CN106874824A - A kind of Frequency Band Selection method and apparatus for personal recognition - Google Patents

Abstract

The invention provides a frequency band selection method and device for palmprint recognition, so as to reduce the calculation amount in the process of palmprint recognition and improve the efficiency of palmprint recognition. The method includes: removing frequency bands corresponding to less information in the palmprint image to obtain a first candidate frequency band; selecting a frequency band corresponding to a lower equal error rate from the palmprint image based on a Gabor filter as the second frequency band. Alternative frequency bands; according to the k-clustering algorithm, several frequency band clusters with good clustering are calculated from the overlapping parts of the first candidate frequency band and the second candidate frequency band, and the frequency band clusters with good clustering The frequency band corresponding to the center of the center is finally used as the frequency band for palmprint recognition. The technical solution provided by the invention can significantly improve the efficiency and accuracy of palmprint recognition.

Description

A frequency band selection method and device for palmprint recognition

technical field

The invention belongs to the field of image processing, in particular to a frequency band selection method and device for palmprint recognition.

Background technique

With the improvement of imaging technology and computing performance, palmprint recognition has gradually developed into one of the mainstream biometric technologies. And the identification of hyperspectral images is also possible. In addition to the complex palm prints on the surface of the human palm, there are intricate arteries, veins and various connective tissues inside. Due to the different optical properties of different tissues under different spectra, different images can be obtained under different spectra. Therefore, combining palm images under various spectra can provide more effective information, thereby improving the accuracy of palm recognition.

In order to further improve the accuracy of palmprint recognition, an existing palmprint recognition method introduces multispectral imaging technology. Multispectral imaging technology can obtain images of palm epidermis and internal tissues and blood vessels under different spectra. However, due to the large number of pictures, the amount of computation increases dramatically. Another existing palmprint recognition method is based on multi-band image fusion at the pixel level, and applies wavelet and curvelet transform. In this method, the choice of frequency band is the key. In view of this, a method for selecting frequency bands is to exhaustively select the optimal combination of frequency bands.

However, since the number of frequency bands of the hyperspectrum is generally on the order of 10^2 or greater, for the palmprint recognition of the hyperspectrum, the efficiency of the above-mentioned palmprint recognition method by exhaustively selecting the optimal frequency band combination is too low.

Contents of the invention

The object of the present invention is to provide a frequency band selection method and device for palmprint recognition, so as to reduce the amount of computation in the process of palmprint recognition and improve the efficiency of palmprint recognition.

The first aspect of the present invention provides a kind of frequency band selection method for palmprint recognition, and described method comprises:

Eliminate frequency bands corresponding to less information in the palmprint image to obtain the first candidate frequency band;

Selecting a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band;

According to the k-clustering algorithm, a number of frequency band clusters with good clustering are calculated from the overlapping parts of the first candidate frequency band and the second candidate frequency band, and the centers of the good cluster frequency band clusters correspond to The frequency band is finally used as the frequency band for palmprint recognition.

A second aspect of the present invention provides a frequency band selection device for palmprint recognition, said device comprising:

The first selection module is used to remove frequency bands corresponding to less information in the palmprint image to obtain the first candidate frequency band;

The second selection module is used to select a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band;

A clustering module, configured to calculate several well-clustered frequency band clusters from the overlapping portion of the first candidate frequency band and the second candidate frequency band according to the k-clustering algorithm, and the clustered good frequency bands The frequency band corresponding to the center of the cluster is finally used as the frequency band for palmprint recognition.

Known from above-mentioned technical scheme of the present invention, on the one hand, before carrying out k-clustering algorithm, by the processing to palmprint image, filter out the frequency band that the information amount of palmprint image is more and the frequency band that waits for error rate is lower, therefore, In the process of palmprint recognition, only processing the palmprint images corresponding to the overlapping parts of these two frequency bands greatly reduces the amount of computation and can significantly improve the efficiency of palmprint recognition; The k-clustering algorithm is performed on the palmprint images corresponding to the frequency bands with a low error rate to calculate the clustering of the frequency bands with good clustering and select the optimal frequency band combination, which can not only reduce the calculation amount of palmprint recognition, but also significantly improve Accuracy of palmprint recognition.

