CN103793692A - Low-resolution multi-spectral palm print and palm vein real-time identity recognition method and system - Google Patents

Abstract

The invention discloses a low-resolution multispectral palmprint and palm vein real-time identification method and system. The system collects palm images under five kinds of spectra, and makes full use of the complementarity of multi-spectral image information to improve the system recognition rate; at the same time, it collects palm vein information under near-infrared light spectrum, so that the system has the ability of live detection and improves the anti-counterfeiting of the system Attack capability; use down-sampling technology based on bi-cubic interpolation to improve the speed of feature extraction and other post-processing, and save the storage space of feature templates; use multi-scale and multi-directional filter banks for feature extraction to reduce the impact of illumination changes on feature extraction , improve the robustness of the system; use the hash table to encode the feature matrix to further improve the system matching speed; use the unique fractional multi-spectral feature fusion method to further improve the recognition rate of the system. The system realized by the invention has the characteristics of high resolution, fast identification speed, stable system, good expansibility, anti-counterfeit attack and the like.

Description

Low-resolution multispectral palmprint and palm vein real-time identification method and system

technical field

The invention relates to the technical field of biological feature identification, in particular to a low-resolution multispectral palmprint and palm vein real-time identification method and system.

Background technique

Identification is an important part of human industrial production, commercial activities and daily life. Currently commonly used identification methods include traditional methods such as keys, smart cards, and passwords. Keys and smart cards have a high degree of recognition, but are easy to lose and copy; passwords are easy to use but easy to forget and crack. Traditional identification methods cannot meet the needs of human production and life. Therefore, biometric identification methods have emerged as the times require.

Biometric identification refers to the technology that computers use human physiological or behavioral characteristics for personal identification. At present, the biometric methods researched and used by people mainly include fingerprint recognition, face recognition, iris recognition, palmprint recognition and so on.

Fingerprint identification is the earliest biometric identification method, which has a long history and is easy to implement. The main problem of the fingerprint identification method is that the fingerprint is the outer layer feature of the body, which is easy to copy, forge and damage. Moreover, the area of the fingerprint image is small and the amount of information contained is small, resulting in a low recognition rate and a small registration database, which limits its large-scale application. In addition, fingers that are too dry, too wet, or dirty will easily cause the fingerprint image to be blurred and cannot be compared normally.

Face recognition is widely used. The features that can be used for recognition include eyes, nose, mouth, eyebrows, face contour and positional relationship. It can be used as an important part of multi-modal biometric feature recognition. Its disadvantage is that the recognition accuracy is low, it is greatly affected by the environment, and its practicability is not strong.

Because the iris feature information is rich and almost unchanged for a lifetime, iris recognition has the lowest error rate among various biometric identification methods, and it has always been the identification technology adopted by high-end security equipment. However, compared with other biometric technologies, the equipment is complex and expensive, and the recognition requires the active cooperation of the person to be recognized, and the fixed face focuses on the pupil, which is the worst acceptability.

Palmprint recognition is a relatively new biometric technology. Palmprint images include rich information features such as palm main lines, wrinkles, fine textures, ridge endings, bifurcation points, etc. These features are clear and stable. Moreover, the system does not require high image resolution during identification, and palmprint image collection is relatively easy, convenient and quick, and it is a non-invasive identification method, which is relatively easy for users to accept. However, compared with fingerprint images, palmprint images are much larger, which brings many difficulties to image feature extraction, matching and storage, and cannot guarantee the real-time requirements of the recognition system. Moreover, single-spectrum palmprint recognition systems cannot prevent counterfeiting attacks.

The palm vein belongs to the characteristics of blood vessels inside the human body, which has good uniqueness and stability; it is not easily affected by pollution, wear, aging, scars and other problems. The internal characteristics of human beings are alive, so they cannot be imitated and stolen by technical means, and have very good security. However, palm vein recognition also has some disadvantages. For example, because the vein is located under the superficial skin, image acquisition is more difficult than palm prints, and there are special requirements for equipment. High, so more complex image processing algorithms are required.

