preprocessing to obtain the preprocessed authenticated user face image h j; The authenticated user face image p j The corresponding personal information is preprocessed to obtain the preprocessed authenticated user personal information g j ;
[0037] Pre-process the authenticated user's personal information g j Perform hash calculation to obtain the authenticated user's personal information hash value c j ;
[0038] The preprocessed authenticated user face image h j and the authenticated user's personal information hash value c j Input into the image steganography encryption model to obtain the encrypted authenticated user face image pm j .
[0039] As a preferred technical solution of the present invention, the image steganography encryption model includes an image input layer, an image convolution fusion layer, a watermarking layer, and an image output layer;
[0040] The image input layer is used to preprocess the authenticated user face image h j and any natural host image in the natural host image set are input into the image convolution fusion layer; the natural host image set includes a plurality of natural host images, and the plurality of natural host images are obtained by preprocessing a plurality of natural information images;
[0041] The image fusion convolution layer includes k convolution layers α u , u=1, 2, …, k; the image fusion convolution layer is used to pre-process the authenticated user face image h j Perform image convolution fusion with the natural host image to obtain the fused authenticated face image γ k ;
[0042] Adding a watermark layer to fuse authenticated face images k Add a digital watermark to the outer circle to get the encrypted authenticated user face image pm j ;
[0043] The image output layer is used to output the encrypted authenticated user face image pm j .
[0044] As a preferred technical solution of the present invention, the specific steps of encrypting an image by the image steganographic encryption model include:
[0045] In the image input layer, any natural host image in the natural host image set is selected and input into the image fusion convolution layer α1, and the pre-processed authenticated user face image h j Input into the image fusion convolution layer α1;
[0046] In the image fusion convolutional layer α1, the preprocessed authenticated user face image h j is subjected to a convolutional operation to obtain the authenticated user face image feature map β1. The authenticated user face image feature map β1 and the natural host image are subjected to image fusion to obtain the fused authenticated face image γ1;
[0047] In the image fusion convolutional layer α d where d = 2, 3, …, k, the authenticated user face image feature map β d-1 is subjected to a convolutional operation to obtain the authenticated user face image feature map β d . The authenticated user face image feature map β d and the fused authenticated face image γ d-1 are subjected to image fusion to obtain a new fused authenticated face image γ d ;
[0048] The fused authenticated face image γ k is obtained in the image fusion convolutional layer α k ;
[0049] In the watermark addition layer, q non-overlapping small blocks of b*b pixels are created to obtain q image small blocks; the hash value c j of the authenticated user personal information is evenly written into the q image small blocks using the LSB technique to obtain q watermarked image small blocks; the q watermarked image small blocks are combined into a "hui" shape to obtain a digital watermark image, the inner length and width of the digital watermark image are the length and width of the fused authenticated face image; the fused authenticated face image γ k and the digital watermark image are spliced to obtain the encrypted authenticated user face image pm j .
[0050] An authentication system for a management cabinet, comprising:
[0051] An image acquisition module, configured to acquire m frames of consecutive face images to be authenticated, combine the m frames of consecutive face images to be authenticated into a face image set to be authenticated; acquire the authentication time at the current moment to obtain a time stamp to be authenticated;
[0052] A face liveness detection module, configured to deploy a face liveness detection model, extract features from the face image set to be authenticated to obtain dynamic features of the face image to be authenticated; perform liveness detection based on the dynamic features of the face image to be authenticated to obtain a liveness detection result to be authenticated;
[0053] The face comparison module extracts key features from the face image set to be authenticated, obtains the static descriptor of the face image to be authenticated, compares the static descriptor of the face image to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtains the comparison result of the face to be authenticated; if the comparison result of the face to be authenticated is successful, it prompts that the authentication is passed and the door of the management cabinet is opened; otherwise, it prompts that the authentication is failed;
[0054] The management cabinet user management module is used to obtain the facial image of the authenticated user, the user code of the authenticated user and the corresponding personal information; the public library of the management cabinet facial authentication comparison library performs feature extraction from the facial image of the authenticated user to obtain the facial feature descriptor of the authenticated user; all the facial feature descriptors of the authenticated user and their corresponding authenticated user codes are combined to obtain; the private library of the management cabinet facial authentication comparison library performs image steganographic encryption from the facial image of the authenticated user and their corresponding personal information to obtain the encrypted facial image of the authenticated user; all the encrypted authenticated user facial images of the authenticated user and their corresponding authenticated user codes are combined to obtain;
[0055] The image encryption module is used to deploy an image steganographic encryption model, and is used to perform image steganographic encryption on the authenticated user's face image and its corresponding personal information to obtain an encrypted authenticated user's face image.
