Palm vein recognition method and device, medium and vehicle
By calculating and updating gradient histogram vectors for units in palm vein images, the problem of recognition errors caused by inconspicuous features in palm vein recognition is solved, and more accurate user recognition is achieved.
Patent Information
- Application Number
- CN202410744495.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-07
- Publication Date
- 2025-04-04
AI Technical Summary
Existing palm vein recognition technology can easily lead to identification errors when the user's palm vein characteristics are not obvious.
By acquiring the palm vein image, determining the cells in the image, and calculating the gradient histogram vector and weight for each cell, updating the gradient histogram vector, and finally user identification is performed based on the updated vector.
Improves the accuracy of palm vein recognition and ensures accurate identification when user characteristics are not obvious.
Smart Images
Figure CN119763159A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer technology, and in particular, to a palm vein recognition method, device, medium and vehicle. Background Art
[0002] The geometric shape of palm veins is a unique biometric feature that can be used for identity recognition and is applied to scenarios such as opening vehicle doors, safe doors, and anti-theft doors. In palm vein recognition technology, the current user is identified by matching the currently detected palm vein image with the palm vein image in the database. Whether the user's palm vein features are obvious directly affects the matching results. If the user's palm vein features are not obvious, there is a possibility of misidentification. Summary of the invention
[0003] The purpose of the present disclosure is to provide a palm vein recognition method, device, medium and vehicle that can accurately identify a user, so as to accurately perform user recognition.
[0004] In order to achieve the above object, the present disclosure provides a palm vein recognition method, the method comprising:
[0005] Get the palm vein image of the current user;
[0006] determining a unit in the palm vein image;
[0007] Determine a gradient histogram vector of each unit, wherein the gradient histogram vector is a directional gradient histogram vector;
[0008] For each unit, determine a weight of the unit, wherein the weight of the unit is related to an accumulated value of a gradient magnitude corresponding to at least one gradient direction interval in a gradient histogram vector of the unit;
[0009] For each unit, updating the gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight;
[0010] The current user is identified according to the updated gradient histogram vector.
[0011] Optionally, for each unit, determining the weight of the unit includes:
[0012] For each unit, the weight of the unit is determined according to the sum of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0013] Optionally, determining the weight of each unit includes:
[0014] For each unit, the weight of the unit is determined according to the maximum value of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0015] Optionally, every predetermined number of adjacent units in the palm vein image constitutes a block, the weight of each block in the palm vein image is equal, and for each block, the weight of the block is the sum of the weights of each unit in the block.
[0016] Optionally, for each unit, updating the gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight includes:
[0017] For each unit, the product of the gradient histogram vector of the unit and the weight of the unit is used as the updated gradient histogram vector of the unit.
[0018] Optionally, for each unit, updating the gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight includes:
[0019] For each unit, a weighted sum of the gradient histogram vector of the unit and the gradient histogram vectors of units adjacent to the unit is used as the updated gradient histogram vector of the unit.
[0020] Optionally, identifying the current user according to the updated gradient histogram vector includes:
[0021] For each unit in the palm vein image, matching the updated gradient histogram vector with a predetermined feature vector to obtain a matching result of the unit;
[0022] The current user is identified according to the matching result.
[0023] Optionally, identifying the current user according to the matching result includes:
[0024] Determine a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or determine a unit whose weight ranking is higher than a predetermined ranking threshold as the high-weight unit;
[0025] If the number of successfully matched units is greater than a predetermined number threshold, and all units determined to be high-weighted are successfully matched, it is determined that the current user has been identified successfully.
[0026] Optionally, the method further comprises: extracting features from the palm vein image according to a local binary pattern histogram (LBPH) method; identifying the current user according to the extracted features;
[0027] The determining of the gradient histogram vector of each unit includes: determining the gradient histogram vector of each unit when the feature extracted according to the LBPH method fails to identify the current user.
[0028] Optionally, identifying the current user according to the extracted features includes:
[0029] According to the chi-square test method, the extracted features are matched with the predetermined features;
[0030] When the matching result is less than the first threshold, determining that the current user has been identified;
[0031] When the matching result is greater than the second threshold, determining that the current user identification fails;
[0032] When the matching result is greater than the first threshold and less than the second threshold, it is determined that the current user identification fails.