Description of drawings

Fig. 1 is the realization flowchart of the frequency band selection method that is used for palmprint recognition that the embodiment of the present invention provides;

FIG. 2 is a schematic diagram of the corresponding relationship between entropy values and frequency bands in an image provided by Embodiment 2 of the present invention;

3 is a schematic diagram of the relationship between different frequency bands and their equal error rates in an image provided by Embodiment 3 of the present invention;

FIG. 4 is a schematic structural diagram of a frequency band selection device for palmprint recognition provided by Embodiment 4 of the present invention;

FIG. 5 is a schematic structural diagram of a frequency band selection device for palmprint recognition provided in Embodiment 5 of the present invention;

FIG. 6 is a schematic structural diagram of a frequency band selection device for palmprint recognition provided by Embodiment 6 of the present invention.

detailed description

In order to make the object, technical solution and beneficial effects of the present invention more clear, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Embodiments of the present invention provide a method for selecting frequency bands for palmprint recognition, said method comprising: removing frequency bands corresponding to less information in a palmprint image to obtain a first candidate frequency band; In the palmprint image, select the frequency band corresponding to the lower error rate as the second candidate frequency band; according to the k clustering algorithm, calculate a number of overlapping parts from the first candidate frequency band and the second candidate frequency band Well-clustered frequency bands are clustered, and the frequency band corresponding to the center of the good-clustered frequency band cluster is finally used as the frequency band for palmprint recognition. The embodiment of the present invention also provides a corresponding frequency band selection device for palmprint recognition. Each will be described in detail below.

Please refer to accompanying drawing 1, it is the realization flowchart of the frequency band selection method for palmprint recognition that embodiment one of the present invention provides, mainly comprises following steps S101 to step S103:

S101. Eliminate frequency bands corresponding to less information in the palmprint image to obtain a first candidate frequency band.

It should be noted that images obtained under different frequency bands contain different amounts of information, and palmprint images also conform to this rule. Moreover, palmprint images with rich information are easier to identify than palmprint images with less information. Therefore, in the embodiment of the present invention, the frequency band corresponding to less information in the palmprint image can be eliminated first to obtain the first candidate frequency band, which can be specifically realized through the following steps S1011 and S1012:

S1011. Calculate the gray distribution value of the palmprint image.

In an embodiment of the present invention, the gray distribution value p of the palmprint image can be calculated by the following _formula (1):

..............Formula 1)

Among them, I _{i, j} represents the palmprint image of 8-bit code, the subscript i and j of I _i, j represent the position of the image, m and n are the size of palmprint image, k is the palmprint image _{Ii, j} grayscale.

S1012. Calculate the entropy corresponding to each frequency band in the palmprint image according to the gray distribution value calculated in step S1011, and use the frequency band with a larger entropy value as the first candidate frequency band.

In the embodiment of the present invention, according to the gray distribution value calculated in step S1011, the entropy E of the palmprint image can be calculated by the following formula (2):

..........Formula (2)

For an image, the greater its entropy value, the more information it contains and the better its quality; palmprint images also conform to this rule. Therefore, in the embodiment of the present invention, when calculating the entropy corresponding to each frequency band in the palmprint image, the frequency band with a larger entropy value can be used as the first candidate frequency band to improve the recognition rate of subsequent palmprint recognition.

As shown in Figure 2, it is the corresponding relationship between the entropy value and the frequency band in the image. In the embodiment of the present invention, obviously, if the palmprint image corresponds to frequency bands with wavelengths below 590nm and above 1020nm, and the remaining frequency bands with wavelengths between 590nm and 1020nm, the entropy value is relatively large, indicating that in this The information content of images within a frequency band range is compared, therefore, the frequency band with a wavelength between 590nm and 1020nm can be used as the first candidate frequency band.