[0056] The present invention has the following advantages:
[0057] 1. The present invention can effectively prevent the use of non-real face images such as photos or videos for authentication through face liveness detection, thereby improving the security of authentication; by extracting the optical flow features and local features of the face image set to be authenticated and combining the optical flow features with the local features, it can capture the changes of the face at different time points, thereby enhancing the accuracy of authentication; after the features of the face image set to be authenticated are extracted as static descriptors, the speed of comparing with the face feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet can be accelerated, thereby improving the authentication efficiency; through the use of the public library of the face authentication comparison library of the management cabinet and the private library of the face authentication comparison library of the management cabinet, the face feature descriptors and images of authenticated users can be flexibly managed according to actual needs, decentralized storage improves security, and protects the personal information of authenticated users.
[0058] 2. The present invention can make the image clearer and reduce noise by preprocessing the face image to be authenticated, which is beneficial to subsequent feature extraction and recognition; the dynamic optical flow features of the face image to be authenticated are obtained by using the optical flow method and sparse feature extraction, which is helpful to distinguish between static images and real living bodies; at the same time, local feature extraction and cascading can capture local information of the face image, thereby increasing the accuracy and robustness of recognition; cascading dynamic optical flow features and local features is helpful to comprehensively utilize a variety of feature information and improve the ability to detect living bodies.
[0059] 3. The present invention uses the SURF algorithm to perform feature extraction on the face images to be authenticated in the face image set, so as to extract the key feature points of the image, which helps to accurately describe the static features of the face images to be authenticated; performing maximum pooling on the static descriptors of all face images to be authenticated helps to screen the most representative feature vectors, thereby improving the efficiency and accuracy of the subsequent comparison process.
[0060] 4. The present invention can fuse the feature information of the authenticated user face image into the natural host image by convolutionally fusing the authenticated user face image and the natural host image, so that the final encrypted authenticated user face image is consistent with the natural host image in appearance, which can prevent unauthorized visitors from obtaining the user's facial information and improve the confidentiality and security of user data; adding a digital watermark to the outer circle of the fused authenticated user face image can improve the security and anti-counterfeiting ability of the image. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 The present invention is a flowchart of an authentication method for managing a cabinet adopted in an embodiment of the present invention.