[0033] The present disclosure also provides a palm vein recognition device, the device comprising:
[0034] An acquisition module is used to acquire a palm vein image of the current user;
[0035] A first determination module, used for determining a unit in the palm vein image;
[0036] A second determination module is used to determine a gradient histogram vector of each unit, wherein the gradient histogram vector is a directional gradient histogram vector;
[0037] A third determination module is used to determine the weight of each unit, wherein the weight of the unit is related to the accumulated value of the gradient magnitude corresponding to at least one gradient direction interval in the gradient histogram vector of the unit;
[0038] An updating module, configured to update, for each unit, a gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight;
[0039] The first identification module is used to identify the current user according to the updated gradient histogram vector.
[0040] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, and the program instructions, when executed by a processor, implement the steps of the above method provided by the present disclosure.
[0041] The present disclosure also provides a palm vein recognition device, comprising:
[0042] processor;
[0043] a memory for storing instructions executable by the processor;
[0044] The processor is configured to implement the steps of the above method provided by the present disclosure.
[0045] The present disclosure also provides a vehicle, comprising the above-mentioned palm vein recognition device provided by the present disclosure.
[0046] Through the above technical solution, on the basis of extracting image features using the HOG method in the related art, a weight is set for each unit in the palm vein image, and the gradient histogram vector of each unit is updated according to these weights. Since the weight is related to the accumulated value of the gradient size corresponding to at least one gradient direction interval in the gradient histogram vector, the characteristic difference of the gradient is more obvious in the gradient histogram vector of each unit after the update, and the current user is identified according to the updated gradient histogram vector, and the identification result is more accurate.
[0047] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:
[0049] Figure 1 is a flowchart of a palm vein recognition method provided by an exemplary embodiment.
[0050] Figure 2 is a schematic diagram of a histogram of directional gradients of a unit provided by an exemplary embodiment.
[0051] Figure 3 It is a schematic diagram of the weights of each unit in a palm vein image provided by an exemplary embodiment.
[0052] Figure 4 is a flowchart of a palm vein recognition method provided by another exemplary embodiment.
[0053] Figure 5 is a block diagram of a palm vein recognition device provided by an exemplary embodiment. DETAILED DESCRIPTION
[0054] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.
[0055] Figure 1 FIG. 1 is a flow chart of a palm vein recognition method provided by an exemplary embodiment. Figure 1 As shown, the method includes steps S101 to S106.
[0056] Step S101 obtains the palm vein image of the current user. The palm of the current user may be scanned by an infrared scanner to obtain the palm vein image of the current user.
[0057] Step S102, determining cells in the palm vein image.
[0058] The palm vein image can be directly determined as a unit, or a region of interest (ROI) can be extracted from the palm vein image and the ROI can be determined as a unit. Alternatively, the palm vein image can be divided into multiple units. Alternatively, the ROI in the palm vein image can be divided into multiple units. A unit is a smaller pixel connected area.
[0059] Step S103, determining the gradient histogram vector of each unit, where the gradient histogram vector is a directional gradient histogram vector.
[0060] The gradient histogram in this scheme is the Histogram of Oriented Gradient (HOG), which divides the gradient direction into multiple intervals evenly. In each unit, the gradient directions of all pixels in each direction interval are counted by histogram to obtain a multi-dimensional feature vector, that is, the gradient histogram vector of each unit. HOG is used to calculate the statistical value of the directional information of the local image gradient.
[0061] Specifically, the gradients along the x-axis and y-axis directions are calculated according to the following formula (1):
[0062]
[0063] Among them, G x (x,y) is the gradient along the x-axis at the position (x,y), G y (x,y) is the gradient along the y-axis at the position (x,y), I(x,y) is the pixel value of the two-dimensional image, 0≤x≤w, 0≤y≤h, w and h are the width and height of the image respectively.
[0064] The magnitude and direction of the gradient are calculated according to the following formula (2):
[0065]
[0066] Among them, G(x,y) is the gradient size at the position (x,y), and θ(x,y) is the gradient direction at the position (x,y). By calculating the gradient, not only can the edge contour, texture and other features of the object be captured, but the influence of lighting can also be further reduced.
[0067] Next, build HOG. Figure 2 FIG. 1 is a schematic diagram of a directional gradient histogram of a unit provided by an exemplary embodiment. Figure 2 In , the histogram is divided into 9 items (0-8), called 9 bins, corresponding to 9 gradient direction intervals. All gradient values of all pixels in a unit are accumulated into the bin interval corresponding to the angle of each gradient value, forming the following Figure 2 The histogram of oriented gradients for this unit is shown.