S102. Select a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as a second candidate frequency band.

In the field of image processing, Equal Error Rate (Equal Error Rate, EER) is derived from the two concepts of False Rejection Rate (FRR, False Rejection Rate) and False Acceptance Rate (FAR, False Acceptation Rate). The rate is the probability of false rejection, which is to explain the problem from within-class matching. If there are 10 samples of volunteers, and each volunteer has 20 samples, then compared with the intra-class test, for example, for No. 1 volunteer, the same class Between these 20 images, match each other (assuming a 1:1 match), and can perform (20*19)/2 times without repeating each other; if 10 volunteers perform such a test, it is 10*(20* 19)/2 times, the above is the total number of intra-class matches. Each match will get a matching value th according to the matching algorithm. The preset threshold is TH. If th>TH, it will be rejected by mistake; the false recognition rate is the error The probability of acceptance is to explain the problem from the matching between classes, that is, the matching between different classes. If the matching value th obtained according to the matching algorithm is less than the preset threshold TH, it will be considered to belong to the same class. This is an error. accept. The formula for calculating FRR is The formula for calculating FAR is where NGRA is the total number of intra-class tests, NIRA is the total number of between-class tests, NFR and NFA are the number of false rejections and false acceptances. Since FRR and FAR contradict each other, the probability when FRR and FAR are equal is EER.

As an embodiment of the present invention, selecting a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band can be realized through the following steps S1021 to S1022:

S1021. Filter the palmprint image based on the Gabor filter.

Specifically, based on the Gabor filter, the palmprint image filtering can be based on the Gabor filter represented by the function G(x, y, θ, u, σ), and the six functions of the function G(x, y, θ, u, σ) Direction to described palmprint image filtering, wherein, function G (x, y, θ, u, σ) is defined as:

in, x and y represent the position coordinates of the Gabor filter, u is the frequency of the sine wave, θ is the direction of the function G(x,y,θ,u,σ), σ is the variance of the Gaussian function, and the function G(x,y ,θ,u,σ) the six directions are 0, with

In the embodiment of the present invention, the two-dimensional Gabor filter can effectively extract the direction information in the palmprint image, and in the filtered palmprint image, different directions are expressed as different gray levels.

S1022. For the filtered palmprint image, calculate the equal error rate of each frequency band, and use the corresponding frequency band whose equal error rate is less than a preset threshold as a second candidate frequency band.

After the palmprint image is passed through the Gabor filter, what is obtained is a coded grayscale image. Each pixel in the picture is coded into a 3-bit (bit) number. For any two palmprint images, you can first Convert it to Gabor Filter Map, and then calculate the Hamming distance between the two, that is, the "matching value" of intra-class or inter-class matching, so as to calculate EER.

Since the lower the EER, the higher the accuracy of the identification system is, after calculating the equal error rate of each frequency band, the corresponding frequency band whose equal error rate is less than the preset threshold can be used as the second candidate frequency band. The relationship between different frequency bands and their equal error rates as shown in accompanying drawing 3, as can be seen from accompanying drawing 3, the frequency band between 580nm to 1080nm wavelength, its equal error rate is relatively low, therefore, as an embodiment of the present invention, A frequency band with a wavelength between 580 nm and 1080 nm may be used as a second candidate frequency band.

S103, according to the k-clustering algorithm, calculate several frequency band clusters with good clustering from the overlapping part of the first candidate frequency band and the second candidate frequency band, and the frequency band corresponding to the center of the good cluster frequency band cluster is used as The frequency band finally used for palmprint recognition.