Based on the respective shortcomings of palmprint and palm vein recognition, although the three-dimensional palmprint recognition method can be used to solve some of these problems, the expensive and bulky devices make it difficult to apply it in practical applications. One of the solutions to the problem is to adopt multispectral palmprint palm vein imaging, that is, images are captured under multiple spectral conditions. The imaging resolution of the existing multi-spectral identification system is generally greater than 300DPI, that is, it has a high-resolution image, but the feature extraction of the high-resolution image cannot meet the real-time requirements of the system.

Contents of the invention

In order to overcome the problems existing in the prior art, the object of the present invention is to provide a low-resolution multispectral palmprint and palm vein real-time identification method and system, which can meet the identification requirements, improve the identification performance and speed up the matching speed, save compression The storage space of the feature code also satisfies the real-time function of the system.

The technical scheme adopted in the present invention is:

Low-resolution multispectral palmprint and palm vein real-time identification method, including registration phase and identification phase.

The registration phase includes:

a. collect palm images to be registered, the palm images include five images collected under white light, red light, green light, blue light, and near-infrared spectrum;

b. Perform ROI extraction on the palm image to be registered, and use bicubic interpolation to down-sample the resulting ROI image;

c. Using a multi-scale and multi-directional filter to extract features from the ROI image obtained, five groups of feature vectors corresponding to different spectra are obtained, and the feature vectors are encoded to generate feature templates and stored in the feature database;

The identification phase includes:

a. collect the palm image to be identified, and the palm image includes five images collected under white light, red light, green light, blue light, and near-infrared spectrum;

b. ROI extraction is performed on the palm image to be recognized, and the resulting ROI image is down-sampled using bicubic interpolation;

c. Using a multi-scale and multi-directional filter to perform feature extraction on the acquired ROI image, obtain five sets of feature vectors corresponding to different spectra, and encode the feature vectors to generate input features;

d. Perform single-spectrum feature matching on the input features and the feature templates stored in the feature database one by one, and perform score-level weight fusion on the five matching scores obtained, and finally use the nearest neighbor algorithm to make decisions based on the fusion scores to obtain the recognition result .

Further, before extracting the ROI of the palm image, it also includes preprocessing and reference coordinate system positioning, wherein the preprocessing includes binarization processing and morphological processing, and the morphological processing includes image erosion, expansion, and closing operations. The positioning of the reference coordinate system includes: extracting the contour curves of the palm and fingers from the binarized image, and determining the reference coordinate system of the image through the Harris corner detection method.

Further, the step of extracting features from the ROI image includes: using a non-downsampled bandpass pyramid filter to convolve with the downsampled image, and then convolve the output response after convolution with a non-downsampled eight-direction filter bank , and finally encode the maximum value of the filter output response in eight directions in a competitive manner to form a binary feature vector of 0 or 1.

Further, the template features are stored in a feature database in the form of a hash table.

The present invention also provides a low-resolution multispectral palmprint and palm vein real-time identification system, including:

An image acquisition module, the image acquisition module includes a multispectral active light source, a CCD image sensor, and a control unit connected with the multispectral active light source;

An image preprocessing module, the input end of the image preprocessing module is connected with the output end of the CCD image sensor, and is used for preprocessing the palm image collected by the CCD image sensor, positioning the reference coordinate system, extracting the ROI, and Sampling processing;

Feature extraction module, the input end of described feature extraction module is connected with the output end of image preprocessing module, is used for carrying out feature extraction to the palm image after preprocessing;

A storage module, the storage module is provided with a feature database, which is connected to the output of the feature extraction module, and is used to store the template features obtained in the registration phase;

Recognition and decision-making module, the recognition and decision-making module is connected with the feature extraction module and the feature database, and is used to perform single-spectrum feature matching on the input features to be identified and the template features, and carry out fractional weight fusion of the matching results of different spectra, and finally according to The fusion score uses the nearest neighbor algorithm for decision-making, and then obtains the recognition result.

Further, the multi-spectral active light source includes a white light source, a red light source, a green light source, a blue light source, and a near-infrared light source that are alternately distributed in a ring around the CCD image sensor. dark.