[0068] In the gradient of each pixel, the value of the gradient direction is [-180°, 180°]. For the gradient of 0-180°, 180° is divided into 9 regions (bins), that is, each bin corresponds to 20°. Therefore, if in this unit, a pixel has a gradient direction between 20°-40°, the bin corresponding to 1 on the histogram is voted, and its number is +1; similarly, if a pixel has a gradient direction between 100°-120°, the number of bins labeled 5 on the histogram is +1. And, according to the size of the calculated gradient, the number of votes for the bin is weighted. For example, the direction of a pixel's gradient is 30°, and the gradient size is 3. Then, a weight of 3 can be assigned to vote 1 on the bin labeled 1, that is, 3×1 is accumulated to the bin labeled 1. Similarly, for the part with a gradient direction of -180°-0, it is first added 180° and then calculated in the same way as the positive gradient direction. One unit corresponds to one gradient histogram vector.
[0069] Before extracting the above vector features, the ROI image can be corrected and preprocessed. The γ correction preprocessing method can be used. The specific steps are as follows:
[0070] 1. Normalization: Convert pixel values to real numbers between 0 and 1. The algorithm is: (I+0.5) / 256, where I is the image pixel.
[0071] 2. Pre-compensation: The pixel normalized data is used to calculate the corresponding value with 1 / γ as the exponent, where γ is the correction coefficient.
[0072] 3. Denormalization: Denormalize the pre-compensated real value to an integer value between 0 and 255. The specific algorithm is: 256f-0.5, where f is the pre-compensated value.
[0073] Feature extraction after γ correction preprocessing can further improve the accuracy of feature extraction.
[0074] Step S104: for each unit, determine the weight of the unit, wherein the weight of the unit is related to the accumulated value of the gradient magnitude corresponding to at least one gradient direction interval in the gradient histogram vector of the unit.
[0075] Since palm vein recognition is to extract the texture of palm veins in the image, the texture features in the image are very important. The accumulated value of the gradient magnitude corresponding to a gradient direction interval (bin interval) is the sum of the gradient magnitudes corresponding to the bin interval in the gradient histogram vector, that is, Figure 2 The height of the column in the bin interval in . When extracting features according to the HOG method, in general, the larger the cumulative value of the gradient magnitudes of each bin in a unit is, the more important the unit is considered to be, and it is likely to be part of the texture. Therefore, the weight of a unit is determined according to the cumulative value of the gradient magnitude corresponding to at least one gradient direction interval in the gradient histogram vector of the unit.
[0076] Step S105 : for each unit, update the gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight.
[0077] The updated gradient histogram vector includes the factors of the original gradient magnitude accumulation value and the weight factors, which is equivalent to amplifying the feature difference of the gradient magnitude accumulation value, making the feature difference of the gradient magnitude accumulation value more obvious.
[0078] Step S106, identifying the current user according to the updated gradient histogram vector.
[0079] The updated gradient histogram vectors of each unit may be concatenated and matched with a predetermined vector feature, and whether the current user is a registered user may be identified based on the matching result.
[0080] Through the above technical solution, on the basis of extracting image features using the HOG method in the related art, a weight is set for each unit in the palm vein image, and the gradient histogram vector of each unit is updated according to these weights. Since the weight is related to the accumulated value of the gradient size corresponding to at least one gradient direction interval in the gradient histogram vector, the characteristic difference of the gradient is more obvious in the gradient histogram vector of each unit after the update, and the current user is identified according to the updated gradient histogram vector, and the identification result is more accurate.
[0081] In yet another embodiment, for each unit, determining the weight of the unit includes: for each unit, determining the weight of the unit according to the sum of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0082] The gradient direction interval is the bin interval mentioned above. Figure 2In the figure, the height of the bar graph corresponding to each bin interval is the cumulative value of the gradient size corresponding to the bin interval. The larger the cumulative value of the gradient size, the more obvious the change of the gradient in the bin interval. In a unit, the larger the sum of the cumulative values of the gradient sizes of all bin intervals, the more obvious the gradient change in the unit, and the more important the unit is in palm vein recognition. Therefore, in one embodiment, the weight of each unit can be determined according to the sum of the cumulative values of the gradient sizes corresponding to all bin intervals in each unit. The larger the sum of the cumulative values of the gradient sizes of all bin intervals in a unit, the greater the weight of the unit.
