CN113963387A - Finger multi-modal feature extraction and fusion method based on optimal coding bit - Google Patents

Finger multi-modal feature extraction and fusion method based on optimal coding bit Download PDF

Info

Publication number
CN113963387A
CN113963387A CN202111186478.4A CN202111186478A CN113963387A CN 113963387 A CN113963387 A CN 113963387A CN 202111186478 A CN202111186478 A CN 202111186478A CN 113963387 A CN113963387 A CN 113963387A
Authority
CN
China
Prior art keywords
finger
image
mode
fusion
coding
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111186478.4A
Other languages
Chinese (zh)
Other versions
CN113963387B (en
Inventor
杨玉清
杨金锋
薛月菊
李树一
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China Agricultural University
Original Assignee
South China Agricultural University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China Agricultural University filed Critical South China Agricultural University
Priority to CN202111186478.4A priority Critical patent/CN113963387B/en
Priority claimed from CN202111186478.4A external-priority patent/CN113963387B/en
Publication of CN113963387A publication Critical patent/CN113963387A/en
Application granted granted Critical
Publication of CN113963387B publication Critical patent/CN113963387B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Abstract

A finger multi-mode image coding and fusion method based on optimal coding bits is provided. The method comprises the steps of enhancing an original finger three-mode image to obtain a finger three-mode enhanced image; carrying out binary coding on the optimal enhancement direction of the finger three-mode enhanced image by using a direction coding method so as to extract effective texture features of the finger three-mode and obtain finger three-mode feature codes; fusing the finger three-mode feature codes by using a feature code fusion method to obtain a final fusion image and the like. The invention has the following effects: effectively highlighting the imaging area of the finger blood vessel and realizing the stable enhancement of the degraded finger image. The problem of redundant information and redundant feature encoding bits that may be generated is solved in order to extract features. The finger tri-modal information can be fully utilized, and the accuracy and robustness of recognition are improved.

Description

Finger multi-modal feature extraction and fusion method based on optimal coding bit
Technical Field
The invention belongs to the technical field of finger multi-modal image recognition, and particularly relates to a finger multi-modal image coding and fusing method based on an optimal coding bit.
Background
With the coming of information age and the rapid development of computer technology, information security has become the premise of social security. At present, products adopting single-mode characteristics for identity authentication are widely applied. In the application of single-mode biological feature recognition, the recognition performance is easily hindered by intra-class variation and deception attack, and is easily limited by actual acquisition conditions and environments, so that the high-performance identity authentication requirement of people in daily life cannot be met. Multimodal biometric identification is always superior to single modality methods in terms of versatility, accuracy and security. The multi-mode fusion can extract complementary and common characteristics among a plurality of modes, more comprehensively and finely describe the characteristic information of the main body, and improve the stability and the safety of the identity recognition system. Therefore, multi-modal recognition is following the development of the trend of the times and becomes an important direction of the current research.
Among the numerous biometric combinations, hand features are of particular interest for finger-based multimodal identity authentication techniques due to their high flexibility and user acceptance. The fingerprint, vein and knuckle print of finger all have very high specificity, and through the combination between the characteristic, can reach the recognition accuracy that far surpasses single characteristic. In addition, the characteristics are concentrated on the finger position, can be uniformly collected, has low requirements on equipment, has low application cost and is easy to accept by users, thereby being beneficial to the technology to quickly realize the productization.
The traditional finger multi-modal feature expression method is insensitive to illumination and has no gray scale invariance and rotation invariance. However, the coding-based feature representation approach provides higher performance in terms of illumination invariance, feature description capability, and feature matching efficiency. Therefore, it becomes a key issue in research to search a robust feature encoding method which is insensitive to illumination variation and has high recognition accuracy. In addition, existing multimodal fusion methods result in large memory space, and these feature expression and fusion methods do not adequately consider the distinguishing features of fingers and do not result in satisfactory recognition performance. Therefore, the method for expressing the characteristics with robustness has great value for improving the recognition performance of the finger multi-mode characteristic fusion system.
Disclosure of Invention
In order to solve the above problems, the present invention aims to provide a finger multimodal image encoding and fusion method based on optimal encoding bits.
In order to achieve the above object, the finger multimodal image coding and fusion method based on the optimal coding bits provided by the present invention comprises the following steps in sequence:
1) enhancing the original finger three-mode image to obtain a finger three-mode enhanced image;
2) carrying out binary coding on the optimal enhancement direction of the finger three-mode enhanced image by using a direction coding method so as to extract effective texture features of the finger three-mode and obtain finger three-mode feature codes;
3) fusing the finger three-mode feature codes obtained in the step 2) by using a feature code fusion method to obtain a final fused image.