The low-resolution multispectral palmprint and palm vein real-time identification method and system provided by the present invention mainly have the following beneficial effects:

(1) Collect palm images under five kinds of spectra, based on the multi-spectral image fusion method, make full use of multi-spectral image information, and improve the system recognition rate;

(2) The down-sampling technology based on bi-cubic interpolation is adopted to increase the speed of feature extraction and matching, save the storage space of feature templates, speed up the recognition speed, ensure the real-time requirements of the system, and reduce the system cost;

(3) Use multi-scale and multi-directional filters for feature extraction to reduce the impact of illumination changes on feature extraction;

(4) The palm vein information of the palm is collected under the near-infrared light spectrum, so that the system has the ability to detect the living body and improve the system's ability to prevent counterfeiting attacks.

Description of drawings

The present invention will be further described below in conjunction with accompanying drawing and example.

Fig. 1 is low-resolution multi-spectral palmprint of the present invention, palm vein real-time identification method and system implementation main process;

Fig. 2 is the main component module of low-resolution multi-spectral palmprint of the present invention, palm vein real-time identification method and system;

Fig. 3 is the palm image preprocessing flow of low-resolution multispectral palmprint of the present invention, palm vein real-time identification method and system;

Fig. 4 is the multi-scale and multi-directional filter bank feature extraction process of the low-resolution multi-spectral palmprint and palm vein real-time identification method and system of the present invention;

Fig. 5 is a detailed flowchart of the identification process of the low-resolution multispectral palmprint and palm vein real-time identification method and system of the present invention.

Detailed ways

Below in conjunction with accompanying drawing and specific example the present invention is described in further detail:

Referring to Fig. 1, the low-resolution multi-spectral palmprint, palm vein real-time identity recognition method and system implementation of the present invention mainly include: (1) collecting images containing palmprint and palm vein information under multi-spectral active light source; (2) ) Binarization, morphological processing, ROI extraction and bi-cubic interpolation downsampling of the palm image; (3) Multi-scale and multi-directional filter bank feature extraction for ROI of various spectral images to obtain corresponding features respectively, and use The hash table encodes the extracted features; (4) Match the features obtained under different spectral images to obtain matching scores, and adopt the score-level fusion method to fuse; (5) According to the fusion scores, use the nearest neighbor decision-making method to obtain recognition result.

Specifically, the identification method of the present invention includes a registration stage and an identification stage.

The registration phase includes:

a. collect palm images to be registered, the palm images include five images collected under white light, red light, green light, blue light, and near-infrared spectrum;

b. Perform ROI extraction on the palm image to be registered, and use bicubic interpolation to down-sample the resulting ROI image;

c. Using a multi-scale and multi-directional filter bank to perform feature extraction on the acquired ROI image, obtain five groups of feature vectors corresponding to different spectra, and encode the feature vectors to generate feature templates and store them in the feature database;

The identification phase includes:

a. collect the palm image to be identified, and the palm image includes five images collected under white light, red light, green light, blue light, and near-infrared spectrum;

b. ROI extraction is performed on the palm image to be recognized, and the resulting ROI image is down-sampled using bicubic interpolation;

c. Using a multi-scale and multi-directional filter bank to perform feature extraction on the acquired ROI image, obtain five sets of feature vectors corresponding to different spectra, and encode the feature vectors to generate input features;

d. Perform single-spectrum feature matching on the input features and the feature templates stored in the feature database one by one, and perform score-level weight fusion on the five matching scores obtained, and finally use the nearest neighbor algorithm to make decisions based on the fusion scores to obtain the recognition result .

Referring to accompanying drawing 2, the system of the present invention is mainly made up of image acquisition module, image preprocessing module, feature extraction module, storage module, recognition decision-making module etc.