[0062] Figure 2 This is a schematic diagram of the structure of an authentication system for managing cabinets adopted in an embodiment of the present invention. DETAILED DESCRIPTION
[0063] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0064] Example 1
[0065] An authentication method for managing cabinets, see Figure 1 As shown, including:
[0066] S1. Obtain m frames of continuous face images to be authenticated o i , i = 1, 2, ..., m; m frames of continuous face images to be authenticated o iCombine into a face image set to be authenticated; obtain the authentication time at the current moment and obtain the timestamp to be authenticated;
[0067] S2. Based on the face liveness detection model, feature extraction is performed on the face image set to be authenticated to obtain dynamic features of the face image to be authenticated; liveness detection is performed according to the dynamic features of the face image to be authenticated to obtain the liveness detection result to be authenticated; if the liveness detection result to be authenticated is liveness, proceed to step S3; otherwise, the authentication is failed;
[0068] The face liveness detection model includes a data preprocessing layer, m feature extraction layers, a feature cascade layer, a face liveness comparison layer, and a result output layer;
[0069] The data preprocessing layer is used to convert the face image to be authenticated into i Perform preprocessing to obtain the preprocessed face image x to be authenticated i ;
[0070] m feature extraction layers are used to extract any preprocessed face image x to be authenticated. i Do the following:
[0071] Preprocess the face image x to be authenticated i Perform optical flow feature extraction to obtain the optical flow vector field t of the face image to be authenticated of the preprocessed face image xi i ; The optical flow vector field t of the face image to be authenticated i Perform sparse feature extraction to obtain the dynamic optical flow feature w of the face image to be authenticated i ; The optical flow method can capture changes in facial expressions and tiny head movements. These motion information can be used to distinguish between living and non-living objects and improve the accuracy of liveness detection; sparse feature extraction can effectively reduce redundant information in the optical flow vector field and retain only the key features that are meaningful to the recognition task, thereby improving the expression efficiency of features and recognition accuracy;
[0072] Preprocess the face image x to be authenticated i Segmentation is performed to obtain 2*2 segmented face images x to be authenticated i1 , x i2 , x i3 , x i4 ; Segment the face image x to be authenticated i1 , x i2 , x i3 , x i4 The LTP operator is used to extract features and obtain the local features y1, y2, y3, y4 of the face image to be authenticated; the local features y1, y2, y3, y4 of the face image to be authenticated are cascaded to obtain the local features Fi of the face image to be authenticated. y = [y1, y2, y3, y4];
[0073] The LTP operator method can capture the local texture information of the pre-processed face image to be authenticated, and has a certain robustness to changes in illumination and expression. In face recognition, it can better cope with pre-processed face images to be authenticated under different illumination conditions and different expressions.
[0074] The dynamic optical flow feature w of the face image to be authenticated i and the local features Fi of the face image to be authenticated y Perform feature cascading to obtain the dynamic feature F of the face image to be authenticated xi =[T i ,Fi y ];
[0075] Feature cascading can comprehensively utilize the advantages of the dynamic optical flow features and local features of the face image to be authenticated, effectively improving the model's discrimination ability and generalization performance;
[0076] The feature cascade layer is used to dynamically identify all face images to be authenticated. xi Combine to obtain a dynamic feature sequence of the face image to be authenticated, perform maximum pooling on the dynamic feature sequence of the face image to be authenticated, and obtain the dynamic features of the face image to be authenticated;
[0077] The result output layer is used to output the liveness detection result to be authenticated according to the dynamic features of the face image to be authenticated. If the liveness detection result to be authenticated is liveness, the process proceeds to step S3; otherwise, the authentication fails.
[0078] Specific steps for training face liveness detection model:
[0079] H1. Collect several live and non-live face image sets, each of which contains m consecutive face images, and the target value of each face image set is live or non-live; combine the face image sets marked with the target values to obtain a live face detection training set;
[0080] H2. Input the face liveness detection training set into the face liveness detection initial model, train with the target value as the target, and obtain the face liveness detection accuracy. If the face liveness detection accuracy is within the error standard range, output the trained face liveness detection initial model, that is, the face liveness detection model; otherwise, continue to train the model using the face liveness detection training set; the error standard range is set by professionals based on actual experience;
[0081] S3. Extract key features from the face image set to be authenticated, obtain static descriptors of the face images to be authenticated, compare the static descriptors of the face images to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtain the comparison results of the faces to be authenticated; if the comparison result of the face to be authenticated is successful, the authentication is passed and the door of the management cabinet is opened; otherwise, the authentication is failed;
[0082] S31. The face image to be authenticated in the authentication face image set o i Perform SURF algorithm feature extraction to obtain the static descriptor z of the face image to be authenticated oi i ;
[0083] S32. Static descriptors z of all face images to be authenticated i Perform maximum pooling to obtain the static descriptor of the face image to be authenticated;
[0084] S33. The static descriptor of the face image to be authenticated is compared with the facial feature descriptors of authenticated users in the public library of the management cabinet face authentication comparison library one by one to obtain the similarity of the face comparison to be authenticated; if the similarity of the face comparison to be authenticated reaches the set threshold, the comparison result of the face to be authenticated is a successful comparison, prompting that the authentication is passed and the door of the management cabinet is opened; otherwise, the comparison result of the face to be authenticated is a failed comparison, prompting that the authentication is not passed; the threshold is set by professionals based on actual experience;
[0085] The SURF algorithm is used for feature extraction, which can quickly and effectively capture the key features of the face image to be authenticated. At the same time, the maximum pooling is used to further reduce the dimension of the static descriptor of the face image to be authenticated, thereby improving the efficiency of feature matching. By comparing the static descriptor of the face image to be authenticated with the facial feature descriptor of the authenticated user one by one, it can accurately determine whether the face to be authenticated belongs to the authenticated user, thereby providing a reliable authentication result.