In step 1), the method for enhancing the original finger three-modality image to obtain the finger three-modality enhanced image includes: firstly, carrying out Gabor filtering on an original finger three-mode image by using a multi-scale, directional and tolerable Gabor filter group; then, establishing directional Weber differential excitation for the filtered image on the basis of Weber's law; and finally, under the multi-scale condition, obtaining an image with strongest scale response and strongest direction response, namely the finger three-modal enhanced image.
In step 2), the method for performing binary coding on the optimal enhancement direction of the finger tri-modal enhanced image by using a direction coding method to extract effective texture features of the finger tri-modal, and obtaining the finger tri-modal feature code includes:
first, 8 directions of a current pixel are represented using 8-bit binary coding; then, defining a main direction for the current pixel, and aligning the optimal enhancement direction with the main direction; then, only comparing the neighborhood value in the optimal enhancement direction of the finger three-modality enhanced image with the size of the current pixel value to obtain a binary code; if the neighborhood value is larger than the current pixel value, setting the binary code between two pixels as 1, otherwise, setting the binary code as 0; the binary code of the current pixel is constantly 1, and the values of other binary code bits are all set to be 0, so that the finger tri-modal feature code is obtained.
In step 3), the method for fusing the finger three-mode feature codes obtained in step 2) by using the feature code fusion method to obtain the final fusion image includes: firstly, defining a main direction of finger three-mode feature coding fusion; aligning the main direction in the direction coding method in the step 2) with the main direction of the finger three-modal feature code fusion, and then fusing according to the fusion sequence by taking the main direction of the finger three-modal feature code fusion as the center to obtain the final fusion image.
The finger multi-mode image coding and fusing method based on the optimal coding bit has the following beneficial effects:
1. the invention provides a finger vein blood vessel region stability enhancement method fusing the Weber's law and Gabor filtering, which solves the problems of low imaging region quality and the like caused by the existence of noise and redundant information on the edge of a captured original finger three-mode image. The directional excitation capability of the Weber local descriptor is amplified through the multi-scale and multi-directional characteristics of Gabor filtering, and the optimal response of the Gabor filtering and the optimal excitation of the Weber law are matched with each other, so that the finger vein imaging area is effectively highlighted, and the stable enhancement of the degraded finger vein image is realized. Experimental results show that the method is also suitable for fingerprints and knuckle prints.
2. A direction coding method based on a Weber local descriptor is provided. Under the multi-scale condition, the Weber excitation response has the characteristics of strongest scale response and strongest direction response, so that the optimal enhancement direction is coded, and the problems of redundant information (including noise) and redundant feature coding bits which may be generated are solved, so that the features are extracted. The optimal direction coding can adopt two coding methods of directivity and non-directivity.
3. A feature code fusion method is provided. The method defines the main direction, aligns the direction position of the three modes of the finger with the main direction, arranges the feature coding information of the three modes of the finger one by one, can fully utilize the information of the three modes of the finger and improve the accuracy and the robustness of identification.
Drawings
Fig. 1 is a schematic diagram of the 3 × 3 neighborhood of WLD.
Fig. 2 is a schematic diagram of three different neighborhoods of a WLD.
Fig. 3 is an example of a partial original finger vein image and a finger vein enhanced image. (a) An original finger vein image; (b) n is 3; (c) n is 5; (d) n is 7.
Fig. 4 is an example of a finger tri-modal enhanced image. (a) A finger vein enhancement image; (b) a knuckle print enhanced image; (c) the fingerprint enhances the image.
Fig. 5 is a schematic diagram of a 5 × 5 neighborhood of the current pixel W.
xc
Fig. 6 is an encoding example.
FIG. 7 is a schematic diagram of a finger tri-modal feature encoding fusion method; (a) directionality; (b) and no directivity.
FIG. 8 is a three-modal image of an original finger and a three-modal feature-coded fused image of a finger, wherein (a) the three-modal image of the original finger and (b) the three-modal feature-coded fused image of a directional finger; (c) and encoding the fused image by using the three-modal characteristic of the non-directional finger.
FIG. 9 is a comparison of recognition performance of different feature expression methods; (a) ROC curves for different feature expression methods; (b) 10 test accuracy curves for different feature expression methods.
FIG. 10 is a ROC curve for different sequencing methods.
FIG. 11 is a comparison of recognition performance for different fusion methods; (a) ROC curves for different fusion methods; (b) 10 test accuracy curves for different fusion methods.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments.