(1) Image acquisition module

The image acquisition module includes a multispectral active light source, a CCD image sensor, and a control unit connected with the multispectral active light source. The multi-spectrum active light source includes a blue light source (470nm), a green light source (525nm), a red light source (660nm), a white light source and a near-infrared light source (850nm), and each light source is alternately distributed in a ring around the CCD camera. Because human skin has different absorption rates for different wavelengths of spectrum, that is to say, different images will be obtained under different spectra; at the same time, human epidermis has a certain thickness, and some spectra can penetrate human epidermis to obtain The texture features of subcutaneous tissue, which are not easy to forge, can improve the anti-spoofing of the recognition system. Therefore, the present invention uses five light sources, including four visible light sources and one near-infrared light source. Under the irradiation of the first four visible light sources, four palm images containing palmprint information are obtained; the near-infrared light source can penetrate the epidermis to obtain the texture characteristics of the subcutaneous tissue, that is, the characteristics of veins, and thus obtain palm vein information. internal image. The special device for palm image acquisition used in the present invention adopts a CCD-based image sensor. The advantages of this acquisition method are fast image acquisition speed, high system integration and good real-time performance. The invention adopts five kinds of LED arrays as the light source to surround the CCD image sensor, and adds a filter in front of the light source to ensure uniform and stable illumination conditions.

(2) Image preprocessing module

The input end of the image preprocessing module is connected to the output end of the CCD image sensor, and is used for preprocessing, reference coordinate system positioning, ROI extraction, and down-sampling processing on the palm image collected by the CCD image sensor.

Shown in conjunction with accompanying drawing 3, the image preprocessing steps among the present invention are as follows:

Firstly, image binarization is performed, and morphological operations such as erosion and expansion are used to improve the image binarization results.

Then extract the contour curves of the palm and fingers from the binary image obtained in the previous step, and then use the Harris corner detection method to extract two points, the valley point P1 between the index finger and the middle finger and the valley point P2 between the ring finger and the little finger. The implementation method is: take a window (5×5) centered on the target pixel point, and move the window up, down, left, and right along the target point, and calculate the gray level changes inside the window in the four directions during the moving process, and put the 4 The minimum value of the difference between grayscale changes is set as the corresponding function value of the corner point of the target pixel point, and when this value is greater than the threshold value, it is used as the corner point. Therefore, the connection line between P1 and P2 is determined as the Y axis, and the midpoint of the line segment from P1 to P2 and perpendicular to the Y axis is used as the X axis. There is an image registration problem, so after the palm image reference coordinate system is determined under white light illumination, the palm coordinate systems under the other four illuminations are also determined accordingly. The image preprocessing before this step is only performed on the white spectrum image to improve the image processing speed.

Then, according to the determined reference coordinate system, ROI extraction is performed on the five palm images respectively, and the central area of the palm is extracted as the region of interest (ROI). The ROI area is a square area, and its side length is about 70% of the width of the palm. Because the ROI image contains the most important features and information in the palm image, the extraction of ROI reduces noise interference and improves the computing speed of the system at the same time. Because the effective image preprocessing method is used in the present invention, the interference caused by factors such as translation and rotation in real time of ROI extraction is reduced, and the recognition rate of the system is improved.

Finally, bicubic interpolation downsampling is performed on the ROI palms under the five illuminations. The bicubic interpolation uses the gray value of 16 points to be interpolated three times, not only considering the gray effect of 4 direct adjacent points, but also considering the effect of the gray value change rate between 12 adjacent points, so a lower The sampling pixels integrate the information of the pixels in the 4×4 area, which not only ensures the recognition accuracy, but also improves the recognition speed. The present invention adopts a downsampling ratio of 4:1. For example, the size of the source image is 128×128, and the size of the downsampled image is 32×32. Obviously, the image processing speed will be increased by 16 times.

(3) Feature extraction module

The input end of the feature extraction module is connected to the output end of the image preprocessing module, and is used for feature extraction of the preprocessed palm image.