[0083] In this embodiment, the weight of the unit is determined according to the sum of the accumulated values of the gradient magnitudes of each interval, the calculation is simple, and the reliability of identifying the palm vein after weighting is high.
[0084] In yet another embodiment, for each unit, determining the weight of the unit includes: for each unit, determining the weight of the unit according to the maximum value of the accumulated values of gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0085] In the gradient histogram vector of a unit, as mentioned above, the larger the gradient magnitude cumulative value is, the more obvious the gradient change in the bin interval is. The gradient magnitude cumulative value of the bin interval with the largest gradient magnitude cumulative value can be used to characterize whether the gradient change of the unit is obvious. The larger the gradient magnitude cumulative value of the bin interval with the largest gradient magnitude cumulative value in a unit is, the greater the weight of the unit is.
[0086] For example, the palm vein image is divided into n units. In the gradient histogram vector of the i-th unit, the maximum value of the accumulated values of the gradient magnitudes corresponding to all gradient direction intervals is Tmax_i. Then the weight Ti of the unit (i-th unit) can be obtained by the following formula:
[0087] Ti=Tmax_i / (Tmax_1+Tmax_2+……+Tmax_i+……+Tmax_n)
[0088] In another embodiment, every predetermined number of adjacent units in the palm vein image constitutes a block, and the weight of each block in the palm vein image is equal. For each block, the weight of the block is the sum of the weights of the units in the block.
[0089] Figure 3 FIG. 1 is a schematic diagram of the weights of each unit in a palm vein image provided by an exemplary embodiment. Figure 3As shown, a small grid represents a unit, and there are 4×4=16 units in total. Among them, 2×2=4 units form a block (indicated by the thick line), and there are four blocks in total. The number in each small grid represents the weight of the unit. For example, the weights of the four units with row number 1 and column numbers 1-4 are 0.2, 0.3, 0.1, and 0.5 respectively. In the block at the upper left position, the weights of the four units are 0.2, 0.3, 0.1, and 0.4 respectively. The sum of the weights of the four units in each block (the weight of each block) is 1.
[0090] For example, the reference weight of each unit in a block can be calculated first, and then the reference weights of each unit in the block can be normalized to obtain the weights of each unit, so that the sum of the weights of each unit in each block is 1. That is, the normalization method is used to make the weights of each unit in each block between (0, 1), and the sum of the weights of each unit in each block is 1.
[0091] The reference weight of each unit can be obtained according to the sum of the accumulated values of the gradient magnitudes of each interval of each unit. For example, the sum of the accumulated values of the gradient magnitudes of each interval of a unit is used as the reference weight of the unit.
[0092] In this embodiment, the sum of the weights of each unit in each block is equal, so that the difference between the updated gradient histogram vector and the original vector will not be too large between blocks. While increasing the weight of the unit to highlight the difference in gradient features, the reliability of recognition is guaranteed.
[0093] In yet another embodiment, for each unit, the gradient histogram vector of the unit is updated according to the gradient histogram vector of the unit and the determined weight, including: for each unit, taking the product of the gradient histogram vector of the unit and the weight of the unit as the updated gradient histogram vector of the unit.
[0094] That is, in each unit, the accumulated value of the gradient size corresponding to each bin interval is directly multiplied by the weight of the unit to obtain the updated gradient histogram. Through this simple multiplication, the gradient histogram vector is quickly updated, the difference of gradient features is increased, and the calculation speed is fast.
[0095] In yet another embodiment, for each unit, the gradient histogram vector of the unit is updated according to the gradient histogram vector of the unit and the determined weight, including: for each unit, taking the weighted sum of the gradient histogram vector of the unit and the gradient histogram vectors of units adjacent to the unit as the updated gradient histogram vector of the unit.
[0096] The units adjacent to a unit may include four adjacent units on the top, bottom, left and right. When a unit itself is at the edge of the image, the number of its adjacent units may also be less than four.
[0097] like Figure 3 As shown in the figure, the weight of the cell with row number 2 and column number 2 is 0.4, and the weights of the four adjacent cells above, below, left, and right are 0.3, 0.3, 0.1, and 0.2 respectively. The weight of the cell with row number 4 and column number 4 is 0.4, and the weights of the two adjacent cells above and left are 0.1 and 0.1 respectively.