The k-clustering (K-Means) algorithm is an unsupervised learning method that can reveal the relationship between various frequency bands. The purpose of the embodiments of the present invention is to select a frequency band that contains more information and has a lower error rate for palmprint recognition. In this way, the first candidate frequency band and the second candidate frequency band mentioned in the foregoing embodiments For example, the frequency band with a wavelength between 590nm and 1020nm illustrated in Figure 2 and the frequency band with a wavelength between 580nm and 1080nm illustrated in Figure 3, the overlapping part, that is, the frequency band with a wavelength between 580nm and 1020nm can take into account both It has the advantages of more information, lower error rate, etc. Therefore, according to the k-clustering algorithm, several well-clustered frequency band clusters can be calculated from the frequency bands with wavelengths between 580nm and 1020nm.

As an embodiment of the present invention, there are 45 frequency bands in the frequency band between 580nm and 1020nm, and according to the k-clustering (K-Means) algorithm, these 45 frequency bands are clustered into 4 classes, and the centers are respectively 600nm and 770nm , 850nm and 1000nm corresponding frequency bands; the calculation process and results of the algorithm show that these 4 classes are well-clustered frequency band clusters, and the frequency band corresponding to the center is finally used for palmprint recognition. The so-called good clustering means that in the process of the k-clustering algorithm, when the number of clusters is set to 4, four values are randomly selected from the frequency bands with wavelengths between 580nm and 1020nm as the initial cluster centers. The sum S of Hamming distances between cluster centers is selected to measure the clustering quality. As the values of the centers of the four classes are different, the value of S obtained is also different. When the value of S remains unchanged, it can be considered that a stable k-clustering center has been found, that is, 600nm, 770nm, 850nm and 1000nm correspond to 4 frequency bands. At this time, the combination of these four frequency bands is optimal, and is the most suitable frequency band for palmprint recognition.

From the frequency band selection method that is used for palmprint recognition of above-mentioned accompanying drawing 1 example as can be known, on the one hand, by the processing to palmprint image, filter out the frequency band that the amount of information of palmprint image is more and the frequency band that etc. error rate are lower, Therefore, in the process of palmprint recognition, only processing the palmprint images corresponding to the overlapping parts of these two frequency bands greatly reduces the amount of computation and can significantly improve the efficiency of palmprint recognition; The k-clustering algorithm is performed on the palmprint images corresponding to the frequency band and the frequency band with a low error rate, and the frequency band clustering with good clustering is calculated, and the optimal frequency band combination is selected, which can not only reduce the calculation amount of palmprint recognition, but also can Significantly improve the accuracy of palmprint recognition.

Please refer to FIG. 4 , which is a schematic structural diagram of a frequency band selection device for palmprint recognition provided by Embodiment 4 of the present invention. For ease of description, Fig. 4 only shows the parts related to the embodiment of the present invention. The frequency band selection device for palmprint recognition illustrated in FIG. 4 may be the subject of execution of the frequency band selection method for palmprint recognition illustrated in FIG. 1 . The frequency band selection device for palmprint identification of accompanying drawing 4 example mainly comprises first selection module 401, the second selection module 402 and clustering module 403, wherein:

The first selection module 401 is used to remove the frequency band corresponding to less information in the palmprint image to obtain the first candidate frequency band;

The second selection module 402 is used to select a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band;

The clustering module 403 is used to calculate several well-clustered frequency band clusters from the overlap between the first candidate frequency band and the second candidate frequency band according to the k-clustering algorithm, and the frequency band clusters with good clustering The frequency band corresponding to the center is finally used as the frequency band for palmprint recognition.