Combined with Figure 4, the processing steps of feature extraction are as follows:

The multi-scale and multi-directional filter bank allows different direction decomposition on each scale, and the aspect ratio of the base support interval changes with the scale, showing "anisotropy" characteristics, which can realize the sparse representation of the image. The feature extraction method used in the present invention is a multi-scale and multi-directional filter bank, and its structure is divided into: non-subsampled pyramid (Non-subsampled Pyramid, NSP) decomposition and non-subsampled directional filter bank (Non-subsampled Directional FilterBank, NSDFB) decomposes two parts. Firstly, NSP is used to decompose the image at multiple scales, and the singular points in the image can be effectively "captured" through NSP decomposition; then, NSDFB is used to decompose the direction of high-frequency components, so as to obtain sub-scales of different scales and directions. with images (coefficients). The difference from the contourlet transform is that in the process of image decomposition and reconstruction, the multi-scale multi-directional filter bank does not perform down-sampling (decimation) after decomposition filtering and up-sampling of the synthesis filter for the signal components decomposed by NSP and NSDFB. Sampling (interpolation), so that it has multi-scale, good spatial and frequency domain local characteristics, and multi-directional characteristics.

The feature extraction process of the present invention is as follows: firstly adopt the non-subsampling bandpass pyramid filter P _f to convolve with the downsampling image I _{x, y} to obtain the sub-image f _{x, y} after the bandpass filter

f _x,y =I _x,y *P _f

The _Pf filter only allows texture information with certain robustness to be preserved for subsequent feature extraction; then, the output response after convolution is convolved with the non-subsampled multi-directional filter bank _Df to obtain multiple directional sub-images

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}$

表示在点(x,y)上各个方向子图像的方向系数。依据

对每一个像素点确定其最大系数对应的方向作为该像素点的方向特征

Indicates the direction coefficient of each direction sub-image on the point (x, y). in accordance with

For each pixel, determine the direction corresponding to its maximum coefficient as the direction feature of the pixel

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}$

According to F _x,y , the present invention uses a hash table to encode the direction feature. The palm images under five kinds of spectra collected by the present invention obtain five groups of directional features after feature extraction, and hash table coding stores multi-directional features in matrix form, and each group of directional features is composed of a column of 0 and 1 An array and an index sequence composed of four columns of common pixel coordinates and direction codes. The five palm images to be recognized are coded as five columns of feature vectors consisting of 0 and 1.

(4) Storage module

The storage module is provided with a feature database, which is connected to the output end of the feature extraction module, and is used for storing the template features acquired in the registration phase. In conjunction with accompanying drawing 2, what the present invention stores is the palm feature hash table, and an image is represented by five columns of binary values, and five columns represent feature vectors under different spectra respectively. Due to down-sampling, the storage space required here greatly reduced.

(5) Identification and decision-making module

The recognition decision-making module is connected with the feature extraction module and the feature database, and is used to perform single-spectrum feature matching on the input features to be identified and the template features, and perform score-level weight fusion on the matching results of different spectra, and finally adopt the nearest Neighboring algorithm to make decisions, and then get the recognition results.

In conjunction with accompanying drawing 5, in the present invention, palm images from five kinds of spectra are collected, and the images under each type of illumination of the palm to be recognized are matched with the feature templates in the database, and different meanings will be obtained according to different matching methods. match score. The present invention proposes two different feature matching methods. According to different matching methods, different score-level fusion methods will be used. When using the same-or feature matching method, the final decision score can be obtained by calculating the sum of the five sets of matching scores

S _F =SUM(S _R ,S _B ,S _G ,S _N ,S _W )

Among them, S _R , S _B , S _G , S _N , and S _W represent the matching scores obtained by the same-matching method under single spectrum, and S _F represents the score after fractional fusion; when using the exclusive-or matching method, it can be calculated by Take the maximum of the five sets of matching scores as the final decision score

S' _F =MAX(S' _R ,S' _B ,S' _G ,S' _N ,S' _W )

Experimental data show that both fusion methods can get high recognition rate.

According to different feature matching methods, matching scores with different meanings will be obtained. When using the same-or feature matching and sum fusion method, the maximum matching score is used as the basis for identification decisions

$\mathrm{<span\; class="notranslate">\; class</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; max</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ;</span>$

When using XOR feature matching and seeking the maximum fusion method, the minimum matching score is used as the basis for identification decisions

$\mathrm{<span\; class="notranslate">\; class</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; .</span>$

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of changes or modifications within the technical scope disclosed in the present invention. Replacement should be covered within the protection scope of the present invention.

