An identity recognition method based on sublingual region image

Abstract

The application provides an identity recognition method based on a sublingual area image, comprising a model training stage and an implementation discrimination stage; wherein the model training stage comprises the following steps: training a sublingual key area extraction model to extract a sublingual key area in a sublingual image by using the sublingual key area extraction model; performing multi-feature representation based on the sublingual key area; and training a discrimination model by using multi-feature representation training data; wherein the implementation discrimination stage comprises the following steps: acquiring a sublingual image to be discriminated by using a visual acquisition device; extracting a sublingual key area from the acquired sublingual key area by using the sublingual key area extraction model; performing multi-feature representation on the sublingual key area; and performing prediction and judgment by using the discrimination model. The application takes the sublingual image as a main object, combines artificial intelligence and feature extraction, and can be used in the field of high-precision personal identity information recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to an identity recognition method based on a sublingual region image. BACKGROUND

[0002] The sublingual region of a human body is located on the lower half of a tongue, and includes a tongue blade, a sublingual vein, a plica fissurae laryngealis, and the like, and covers and expresses rich identity discrimination information (for example, color of the sublingual vein, texture information of the tongue blade, and the like). Therefore, aiming at the sublingual region condition, the personal identity can be inferred based on artificial intelligence and computer vision technology. Because the position of the sublingual region is special, the rich recognition is difficult to be damaged and hurt, so the sublingual region image can be used as an important identifier for recognizing personal identity information, and has a broad application prospect in the field of personal identity information recognition and security. However, there is no method for accurately recognizing personal identity information by using the sublingual region technology at present. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides an identity recognition method based on a sublingual region image, which takes the sublingual image as a main object, and combines artificial intelligence and feature extraction and other advanced methods, and can be used in the field of high-precision personal identity information recognition and other security applications.

[0004] The above technical purpose of the present application is realized by the following technical scheme:

[0005] An identity recognition method based on a sublingual region image, comprising a model training stage and an implementation discrimination stage.

[0006] The model training stage comprises the following steps:

[0007] Step 1, training a sublingual key region extraction model

[0008] Step 2, using the sublingual key region extraction model extracting a sublingual key region in the sublingual image;

[0009] Step 3, performing multi-feature representation based on the sublingual key region;

[0010] Step 4, training a discrimination model using multi-feature representation training data

[0011] The implementation discrimination stage comprises the following steps:

[0012] Step a, using a visual acquisition device to acquire a sublingual image to be discriminated

[0013] Step b, using the sublingual key area extraction model From Obtain the sublingual key area

[0014] Step c, multi-feature representation is performed on the sublingual key area ;

[0015] Step d, using the discriminant model to make a prediction judgment.

[0016] The application is further provided: in step 1, (1) using a visual acquisition device to collect N sublingual images of C groups to form a training set (2) the training set is divided into a sublingual global image training set and a sublingual key area training set (3) a part of the sublingual key area training set is labeled to obtain a labeled set (4) the training set and the labeled set are input into the YOLOv5x framework for end-to-end training to obtain a sublingual key area extraction model

[0017] The application is further provided: in step 2, the is input into the sublingual key area extraction model to obtain a key area feature training set

[0018] The application is further provided: in step 3, for the sublingual key area training set, the following four features are extracted: color feature, texture feature, geometric feature and latent representation feature.

[0019] The application is further provided: when extracting the color feature, (1) for each image in the , three channel color histograms H R , H G , H B are extracted; (2) for each image in the , first-order color moments, second-order color moments and third-order color moments are extracted respectively:

[0020]

[0021]

[0022]

[0023] wherein N represents the number of pixels, M1，ch , M 2，ch , M 3，ch are the first, second and third order color moments on the ch channel, respectively, P i represents the i-th pixel; (3) stitch the color histogram and the first, second and third order color moments:

[0024] FC i = [H R , H G , H B ,...

[0025] M 1，R , M 1，G , M 1，B ,...

[0026] M 2，R , M 2，G , M 2，B ,...