It should be noted that, in the embodiment of the frequency band selection device for palmprint recognition illustrated in the accompanying drawing 4 above, the division of each functional module is only an example, and in actual applications, it can be based on needs, such as configuration requirements of corresponding hardware or software Considering the convenience of realization, the above-mentioned function allocation is completed by different functional modules, that is, the internal structure of the frequency band selection device for palmprint recognition is divided into different functional modules to complete all or part of the functions described above. Moreover, in practical applications, the corresponding functional modules in this embodiment may be implemented by corresponding hardware, or may be completed by corresponding hardware executing corresponding software. For example, the aforementioned second selection module may be capable of executing the aforementioned The Gabor filter selects the hardware corresponding to the lower frequency band of the equal error rate as the second alternative frequency band from the palmprint image, such as the second selector, and can also be the general processing that can execute the corresponding computer program so as to complete the aforementioned functions device or other hardware devices; as the aforementioned clustering module, it can be performed according to the k-clustering algorithm, from the overlapping part of the first candidate frequency band and the second candidate frequency band to calculate several well-clustered frequency band clusters Hardware, such as a clusterer, may also be a general processor or other hardware device capable of executing corresponding computer programs to complete the aforementioned functions (the above description principles can be applied to each embodiment provided in this specification).

The first selection module 401 of accompanying drawing 4 example can comprise gray level calculation unit 501 and entropy calculation unit 502, as shown in accompanying drawing 5 the frequency band selection device for palmprint recognition that embodiment five of the present invention provides, wherein:

A grayscale calculation unit 501, used to calculate the grayscale distribution value of the palmprint image;

The entropy calculation unit 502 is used to calculate the entropy corresponding to each frequency band in the palmprint image according to the gray distribution value calculated by the gray scale calculation unit 501, and use the frequency band with a larger entropy value as the first candidate frequency band.

The second selection module 402 of accompanying drawing 4 example can comprise filtering unit 601 and equal error rate computing unit 602, as shown in accompanying drawing 6 embodiment six of the present invention provides the frequency band selection device for palmprint identification, wherein:

Filtering unit 601, for palmprint image filtering based on Gabor filter;

The equal error rate calculation unit 602 is configured to calculate the equal error rate of each frequency band for the filtered palmprint image, and use the corresponding frequency band whose equal error rate is less than a preset threshold as a second candidate frequency band.

In the frequency band selection device for palmprint recognition shown in accompanying drawing 6, the filtering unit 601 is specifically used for the Gabor filter based on the function G (x, y, θ, u, σ), and in the function G (x, y , θ, u, σ) to filter the palmprint image in six directions, where the function G(x, y, θ, u, σ) is defined as follows:

in, x and y represent the position coordinates of the Gabor filter, u is the frequency of the sine wave, θ is the direction of the function G(x,y,θ,u,σ), σ is the variance of the Gaussian function, and the function G(x,y ,θ,u,σ) the six directions are 0, with

It should be noted that the information interaction and execution process between the modules/units of the above-mentioned device are based on the same idea as the method embodiment of the present invention, and the technical effect it brings is the same as that of the method embodiment of the present invention. The specific content can be Refer to the descriptions in the method embodiments of the present invention, and details are not repeated here.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: Read Only Memory (ROM, Read Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

Above, the frequency band selection method and device for palmprint recognition provided by the embodiments of the present invention have been introduced in detail. In this paper, specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for To help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation and scope of application. In summary, the content of this specification It should not be construed as a limitation of the invention.

Claims (10)