The finger multi-modal image coding and fusion method based on the optimal coding bit comprises the following steps in sequence:
1) enhancing the original finger three-mode image to obtain a finger three-mode enhanced image;
compared with the global descriptor, the image content expression and enhancement method of the local descriptor is good at capturing local changes of the image content, and is beneficial to describing local characteristics of the target. In order to overcome the limitations of the existing Local descriptors in the aspects of illumination change, rotation transformation, single scale and the like, the step adopts a Local feature Descriptor (MGWLD) with good generalization capability to enhance the original finger three-mode image. MGWLD organically integrates the neighborhood characteristics of a Weber Local Descriptor (WLD) and the multi-scale and multi-direction characteristics of a Gabor filter, and effectively gives consideration to the direction randomness of a finger vascular network and the local direction expression capability of the WLD.
WLD is characterized by being simple and efficient and robust to illumination variations. As shown in fig. 1, WLD may use a 3 × 3 neighborhood filter to calculate the gray-level value of the current pixel, and the calculation formula of the original differential excitation strength is shown in equation (1):
Figure BDA0003299438270000051
where I denotes the original excitation intensity, Δ I denotes the excitation difference,
Figure BDA0003299438270000056
representing spatial coordinates x, y]，
Figure BDA0003299438270000052
Representing a current pixel
Figure BDA0003299438270000053
P represents the number of neighborhood pixels. If the original differential excitation strength
Figure BDA0003299438270000054
Indicating that the neighborhood pixel value is greater than the current pixel value; if the original differential excitation strength
Figure BDA0003299438270000055
Indicating that the current pixel is of lower brightness in this region. In order to prevent the input value from being too large or too small in the calculation process, the output is mapped into a reasonable value range by utilizing an arctan (·) function so as to prevent the ratio from being too large, and therefore the influence of partial noise can be inhibited.
The original differential excitation intensity will be poor
Figure BDA0003299438270000061
Considering a scalar, summing up only the gray differences in eight directions is essentially an isotropic laplacian, resulting in insufficient applicability to gray change information and sensitivity to image noise. For finger vein vessel regions with more directional image content, the enhancement effect of the current pixel should be closely related to the directional excitation. Therefore, simply representing the differential component as a scalar quantity is not favorable for the directional enhancement of the finger vein image.
In order to calculate the directional difference of the differential excitation, the calculation formula of the directional differential excitation strength is redefined as shown in formula (2):
Figure BDA0003299438270000062
wherein, thetakRepresenting the k-th direction of the WLD neighborhood,
Figure BDA0003299438270000063
representing a current pixel
Figure BDA0003299438270000064
Differential excitation strength in the k-th direction. Because the finger vein image is seriously degraded and the blood vessel trend is not obvious, the formula (2) is used for depicting the directional excitation on the basis of the original pixel informationThe effect is not ideal in terms of difference. Therefore, in order to highlight the directionality of the blood vessel network, the original finger vein image needs to be subjected to directional filtering.
In the aspect of enhancement of finger vein blood vessel region, the Gabor function has outstanding performance as a directional filter. The Weber law and the Gabor filter are organically combined, and the method has important value for stably enhancing the finger vein area. In the k-th direction theta of WLD neighborhoodkAbove, the finger vein image after Gabor filtering is:
Figure BDA0003299438270000065
wherein the content of the first and second substances,
Figure BDA0003299438270000066
representing a directional multi-scale Gabor filter bank, symbol
Figure BDA0003299438270000067
Which represents a 2D convolution of the image,
Figure BDA0003299438270000068
representing the original finger vein image,
Figure BDA0003299438270000069
representing a Gabor filtered finger vein image. Since the original Gabor wavelet is not tolerable, the Gabor wavelet needs to be made tolerable in directionality in order to weaken the response deviation caused by the illumination change of the image. A multi-scale, directional, and tolerable Gabor filter bank is defined as:
Figure BDA0003299438270000071
Figure BDA0003299438270000072
wherein m represents the scale change of the Gabor wavelet,
Figure BDA0003299438270000073
Δφ∈[1,1.5]the bandwidth indicates the bandwidth of the octave at half maximum, and a ═ diag [1, vsin (pi/16) (2ln2)-0.5]) Represents a 2 x 2 diagonal matrix, embodying the anisotropy of the Gabor wavelet,
Figure BDA0003299438270000074
representing the center frequency, σ, of the complex exponentialm(m ═ 1,2, or 3) corresponds to the scale of the Gabor wavelet, θkDenotes the K-th direction of the Gabor wavelet, which is the same as the K-th direction of the WLD neighborhood expressed by equation (2), where K is 1,2,3, …, K is 2(n-1), and n is 3, 5, or 7, and denotes the neighborhood size of the WLD. When n is 3, the neighborhood of WLD expresses 4 directions, when n is 5, 8 directions, and when n is 7, 12 directions, as shown in fig. 2.