Claims (6)

1. The low-resolution multispectral palmprint and palm vein real-time identification method mainly includes a registration stage and a recognition stage, and is characterized in that, The registration phase includes: a. Collect the palm image to be registered, the palm image includes five images collected under white light, red light, green light, blue light, and near-infrared spectrum; b. Perform ROI extraction on the palm image to be registered, and use bicubic interpolation to down-sample the resulting ROI image; c. Use multi-scale and multi-directional filter banks to extract features from the acquired ROI image, obtain five sets of feature vectors corresponding to different spectra, and encode the feature vectors to generate feature templates and store them in the feature database; The identification phase includes: a. Collect the palm image to be identified, the palm image includes five images collected under white light, red light, green light, blue light, and near-infrared spectrum; b. Perform ROI extraction on the palm image to be recognized, and use bicubic interpolation to down-sample the resulting ROI image; c. Use multi-scale and multi-directional filter banks to extract features from the acquired ROI image, obtain five sets of feature vectors corresponding to different spectra, and encode the feature vectors to generate input features; d. Match the input features with the feature templates stored in the feature database one by one for single-spectrum feature matching, and perform fractional weight fusion of the five matching scores obtained, and finally use the nearest neighbor algorithm to make decisions based on the fusion scores to obtain the recognition result . 2. low-resolution multispectral palmprint according to claim 1, palm vein real-time identity recognition method, it is characterized in that, before described palm image is carried out ROI extracting, also comprise preprocessing and reference coordinate system location, wherein said The preprocessing includes binarization processing and morphological processing, and the morphological processing includes image erosion, expansion, and closing operations; The positioning of the reference coordinate system includes: extracting the contour curves of the palm and fingers from the binarized image, and determining the reference coordinate system of the image through the Harris corner detection method. 3. low-resolution multi-spectral palmprint according to claim 1, palm vein real-time identity recognition method and system, it is characterized in that, described ROI image is carried out the step of feature extraction comprising: adopt non-subsampling band-pass pyramid filtering The filter is convolved with the downsampled image, and the output response after convolution is convolved with the non-downsampled eight-direction filter bank. Finally, the maximum value of the filter output response is encoded in the eight directions in a competitive manner to form 0 or A binary eigenvector of 1. 4. low-resolution multispectral palmprint according to claim 1, palm vein real-time identification method, is characterized in that, described template feature is stored in the palm feature database with hash table form. 5. A low-resolution multi-spectral palmprint and palm vein real-time identification system, characterized in that it includes: An image acquisition module, the image acquisition module includes a multispectral active light source, a CCD image sensor, and a control unit connected with the multispectral active light source; An image preprocessing module, the input end of the image preprocessing module is connected with the output end of the CCD image sensor, and is used for preprocessing the palm image collected by the CCD image sensor, positioning the reference coordinate system, extracting the ROI, and Sampling processing; Feature extraction module, the input end of described feature extraction module is connected with the output end of image preprocessing module, is used for carrying out feature extraction to the palm image after preprocessing; A storage module, the storage module is provided with a feature database, which is connected to the output of the feature extraction module, and is used to store the template features obtained in the registration phase; Recognition and decision-making module, the recognition and decision-making module is connected with the feature extraction module and the feature database, and is used to perform single-spectrum feature matching on the input features to be identified and the template features, and carry out fractional weight fusion of the matching results of different spectra, and finally according to The fusion score uses the nearest neighbor algorithm for decision-making, and then obtains the recognition result. 6. The low-resolution multi-spectral palmprint and palm vein real-time identification system according to claim 1, characterized in that: the multi-spectral active light source comprises a white light source, a red light source, and a red light source that are alternately distributed in a ring around the CCD image sensor. The light source, the green light source, the blue light source, and the near-infrared light source, the control unit is used to control the brightness and darkness of each light source in turn.

Images