[0086] Specific steps for building a management cabinet face authentication comparison library:
[0087] Obtain the face images of n authenticated users, the user codes of the authenticated users and the corresponding personal information, and obtain the face image p of the authenticated user j , where j = 1, 2, ..., n; n is the number of authenticated users in the management cabinet;
[0088] The authenticated user face image pj and its corresponding personal information are encrypted by image steganography to obtain the encrypted authenticated user face image pm j ;
[0089] The authenticated user's face image p jPerform SURF algorithm feature extraction to obtain the authenticated user's face feature descriptor pt j ;
[0090] PM all encrypted authenticated user face images j and its corresponding authenticated user code storage, obtain the face authentication comparison private library of the management cabinet, and upload the face authentication comparison private library of the management cabinet to the blockchain for storage;
[0091] All authenticated user face feature descriptors pt j and its corresponding authenticated user code storage, and obtain the management cabinet face authentication comparison library public library, and the management cabinet face authentication comparison library public library is stored locally;
[0092] The user code of the same authenticated user in the public database of the management cabinet face authentication comparison database and the private database of the management cabinet face authentication comparison database is the same;
[0093] The public database of the management cabinet face authentication comparison database and the private database of the management cabinet face authentication comparison database are combined to obtain the management cabinet face authentication comparison database;
[0094] The facial image of the authenticated user is obtained and steganographically encrypted. At the same time, the facial feature descriptor is extracted and stored in the private library of the management cabinet face authentication comparison library and the public library of the management cabinet face authentication comparison library respectively. The private library of the management cabinet face authentication comparison library is uploaded to the blockchain for safe storage, ensuring the security of user privacy. The user image and personal information are protected by encryption. The feature descriptors stored in the public library provide efficient matching and comparison functions for face authentication. The integration of the public and private libraries makes the management cabinet face authentication comparison library more complete, further improving security and authentication efficiency.