[0027] M 3，R , M 3，G , M 3，B ];

[0028] When extracting the texture feature: for each image in , two features are extracted: LBP feature and Gabor feature;

[0029] (1) feature F LBP :

[0030]

[0031]

[0032]

[0033] where the δ(·,·) function is expressed as follows:

[0034]

[0035] (2) feature F gabor is obtained by defining the following Gabor filters:

[0036]

[0037] where θ = {0°, 22.5°, 45°, …, 180°},

[0038] merge the LBP feature and the Gabor feature to form the texture feature FT = [F LBP , F gabor ];

[0039] When the geometric features are extracted: the geometric features are composed of length (l), width (w), length-width ratio (lwr), area (A), center distance (ζ), circular area (α), circular area ratio (ρ), square area (Φ) and square area ratio (β) of the sublingual key area image; wherein,

[0040] l = nrow(B)

[0041] w = neol(N)

[0042]

[0043] A = l * w

[0044]

[0045] α = πζ 2

[0046]

[0047] Φ = 4ζ 4

[0048]

[0049] At this time, the geometric features FG = [l, w, lwr, A, ζ, α, ρ, Φ, β].

[0050] When the latent representation features are extracted: the sublingual key area image is input into a deep autoencoder

[0051] The final multi-feature representation is F i = [FC i , FT i , FG i , FL i ].

[0052] The application is further provided as: in step 4, first extract different features according to all images in the sublingual key area training set in step 3, prepare the sublingual key area feature training set Then select the base classifier Finally, according to the sublingual key area feature training set F tr and the base classifier train the discriminant model

[0053] The application is further provided as: in step c,

[0054] The application is further provided as: in step d,

[0055] The present application has the following advantages:

[0056] The present technology can directly utilize human sublingual images for human identity information recognition or as an auxiliary discrimination technology. The present technology automates the processes of sublingual key region acquisition, feature extraction, and discrimination analysis, and can achieve a relatively accurate discrimination effect. Since the present technology is an image analysis method for the sublingual region, it can take into account the rich key discrimination information contained in the sublingual region to ensure the effect of automated discrimination. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is a schematic diagram of the sublingual vein key region;

[0058] Figure 2 is a flowchart of the model training stage;

[0059] Figure 3 is a flowchart of the real-time discrimination stage. DETAILED DESCRIPTION

[0060] The technical solutions in the present application will be further described below in conjunction with the drawings and examples.

[0061] A sublingual region image-based identity recognition method, comprising:

[0062] Stage One, Model Training Stage:

[0063] Step 1, Training the Sublingual Key Region Extraction Model

[0064] (1) Using a visual acquisition device to collect N sublingual images of C groups to form a training set

[0065] (2) Dividing the training set into a sublingual global image training set and a sublingual key region training set

[0066] (3) Labeling part of the sublingual vein key region of the patient (as shown below) Figure 1 to obtain a labeled set

[0067] (4) Inputting the training set and the labeled set into the YOLOv5x framework for end-to-end training to obtain a sublingual key region extraction model

[0068] Step 2, (One of the Two-Step Method) Using the Model Extracting the sublingual key region in the sublingual image: taking Inputting the sublingual key region extraction model Obtaining the key region feature training set

[0069] Step 3, (two-step method two) multi-feature representation based on the sublingual key region: for the sublingual key region training set, the following four features are extracted:

[0070] Color feature: color feature is an important feature indicating sublingual vein discrimination information.

[0071] (1) For each image in , extract three channels of color histogram H R , H G , H B .

[0072] (2) For each image in , extract first-order color moment, second-order color moment, and third-order color moment respectively:

[0073]

[0074]

[0075]

[0076] N represents the number of pixels, M 1，ch is the first-order color moment, the second-order color moment, and the third-order color moment on the ch channel respectively, P i represents the i-th pixel.

[0077] (3) Concatenate the color histogram and the first-order, second-order, and third-order color moments:

[0078] FC i = [H R , H G , H B ,...

[0079] M 1，R , M 1，G , M 1，B ,...

[0080] M 2，R , M 2，G , M 2，B ,...

[0081] M 3，R , M 3，G , M 3，B ]

[0082] Texture feature: For each image in , two texture features are extracted:

[0083] (1) LBP feature F LBP :

[0084]

[0085] δ(T i，j-1 , T i，j ), δ(T i，j+1 , T i，j ), δ(T i+1，j-1 , T i，j ),

[0086] δ(T i+1，j , T i，j ), δ(T i+1，j+1 , T i，j )

[0087] where the δ(·, ·) function is expressed as follows:

[0088]

[0089] (2) Gabor feature: The Gabor feature is obtained by convolving the Gabor filter with the image of the sublingual key area F gabor :

[0090]

[0091] where θ = {0°, 22.5°, 45°,..., 180°},

[0092] Then the LBP feature and the Gabor feature are combined to form the texture feature FT:

[0093] FT = [F LBP， F gabor ]

[0094] Geometric feature: The geometric feature is composed of 9 kinds of features combined by the length (l), width (w), length-width ratio (lwr), area (A), center distance (ζ), circular area (α), circular area ratio (ρ), square area (Φ), and square area ratio (β) of the image of the sublingual key area. The 9 kinds of features are described as follows:

[0095] l = nrow(B)

[0096] w = ncol(N)

[0097]

[0098] A = l x w

[0099]

[0100] a = p z 2

[0101]

[0102] F = 4 z 4

[0103]

[0104] Finally, the geometric feature FG = [l, w, lwr, A, z, a, p, F, b].