[0098] The updated gradient histogram vector of the cell with row number 2 and column number 2 can be expressed as: A = 0.3A 上 +0.3A 下 +0.1A 左 +0.2A 右 +0.4A0, where A is the updated gradient histogram vector, A0 is the gradient histogram vector before the update, and A 上 , A 下 , A 左 , A 右 They are the gradient histogram vectors of the four adjacent units above, below, left and right (before the gradient histogram vectors of these four units are updated).
[0099] In this embodiment, when updating the gradient histogram vector of a unit, the weight of the unit itself and the weights of adjacent units are considered, so that the correlation between adjacent units is retained and the reliability of the recognition result is guaranteed.
[0100] In yet another embodiment, identifying the current user according to the updated gradient histogram vector includes:
[0101] For each unit in the palm vein image, the updated gradient histogram vector is matched with a predetermined feature vector to obtain a matching result of the unit; and the current user is identified according to the matching result.
[0102] For example, all units in the ROI may be matched one by one, and each unit has a matching result. The matching result may include success and failure. The chi-square test method or the Euclidean distance matching method may be applied for matching. The predetermined feature vector may be a gradient histogram vector obtained from the palm vein image of the user during registration.
[0103] In this embodiment, each unit is matched one by one, and when the matching results of each unit are used to perform user identification, the identification result is more accurate.
[0104] In yet another embodiment, identifying the current user according to the matching result includes:
[0105] Determine a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or determine a unit whose weight ranking is higher than a predetermined ranking threshold as a high-weight unit;
[0106] If the number of successfully matched units is greater than a predetermined number threshold, and all units with high weights are determined to be successfully matched, it is determined that the current user has been identified successfully.
[0107] On the basis that the number of successfully matched units is greater than a predetermined number threshold, in order to determine that the current user is an authorized user (identification passed), it is also necessary to meet the condition that all high-weighted units are successfully matched.
[0108] High-weighted units are units with higher weights. Due to the high weights, the gradient features in these units are obvious, and they should play an important role in palm vein recognition. If these high-weighted units are not matched successfully, it is considered that the ROI as a whole should not be matched successfully. This ensures the accuracy of the matching results.
[0109] In yet another embodiment, the method further comprises: determining a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or determining a unit whose weight ranking is higher than a predetermined ranking threshold as a high-weight unit.
[0110] The predetermined weight threshold and the predetermined sorting threshold can be obtained based on experiments. The units higher than the predetermined sorting threshold are the thresholds at the front in the sorting from large to small weights, that is, the units with the largest weights. In this embodiment, high-weight units can be quickly screened out based on this condition, and the matching results can be limited accordingly.
[0111] In yet another embodiment, the method may further include: extracting features from the palm vein image according to a local binary pattern histogram (LBPH) method; and identifying the current user according to the extracted features.
[0112] The step of determining the gradient histogram vector of each unit includes: determining the gradient histogram vector of each unit when the feature extracted according to the LBPH method fails to identify the current user.
[0113] That is, after acquiring the palm vein image of the current user, the LBPH method is first used to extract features from the palm vein image. If recognition based on the LBPH feature fails, HOG is used for recognition according to the method in the above embodiment.
[0114] Failure of identification based on LBPH features means that when matching based on LBPH features, it is not possible to clearly determine whether the current user is an authorized user. Identification based on LBPH features that the current user is an authorized user (identification passed) and identification that the current user is not an authorized user (identification failed) are both considered successful identification based on LBPH features.
[0115] Matching based on LBPH features can usually ensure the accuracy of recognition in most cases. In this embodiment, only in a few cases, when the result of matching based on LBPH features is inaccurate, the HOG method is added for secondary recognition to reduce the recognition error. Compared with directly using two feature extraction methods, in most cases, the computing power is reduced and unnecessary operations are avoided.
[0116] Since the overall picture of the infrared palm vein image is dark, the image denoising effect is not obvious in the time domain space, that is, if the image is directly filtered, it can be converted to the frequency domain space to distinguish the palm vein and noise in the frequency domain space. Before LBPH feature extraction, the ROI can be enhanced and preprocessed. The specific steps are as follows:
[0117] 1) Use Fourier transform to obtain the spectrum of the ROI image;
[0118] 2) Apply a Butterworth filter to low-pass filter the spectrum obtained in step 1). The parameters of the Butterworth low-pass filter can be obtained based on statistical experience. For example, the cutoff frequency is 30 and the order is 1. The characteristics of the Butterworth filter are that the transfer function is relatively smooth and continuously attenuated. Using this filter to filter can greatly reduce the blurring degree of the image edge while suppressing noise.