[0095] The specific steps of image steganography encryption include:
[0096] The authenticated user's face image p j Perform preprocessing to obtain the preprocessed authenticated user face image h j ; The authenticated user face image p j The corresponding personal information is preprocessed to obtain the preprocessed authenticated user personal information g j ;
[0097] Pre-process the authenticated user's personal information g j Perform hash calculation, which is completed by the hash function, and obtain the hash value c of the authenticated user's personal information j ;
[0098] The preprocessed authenticated user face image h j and the authenticated user's personal information hash value c jInput into the image steganography encryption model to obtain the encrypted authenticated user face image pm j ;
[0099] The image steganography encryption model includes image input layer, image convolution fusion layer, watermarking layer, and image output layer;
[0100] The image input layer is used to preprocess the authenticated user face image h j and any natural host image in the natural host image set are input into the image convolution fusion layer; the natural host image set includes a plurality of natural host images, and the plurality of natural host images are obtained by preprocessing a plurality of natural information images;
[0101] The image fusion convolution layer includes k convolution layers α u , u=1, 2, …, k; the image fusion convolution layer is used to pre-process the authenticated user face image h j Perform image convolution fusion with the natural host image to obtain the fused authenticated face image γ k ;
[0102] Adding a watermark layer to fuse authenticated face images k Add a digital watermark to the outer circle to get the encrypted authenticated user face image pm j ;
[0103] The image output layer is used to output the encrypted authenticated user face image pm j ;
[0104] The specific steps of the image steganography encryption model to encrypt an image include:
[0105] In the image input layer, any natural host image in the natural host image set is selected and input into the image fusion convolution layer α1, and the pre-processed authenticated user face image h j Input into the image fusion convolution layer α1;
[0106] In the image fusion convolution layer α1, the pre-processed authenticated user face image h j Perform a convolution operation to obtain a feature map β1 of the authenticated user face image, and perform image fusion on the feature map β1 of the authenticated user face image and the natural host image to obtain a fused authenticated face image γ1;
[0107] In the image fusion convolution layer α d , d = 2, 3, ..., k, for the authenticated user face image feature map β d-1 Perform convolution operation to obtain the authenticated user face image feature map β d , the authenticated user face image feature map β d and fusion of authenticated face images γd-1 Perform image fusion to obtain a new fused authenticated face image γ d ;
[0108] In the image fusion convolutional layer α k obtain the fused authenticated face image γ k ;
[0109] By embedding the pre - processed authenticated user face image into the natural host image, the image storage data becomes more concealed and is not easily detected and intercepted; using the convolutional layer to continuously perform convolutional fusion on the authenticated user face image feature map and the natural host image can not only retain the feature information of the pre - processed authenticated user face image but also hide the feature information of the pre - processed authenticated user face image;
[0110] In the watermark addition layer, create q non - overlapping small blocks each with b*b pixels to obtain q image small blocks; use the LSB technology to evenly write the hash value c of the authenticated user personal information j into the q image small blocks to obtain q watermark image small blocks; combine the q watermark image small blocks into a "hui" - shaped pattern to obtain a digital watermark image, where the inner length and width of the digital watermark image are the same as those of the fused authenticated face image; splice the fused authenticated face image γ k and the digital watermark image to obtain the encrypted authenticated user face image pm j ;
[0111] By using the LSB technology to embed the hash value of the authenticated user personal information into the least significant bits of the pixels of the q image small blocks to keep the visual perception of the image small blocks unchanged, and combining the q watermark image small blocks into a digital watermark image, the uniqueness and anti - tampering property of the encrypted authenticated user face image are improved.
[0112] Embodiment 2
[0113] An authentication system for a management cabinet, as shown in Figure 2 includes:
[0114] An image acquisition module, used to obtain m consecutive face images to be authenticated, combine the m consecutive face images to be authenticated into a set of face images to be authenticated; obtain the authentication time at the current moment to obtain a timestamp to be authenticated;
[0115] A face liveness detection module, used to deploy a face liveness detection model, extract features from the set of face images to be authenticated to obtain dynamic features of the face images to be authenticated; perform liveness detection based on the dynamic features of the face images to be authenticated to obtain a liveness detection result to be authenticated;
[0116] The face comparison module extracts key features from the face image set to be authenticated, obtains the static descriptor of the face image to be authenticated, compares the static descriptor of the face image to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtains the comparison result of the face to be authenticated; if the comparison result of the face to be authenticated is successful, it prompts that the authentication is passed and the door of the management cabinet is opened; otherwise, it prompts that the authentication is failed;
[0117] The management cabinet user management module is used to obtain the facial image of the authenticated user, the user code of the authenticated user and the corresponding personal information; the public library of the management cabinet facial authentication comparison library performs feature extraction from the facial image of the authenticated user to obtain the facial feature descriptor of the authenticated user; all the facial feature descriptors of the authenticated user and their corresponding authenticated user codes are combined to obtain; the private library of the management cabinet facial authentication comparison library performs image steganographic encryption from the facial image of the authenticated user and their corresponding personal information to obtain the encrypted facial image of the authenticated user; all the encrypted authenticated user facial images of the authenticated user and their corresponding authenticated user codes are combined to obtain;
[0118] The image encryption module is used to deploy an image steganographic encryption model, and is used to perform image steganographic encryption on the authenticated user's face image and its corresponding personal information to obtain an encrypted authenticated user's face image.