1. a kind of frequency band selection method for palmprint recognition, is characterized in that, described method comprises: Eliminate frequency bands corresponding to less information in the palmprint image to obtain the first candidate frequency band; Selecting a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band; According to the k-clustering algorithm, a number of frequency band clusters with good clustering are calculated from the overlapping parts of the first candidate frequency band and the second candidate frequency band, and the centers of the good cluster frequency band clusters correspond to The frequency band is finally used as the frequency band for palmprint recognition. 2. method as claimed in claim 1, is characterized in that, described rejecting in the palmprint image to be identified corresponds to the less frequency band of information, obtains the first alternative frequency band, comprising: Calculate the gray distribution value of the palmprint image; According to the gray distribution value, calculate the entropy corresponding to each frequency band in the palmprint image, and use the frequency band with a larger entropy value as the first candidate frequency band. 3. method as claimed in claim 1, is characterized in that, described based on Gabor filter selects from described palmprint image corresponding to the lower frequency band of equal error rate as the second alternative frequency band, comprising: Filtering the palmprint image based on a Gabor filter; For the filtered palmprint image, calculate the equal error rate of each frequency band, and use the corresponding frequency band whose equal error rate is less than a preset threshold as the second candidate frequency band. 4. method as claimed in claim 3, is characterized in that, described based on Gabor filter to described palmprint image filtering, comprises: based on the Gabor filter that function G (x, y, θ, u, σ) represents , the palmprint image is filtered in six directions of the function G (x, y, θ, u, σ), and the function G (x, y, θ, u, σ) is defined as follows: $G\left(x,\mathrm{the\; y},\mathrm{\θ},u,\mathrm{\σ}\right)=\frac{1}{2{\mathrm{\π\σ}}^2}\exp \left\{-\frac{x^2+{\mathrm{the\; y}}^2}{2{\mathrm{\σ}}^2}\right\}*\exp \left\{2\mathrm{\π}i\left(ux\cos \mathrm{\θ}+u\mathrm{the\; y}\sin \mathrm{\θ}\right)\right\},$ said The x and y represent the position coordinates of the Gabor filter, the u is the frequency of the sine wave, the θ is the direction of the function G(x, y, θ, u, σ), and the σ is a Gaussian function Variance. 5. method as claimed in claim 4 is characterized in that, six directions of described function G (x, y, θ, u, σ) are respectively 0, with 6. A frequency band selection device for palmprint identification, characterized in that said device comprises: The first selection module is used to remove frequency bands corresponding to less information in the palmprint image to obtain the first candidate frequency band; The second selection module is used to select a frequency band corresponding to a lower equal error rate from the palmprint image based on the Gabor filter as the second candidate frequency band; A clustering module, configured to calculate several well-clustered frequency band clusters from the overlapping portion of the first candidate frequency band and the second candidate frequency band according to the k-clustering algorithm, and the clustered good frequency bands The frequency band corresponding to the center of the cluster is finally used as the frequency band for palmprint recognition. 7. The device according to claim 6, wherein the first selection module comprises: A grayscale calculation unit, used to calculate the grayscale distribution value of the palmprint image; The entropy calculation unit is configured to calculate the entropy corresponding to each frequency band in the palmprint image according to the gray distribution value, and use the frequency band with a larger entropy value as the first candidate frequency band. 8. The device according to claim 6, wherein the second selection module comprises: A filtering unit, for filtering the palmprint image based on a Gabor filter; The equal error rate calculation unit is used to calculate the equal error rate of each frequency band for the filtered palmprint image, and use the corresponding frequency band whose equal error rate is less than a preset threshold as the second candidate frequency band. 9. The device according to claim 8, wherein the filtering unit is specifically used for a Gabor filter based on a function G (x, y, θ, u, σ), and in the function G (x, y, θ, u, σ) to the palmprint image filtering in six directions, the function G (x, y, θ, u, σ) is defined as follows: $G\left(x,\mathrm{the\; y},\mathrm{\θ},u,\mathrm{\σ}\right)=\frac{1}{2{\mathrm{\π\σ}}^2}\exp \left\{-\frac{x^2+{\mathrm{the\; y}}^2}{2{\mathrm{\σ}}^2}\right\}*\exp \left\{2\mathrm{\π}i\left(ux\cos \mathrm{\θ}+u\mathrm{the\; y}\sin \mathrm{\θ}\right)\right\},$ said The x and y represent the position coordinates of the Gabor filter, the u is the frequency of the sine wave, the θ is the direction of the function G(x, y, θ, u, σ), and the σ is a Gaussian function Variance. 10. The device according to claim 9, wherein the six directions of the function G (x, y, θ, u, σ) are respectively 0, with