[0119] 3) Performing inverse Fourier transform on the filtered spectrum in step 2) to obtain an image enhanced in time domain and space.
[0120] After the above enhancement preprocessing, the image denoising effect is good.
[0121] When extracting LBPH features, the LBPH method of circular rotation invariant pattern can be applied. The LBPH features of circular rotation invariant pattern are extracted on the ROI, so that the features extracted from different people are more distinguishable.
[0122] For example, the circular rotation invariant pattern LBP algorithm is applied to obtain a texture map, which is divided into 8×8 blocks, and histogram statistics are performed on each texture map. Then, the histograms of the 8×8 blocks are spliced so that the spliced feature vector retains the palm vein features that can distinguish different people as much as possible. The 8×8 block parameters and the radius and neighborhood settings of the circular LBP can be obtained based on experiments. Applying histogram statistics instead of using the entire feature map can reduce the number of features and improve the operating efficiency during matching.
[0123] In yet another embodiment, identifying the current user according to the extracted features includes:
[0124] According to the chi-square test method, the extracted features are matched with the predetermined features;
[0125] When the matching result is less than the first threshold, it is determined that the current user has been identified;
[0126] When the matching result is greater than the second threshold, determining that the current user identification fails;
[0127] When the matching result is greater than the first threshold and less than the second threshold, it is determined that the current user identification fails.
[0128] That is, when the matching result according to the LBPH feature is less than the first threshold, it is considered that according to the LBPH feature, the palm vein image of the current user can well match the palm vein image of the authorized user, and the current user is the authorized user and the identification is passed.
[0129] When the matching result according to the LBPH feature is greater than the second threshold, it is considered that according to the LBPH feature, the palm vein image of the current user is significantly different from the palm vein image of the authorized user, the current user is not the authorized user, and the recognition fails. The second threshold is greater than the first threshold.
[0130] When the matching result is greater than the first threshold and less than the second threshold, it is considered that according to the LBPH feature, the palm vein image of the current user has a certain similarity with the palm vein image of the authorized user, but the similarity is not too high and is in an ambiguous state, that is, the recognition fails.
[0131] In this embodiment, whether the recognition is successful is determined by using the threshold of the matching result, which is simple and fast.
[0132] Figure 4 FIG. 1 is a flow chart of a palm vein recognition method provided by another exemplary embodiment. Figure 4 As shown, when applied to opening a vehicle door, the palm vein recognition method may include the following steps.
[0133] Registration stage:
[0134] 1) Obtaining palm vein images of authorized users;
[0135] 2) Extract ROI;
[0136] 3) performing enhancement preprocessing and correction preprocessing (e.g., γ correction preprocessing) on the ROI image respectively;
[0137] 4) Perform LBPH feature extraction on the data after enhancement preprocessing, and perform improved feature extraction on the data after correction preprocessing, where the improved features are the features extracted according to the above-mentioned HOG method;
[0138] 5) Enter both extracted features into the database.
[0139] Identification phase:
[0140] 1) Obtain the palm vein image of the current user;
[0141] 2) Extract ROI;
[0142] 3) Perform enhancement preprocessing on the ROI image;
[0143] 4) Extract LBPH features from the enhanced preprocessed data;
[0144] 5) LBPH feature matching with LBPH features in the database;
[0145] 6) Determine whether the current user is an authorized user. If the matching result is less than or equal to the first threshold value T1, it is determined that the current user identification is passed and the door is allowed to open; if the matching result is greater than or equal to the second threshold value T2, it is determined that the current user identification is not passed and the door is prohibited from opening; if the matching result is greater than the first threshold value T1 and less than the second threshold value T2, it is determined that the current user identification fails;
[0146] 7) When the recognition based on the LBPH feature fails, the ROI image is corrected and preprocessed;
[0147] 8) Improve feature extraction of the correction preprocessed data;
[0148] 9) Matching the improved features with the improved features in the database;
[0149] 10) If the current user is determined to be an authorized user based on the improved features, the door is allowed to open; if the current user is determined to be not an authorized user based on the improved features, the door is prohibited from opening.