[0119] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.
Claims
1. An authentication method for a management cabinet, It is characterized in that include: S1. Obtain m frames of continuous face images to be authenticated o i , i=1,2,…,m; Take m frames of continuous face images to be authenticated o i Combine into a face image set to be authenticated; obtain the authentication time at the current moment and obtain the timestamp to be authenticated; S2. Based on the face liveness detection model, feature extraction is performed on the face image set to be authenticated to obtain dynamic features of the face image to be authenticated; Perform liveness detection based on the dynamic features of the face image to be authenticated, and obtain the liveness detection result to be authenticated; if the liveness detection result to be authenticated is liveness, proceed to step S3; otherwise, it is prompted that the authentication fails; S3. Extract key features from the face image set to be authenticated, obtain static descriptors of the face images to be authenticated, compare the static descriptors of the face images to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtain the comparison results of the faces to be authenticated; if the comparison result of the face to be authenticated is successful, the authentication is passed and the door of the management cabinet is opened; otherwise, the authentication is failed; Obtain the authenticated user's face image, the authenticated user's user code and corresponding personal information; The public database of the face authentication comparison library of the management cabinet performs feature extraction from the face image of the authenticated user to obtain the face feature descriptor of the authenticated user; all the face feature descriptors of the authenticated user and their corresponding authenticated user codes are combined to obtain; the private database of the face authentication comparison library of the management cabinet performs image steganographic encryption from the face image of the authenticated user and its corresponding personal information to obtain the encrypted face image of the authenticated user; all the encrypted authenticated user face images of the authenticated user and their corresponding authenticated user codes are combined to obtain; The public database of the management cabinet face authentication comparison database and the private database of the management cabinet face authentication comparison database are combined to obtain the management cabinet face authentication comparison database; Specific steps for building a management cabinet face authentication comparison library: Obtain the face images of n authenticated users, the user codes of the authenticated users and the corresponding personal information, and obtain the face image p of the authenticated user j , where j = 1, 2, ..., n; n is the number of authenticated users in the management cabinet; The authenticated user's face image p j The corresponding personal information is encrypted by image steganography to obtain the encrypted authenticated user face image pm j ; The authenticated user's face image p j Perform SURF algorithm feature extraction to obtain the authenticated user's face feature descriptor pt j ; PM all encrypted authenticated user face images j and its corresponding authenticated user code storage, obtain the face authentication comparison private library of the management cabinet, and upload the face authentication comparison private library of the management cabinet to the blockchain for storage; All authenticated user face feature descriptors pt j and its corresponding authenticated user code storage, and obtain the management cabinet face authentication comparison library public library, and the management cabinet face authentication comparison library public library is stored locally; The user code of the same authenticated user in the public database of the management cabinet face authentication comparison database and the private database of the management cabinet face authentication comparison database is the same; The specific steps of image steganography encryption include: The authenticated user's face image p j Perform preprocessing to obtain the preprocessed authenticated user face image h j ; The authenticated user face image p j The corresponding personal information is preprocessed to obtain the preprocessed authenticated user personal information g j ; Pre-process the authenticated user's personal information g j Perform hash calculation to obtain the authenticated user's personal information hash value c j ; The preprocessed authenticated user face image h j and the authenticated user's personal information hash value c j Input into the image steganography encryption model to obtain the encrypted authenticated user face image pm j ; The image steganography encryption model includes image input layer, image convolution fusion layer, watermarking layer, and image output layer; The image input layer is used to preprocess the authenticated user face image h j and any natural host image