[0105] Latent representation features: extract the image of the sublingual key area Input depth autoencoder

[0106] Perform multi-feature representation F i = [FC i , FT i , FG i , FL i ];

[0107] Step 4, train the discriminant model using the multi-feature representation training data

[0108] (1) Extract different features from all images in the sublingual key area training set according to step 3 Prepare the sublingual key area feature training set F

[0109] : tr

[0110]

[0111] (2) Select a base classifier

[0112] (3) Train the discriminant model according to the sublingual key area feature training set F tr and the base classifier

[0113]

[0114] Phase two, real-time discrimination phase:

[0115] Step a, use a visual acquisition device to collect the sublingual image to be discriminated

[0116] ​​Step b, using the sublingual key region extraction model From Acquire the sublingual key region

[0117] Step c, multi-feature representation of the sublingual key region

[0118]

[0119] Step d, using the discriminant model Make a prediction:

[0120]

[0121] The present technology can directly use human sublingual images for human identity information recognition or as an auxiliary discrimination technology. The present technology automates the processes of sublingual key region acquisition, feature extraction, discriminant analysis, etc. and can achieve a relatively accurate discrimination effect. Since the present technology is an image analysis method for the sublingual part, it can take into account the rich key discrimination information contained in the sublingual part, ensuring the effect of automated discrimination.

[0122] Finally, it should be pointed out that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered by the scope of the claims of the present application.​

Claims

1. A method for identity recognition based on sublingual region images, characterized in that: The method comprises a model training stage and an implementation discrimination stage; The model training stage comprises the following steps: Step 1, training a sublingual key region extraction model ; Step 2, extracting the sublingual key region using a sublingual key region extraction model extracting a sublingual key region in the sublingual image; Step 3, multi-feature representation based on the sublingual key region; Step 4, training a discriminative model using the multi-feature representation training data ; The implementation discrimination stage comprises the following steps: Step a, collecting the sublingual image to be discriminated using a visual acquisition device ; Step b, using sublingual key region extraction model From Acquiring sublingual key region ; Step c. Sublingual key area performing multi-feature representation; Step d, using a discriminant model making a predictive determination; In step 1, (1) a visual acquisition device is used to acquire sublingual images of one group to form a training set ; (2) the training set is divided into a sublingual global image training set and a sublingual key region training set ; (3) part of the sublingual key region training set is labeled to obtain a labeled set , ; (4) the training set and the labeled set are input into a YOLOv5x framework for end-to-end training to obtain a sublingual key region extraction model ;​ In step 2, the input sublingual key region extraction model , to obtain a key region feature training set ; In step 3, the following four features are extracted for the sublingual key region training set: color feature, texture feature, geometric feature and latent representation feature; When extracting the color features: (1) for each image in , three channel color histograms , , are extracted; (2) for each image in , first-order color moments, second-order color moments and third-order color moments are extracted respectively: wherein, representing the number of pixels, respectively, the first, second and third order chromatic moments on the channel, representing the i-th pixel; (3) the stitched color histogram and the first, second and third order chromatic moments: ； When extracting the texture features: for each image in two features are extracted: LBP features and Gabor features; (1) Feature : wherein The function is expressed as follows: ； (2) Feature is obtained by defining the following Gabor filter: wherein , ; combining the lbp features and the gabor features to form texture features ; When extracting the geometric features: the geometric features are composed of the length , width , length-width ratio , area , center distance , circular area , circular area ratio , square area , and square area ratio of the sublingual key region image; wherein, At this time, the geometric feature ; When extracting the features of the submarine, the image of the key area of the tongue is input into a deep autoencoder ; The final multi-feature representation is ; In step 4, the training set for the key sublingual regions is first established based on the data from step 3. Extract different features from all images to prepare a training set of features for the key sublingual region. Then select the base classifier. Finally, a training set was developed based on the features of key sublingual regions. With base classifier Training the discriminant model .

2. The method of claim 1, wherein the method comprises: In step c, . 3.The method of claim 2, wherein: In step d, .

Images