[0150] The steps in the dotted part are performed in a small number of cases. In most cases, the computing power of palm vein recognition is small, and the second method is used for recognition when necessary. When the palm vein recognition method disclosed in the present invention is applied to an on-board chip, it can reduce the dependence on the computing performance of the on-board chip.
[0151] Based on the same inventive concept, the present disclosure also provides a palm vein recognition device. Figure 5 FIG. 1 is a block diagram of a palm vein recognition device provided by an exemplary embodiment. Figure 5 As shown, the palm vein recognition device 500 includes an acquisition module 501 , a first determination module 502 , a second determination module 503 , a third determination module 504 , an update module 505 and a first recognition module 506 .
[0152] The acquisition module 501 is used to acquire the palm vein image of the current user.
[0153] The first determination module 502 is used to determine the unit in the palm vein image.
[0154] The second determination module 503 is used to determine the gradient histogram vector of each unit, where the gradient histogram vector is a directional gradient histogram vector.
[0155] The third determination module 504 is used to determine the weight of each unit, wherein the weight of the unit is related to the accumulated value of the gradient magnitude corresponding to at least one gradient direction interval in the gradient histogram vector of the unit.
[0156] The updating module 505 is used to update the gradient histogram vector of each unit according to the gradient histogram vector of the unit and the determined weight.
[0157] The first identification module 506 is used to identify the current user according to the updated gradient histogram vector.
[0158] Optionally, the third determination module 504 is used to: for each unit, determine the weight of the unit according to the sum of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0159] Optionally, the third determination module 504 is used to: for each unit, determine the weight of the unit according to the maximum value of the accumulated values of gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
[0160] Optionally, every predetermined number of adjacent units in the palm vein image constitutes a block, and the weight of each block in the palm vein image is equal. For each block, the weight of the block is the sum of the weights of each unit in the block.
[0161] Optionally, the updating module 505 is used to: for each unit, use the product of the gradient histogram vector of the unit and the weight of the unit as the updated gradient histogram vector of the unit.
[0162] Optionally, the updating module 505 is used to: for each unit, use the weighted sum of the gradient histogram vector of the unit and the gradient histogram vectors of the units adjacent to the unit as the updated gradient histogram vector of the unit.
[0163] Optionally, the first identification module 506 is used to:
[0164] For each unit in the palm vein image, the updated gradient histogram vector is matched with a predetermined feature vector to obtain a matching result of the unit;
[0165] The current user is identified according to the matching result.
[0166] Optionally, the first identification module 506 is used to:
[0167] Determine a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or determine a unit whose weight ranking is higher than a predetermined ranking threshold as a high-weight unit;
[0168] If the number of successfully matched units is greater than a predetermined number threshold, and all units with high weights are determined to be successfully matched, it is determined that the current user has been identified successfully.
[0169] Optionally, the palm vein recognition device 500 further includes a judgment module.
[0170] The judgment module is used to: judge a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or judge a unit whose weight ranking is higher than a predetermined ranking threshold as a high-weight unit.
[0171] Optionally, the palm vein recognition device 500 further includes a second recognition module.
[0172] The second recognition module is used to: extract features from the palm vein image according to the LBPH method; and recognize the current user according to the extracted features.
[0173] The second determination module 503 is used to determine the gradient histogram vector of each unit when the features extracted according to the LBPH method fail to identify the current user.
[0174] Optionally, the second identification module is used for:
[0175] According to the chi-square test method, the extracted features are matched with the predetermined features;
[0176] When the matching result is less than the first threshold, it is determined that the current user has been identified;
[0177] When the matching result is greater than the second threshold, determining that the current user identification fails;
[0178] When the matching result is greater than the first threshold and less than the second threshold, it is determined that the current user identification fails.
[0179] Through the above technical solution, on the basis of extracting image features using the HOG method in the related art, a weight is set for each unit in the palm vein image, and the gradient histogram vector of each unit is updated according to these weights. Since the weight is related to the accumulated value of the gradient size corresponding to at least one gradient direction interval in the gradient histogram vector, the characteristic difference of the gradient is more obvious in the gradient histogram vector of each unit after the update, and the current user is identified according to the updated gradient histogram vector, and the identification result is more accurate.
[0180] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, and the program instructions, when executed by a processor, implement the steps of the above method provided by the present disclosure.