in the natural host image set are input into the image convolution fusion layer; the natural host image set includes a plurality of natural host images, and the plurality of natural host images are obtained by preprocessing a plurality of natural information images; The image fusion convolution layer includes k convolution layers α u , u=1, 2, …, k; the image fusion convolution layer is used to pre-process the authenticated user face image h j Perform image convolution fusion with the natural host image to obtain the fused authenticated face image γ k ; Add a watermark layer to fuse the authenticated face image γ k Add a digital watermark to the outer circle to get the encrypted authenticated user face image pm j ; The image output layer is used to output the encrypted authenticated user face image pm j ; The specific steps of the image steganography encryption model to encrypt an image include: In the image input layer, any natural host image in the natural host image set is selected and input into the image fusion convolution layer α 1 In the process, the authenticated user face image h is preprocessed. j Input to the image fusion convolution layer α 1 middle; In the image fusion convolution layer α 1 In the preprocessing process, the authenticated user face image h j Perform convolution operation to obtain the authenticated user face image feature map β 1 , the authenticated user face image feature map β 1 Perform image fusion with the natural host image to obtain the fused authenticated face image γ 1 ; In the image fusion convolution layer α d , d = 2, 3, ..., k, for the authenticated user face image feature map β d-1 Perform convolution operation to obtain the authenticated user face image feature map β d , the authenticated user face image feature map β d and fusion of authenticated face images γ d-1 Perform image fusion to obtain a new fused authenticated face image γ d ; In the image fusion convolution layer α k The fused authenticated face image γ is obtained k ; In the watermark layer addition, q non-overlapping small blocks of b*b pixels are created to obtain q image small blocks; the hash value c of the authenticated user's personal information is evenly written into the q image small blocks by using the LSB technology to obtain q watermark image small blocks; the q watermark image small blocks are combined into a "hui" shape to obtain a digital watermark image, and the inner length and width of the digital watermark image are the length and width of the fused authenticated face image; the fused authenticated face image γ j and the digital watermark image are spliced to obtain the encrypted authenticated user face image pm k j . 2. According to claim 1, an authentication method for a management cabinet, It is characterized in that The face liveness detection model includes a data preprocessing layer, m feature extraction layers, a feature cascade layer, a face liveness comparison layer, and a result output layer; The data preprocessing layer is used to convert the face image to be authenticated into i Perform preprocessing to obtain the preprocessed face image x to be authenticated i ; m feature extraction layers are used to extract any preprocessed face image x to be authenticated. i Do the following: Preprocess the face image x to be authenticated i Perform optical flow feature extraction to obtain the pre-processed face image x to be authenticated i The optical flow vector field t of the face image to be authenticated i ; The optical flow vector field t of the face image to be authenticated i Perform sparse feature extraction to obtain the dynamic optical flow feature w of the face image to be authenticated i ; Preprocess the face image x to be authenticated i Perform segmentation to obtain 2*2 segmented face images x to be authenticated i1 , x i2 , x i3 , x i4 ; Segment the face image x to be authenticated i1 , x i2 , x i3 , x i4 The LTP operator is used to extract features and obtain the local features y of the face image to be authenticated. 1 ,y 2 ,y 3 ,y 4 ; Treat the local features y of the authenticated face image 1 ,y 2 ,y 3 ,y 4 Perform feature cascading to obtain the local features Fi of the face image to be authenticated y =[y 1 ,y 2 ,y 3 ,y 4 ]; The dynamic optical flow feature w of the face image to be authenticated i and the local features Fi of the face image to be authenticated y Perform feature cascading to obtain the dynamic feature F of the face image to be authenticated xi =[T i ,Fi y ]; The feature cascade layer is used to dynamically identify all face images to be authenticated. xi Combine to obtain a dynamic feature sequence of the face image to be authenticated, perform maximum pooling on the dynamic feature sequence of the face image to be authenticated, and obtain the dynamic features of the face image to be authenticated; The result output layer is used to output the liveness detection result to be authenticated according to the dynamic features of the face image to be authenticated. If the liveness detection result to be authenticated is liveness, go to step S3; otherwise, it prompts that the authentication fails.