[0181] The present disclosure also provides a palm vein recognition device, comprising a processor and a memory for storing instructions executable by the processor, wherein the processor is configured to implement the steps of the above method provided by the present disclosure.
[0182] The present disclosure also provides a vehicle, comprising the above-mentioned palm vein recognition device provided by the present disclosure.
[0183] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.
[0184] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.
[0185] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.
Claims
1. A palm vein recognition method, characterized in that: The method comprises: Get the palm vein image of the current user; determining a unit in the palm vein image; Determine a gradient histogram vector of each unit, wherein the gradient histogram vector is a directional gradient histogram vector; For each unit, determine a weight of the unit, wherein the weight of the unit is related to an accumulated value of a gradient magnitude corresponding to at least one gradient direction interval in a gradient histogram vector of the unit; For each unit, updating the gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight; The current user is identified according to the updated gradient histogram vector.
2. The method according to claim 1, characterized in that The step of determining the weight of each unit includes: For each unit, the weight of the unit is determined according to the sum of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
3. The method according to claim 1, characterized in that The step of determining the weight of each unit includes: For each unit, the weight of the unit is determined according to the maximum value of the accumulated values of the gradient magnitudes corresponding to each gradient direction interval in the gradient histogram vector of the unit.
4. The method according to claim 1, characterized in that: Every predetermined number of adjacent units in the palm vein image forms a block, and the weight of each block in the palm vein image is equal. For each block, the weight of the block is the sum of the weights of the units in the block.
5. The method according to claim 1, characterized in that The step of updating the gradient histogram vector of each unit according to the gradient histogram vector of the unit and the determined weight comprises: For each unit, the product of the gradient histogram vector of the unit and the weight of the unit is used as the updated gradient histogram vector of the unit.
6. The method according to claim 1, characterized in that The step of updating the gradient histogram vector of each unit according to the gradient histogram vector of the unit and the determined weight comprises: For each unit, a weighted sum of the gradient histogram vector of the unit and the gradient histogram vectors of units adjacent to the unit is used as the updated gradient histogram vector of the unit.
7. The method according to claim 1, characterized in that The identifying the current user according to the updated gradient histogram vector includes: For each unit in the palm vein image, matching the updated gradient histogram vector with a predetermined feature vector to obtain a matching result of the unit; The current user is identified according to the matching result.
8. The method according to claim 7, characterized in that The identifying the current user according to the matching result includes: Determine a unit whose weight is greater than a predetermined weight threshold as a high-weight unit; or determine a unit whose weight ranking is higher than a predetermined ranking threshold as the high-weight unit; If the number of successfully matched units is greater than a predetermined number threshold, and all units determined to be high-weighted are successfully matched, it is determined that the current user has been identified successfully.
9. The method according to claim 1, characterized in that: The method further comprises: extracting features from the palm vein image according to a local binary pattern histogram (LBPH) method; identifying the current user according to the extracted features; The determining of the gradient histogram vector of each unit includes: determining the gradient histogram vector of each unit when the feature extracted according to the LBPH method fails to identify the current user.
10. The method according to claim 9, characterized in that The identifying the current user according to the extracted features includes: According to the chi-square test method, the extracted features are matched with the predetermined features; When the matching result is less than the first threshold, determining that the current user has been identified; When the matching result is greater than the second threshold, determining that the current user identification fails; When the matching result is greater than the first threshold and less than the second threshold, it is determined that the current user identification fails.
11. A palm vein recognition device, characterized in that: The device comprises: An acquisition module is used to acquire a palm vein image of the current user; A first determination module, used to determine a unit in the palm vein image; A second determination module is used to determine a gradient histogram vector of each unit, wherein the gradient histogram vector is a directional gradient histogram vector; A third determination module is used to determine the weight of each unit, wherein the weight of the unit is related to the accumulated value of the gradient magnitude corresponding to at least one gradient direction interval in the gradient histogram vector of the unit; An updating module, configured to update, for each unit, a gradient histogram vector of the unit according to the gradient histogram vector of the unit and the determined weight; The first identification module is used to identify the current user according to the updated gradient histogram vector.
12. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 10 are implemented.
13. A palm vein recognition device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to implement the steps of the method according to any one of claims 1 to 10.
14. A vehicle, characterized in that: Including the palm vein recognition device as described in claim 13.