3. According to claim 2, an authentication method for a management cabinet, It is characterized in that Specific steps for training face liveness detection model: H1. Collect several live and non-live face image sets, each of which contains m consecutive face images, and the target value of each face image set is live or non-live; combine the face image sets marked with the target values to obtain a live face detection training set; H2. Input the face liveness detection training set into the face liveness detection initial model, train it with the target value as the target, and obtain the face liveness detection accuracy. If the face liveness detection accuracy is within the error standard range, output the trained face liveness detection initial model, that is, the face liveness detection model; otherwise, continue model training using the face liveness detection training set.
4. According to claim 3, an authentication method for a management cabinet, It is characterized in that The specific steps of step S3 include: S31. The face image to be authenticated in the face image set o i Perform SURF algorithm feature extraction to obtain the face image to be authenticated o i The static descriptor z of the face image to be authenticated i ; S32. Static descriptors z of all face images to be authenticated i Perform maximum pooling to obtain the static descriptor of the face image to be authenticated; S33. The static descriptor of the face image to be authenticated is compared with the facial feature descriptors of the authenticated user in the public library of the management cabinet face authentication comparison library one by one to obtain the similarity of the face comparison to be authenticated; if the similarity of the face comparison to be authenticated reaches the set threshold, the comparison result of the face to be authenticated is a successful comparison, indicating that the authentication is passed, and the door of the management cabinet is opened; Otherwise, the comparison result of the face to be authenticated is comparison failure, indicating that the authentication has failed.
5. An authentication system for managing cabinets, It is characterized in that The system applies an authentication method for managing a cabinet as described in any one of claims 1 to 4, including: The image acquisition module is used to obtain m frames of continuous face images to be authenticated, and combine the m frames of continuous face images to be authenticated into a set of face images to be authenticated; obtain the authentication time at the current moment, and obtain the timestamp to be authenticated; The face liveness detection module is used to deploy the face liveness detection model, extract features from the face image set to be authenticated, and obtain dynamic features of the face image to be authenticated; perform liveness detection based on the dynamic features of the face image to be authenticated, and obtain liveness detection results to be authenticated; The face comparison module extracts key features from the face image set to be authenticated, obtains the static descriptor of the face image to be authenticated, compares the static descriptor of the face image to be authenticated with the facial feature descriptors of authenticated users in the public library of the face authentication comparison library of the management cabinet one by one, and obtains the comparison result of the face to be authenticated; if the comparison result of the face to be authenticated is successful, it prompts that the authentication is passed and the door of the management cabinet is opened; otherwise, it prompts that the authentication is failed; The management cabinet user management module is used to obtain the facial image of the authenticated user, the user code of the authenticated user and the corresponding personal information; the public library of the management cabinet facial authentication comparison library performs feature extraction from the facial image of the authenticated user to obtain the facial feature descriptor of the authenticated user; all the facial feature descriptors of the authenticated user and their corresponding authenticated user codes are combined to obtain; the private library of the management cabinet facial authentication comparison library performs image steganographic encryption from the facial image of the authenticated user and their corresponding personal information to obtain the encrypted facial image of the authenticated user; all the encrypted authenticated user facial images of the authenticated user and their corresponding authenticated user codes are combined to obtain; The image encryption module is used to deploy an image steganographic encryption model, and is used to perform image steganographic encryption on the authenticated user's face image and its corresponding personal information to obtain an encrypted authenticated user's face image.
Citation Information
Patent Citations
Face feature and dynamic attribute authentication method based on WeChat applet platform
CN111460416A
Vehicle-mounted system security authentication management method considering face fraud
CN114663938A
Face authentication method and system based on public block chain
CN117272269A
Non-interactive face living body detection and face verification method and system
CN117351578A
Improved face liveness detection using background / foreground motion analysis
US20230222842A1