Combining equation (3) and equation (4), a new calculation equation of WLD differential excitation strength can be obtained as shown in equation (6):
Figure BDA0003299438270000075
because the filtering response is strongest when the main excitation lobe of the Gabor wavelet is locally vertical to the ridge line of the finger vein, the new WLD differential excitation strength is obtained at the moment
Figure BDA0003299438270000076
Should also perform optimally. Thus, at some scale m, the optimal function for a finger vein enhancement image is:
Figure BDA0003299438270000077
thus, by using equation (7), taking the scale m as 1, a finger vein enhanced image with an optimal single-scale direction can be obtained, as shown in fig. 3. As can be seen from fig. 3, the MGWLD has a very significant effect of enhancing the finger vein region, and especially under the condition of 7 × 7 neighborhood, the calculation of combining the Gabor filtering in 12 directions and the directional excitation of weber effectively suppresses noise, and has a very significant effect of enhancing the main blood vessels of the finger veins.
Since the diameter variation of the finger vein in a neighborhood is random, the Gabor wavelet has the multi-scale characteristic, and the optimal filtering response can be obtained, as shown in formulas (3) and (4). Therefore, in the multi-scale case, considering that the weber excitation response should have the strongest scale response and the strongest direction response at the same time, the enhancement function of the finger vein image is:
Figure BDA0003299438270000081
therefore, with equation (8), a finger vein enhanced image can be obtained; similarly, a fingerprint and knuckle-print enhanced image, i.e., a finger tri-modal enhanced image, may be obtained, as shown in fig. 4.
2) Carrying out binary coding on the optimal enhancement direction of the finger three-mode enhanced image by using a direction coding method so as to extract effective texture features of the finger three-mode and obtain finger three-mode feature codes;
taking the 5 x 5 neighborhood of WLD as an example, it defines 8 directions. The 8 directions of the current pixel are represented by using 8-bit binary coding to obtain efficient texture features. The main idea of the direction coding method is as follows: the binary encoding is obtained by comparing only the neighborhood value in the optimal enhancement direction with the magnitude of the current pixel value. The binary encoding between two pixels is set to 1 if the neighborhood value is greater than the current pixel value, and is set to 0 otherwise. The directional encoding method can improve recognition performance and reduce time cost compared to other encoding methods. In addition, the above direction encoding method includes two kinds: (1) a directional coding method; (2) a non-directional coding method. The specific process is as follows:
(1) directional coding method
In order to distinguish the binary encoding of the optimal enhancement direction of the current pixel, a main direction is defined for the current pixel, and any direction can be defined as a main direction. The present invention defines the direction 5 as the principal direction, as shown in fig. 5. The specific encoding process is as follows:
when 1 < thetakWhen the ratio is less than 8, the reaction solution is,
Figure BDA0003299438270000091
wherein the content of the first and second substances,
Figure BDA0003299438270000092
Figure BDA0003299438270000093
wherein, thetak(-1, 2,3 … 8) represents the optimal direction of enhancement of the image,
Figure BDA0003299438270000094
representing the enhanced current pixel. As can be clearly seen from figure 5 of the drawings,
Figure BDA0003299438270000095
and
Figure BDA0003299438270000096
respectively representing two different pixels in the same direction in a 5 x 5 neighborhood of the WLD. TQ and TP respectively represent the difference between two adjacent pixel values on both sides of the main direction of the optimal enhancement direction and the current pixel value.
Figure BDA0003299438270000097
And ccRepresenting the binary encoding of the current pixel. c. CklAnd ckrRespectively represent the optimal enhancement directions thetakTwo different binary encodings in the WLD neighborhood. Optimum direction of enhancement thetakThe binary code of (a) is always 1. When 1 is<θk<At time 8, its binary encoding depends on the size of the TP and TQ. If TP>0, then binary code cklIs 1, similarly, if TQ>0, then binary code ckrIs 1. For 8-bit binary codingIn other words, θ k1 and θkThe special direction is 8. Therefore, given that the optimal enhancement direction is 1, the binary encoding of the current pixel needs to be shifted by one bit to the high bits in the binary encoding. Given an optimal enhancement direction of 8, the binary encoding of the current pixel needs to be shifted one bit to the lower bits in the binary encoding. Then according to 1<θk<The rule at 8 is calculated. The current pixel is in the optimal enhancement direction thetakFeature code of (a) F (θ)k) That is, the binary codes are summed, and the calculation formula is shown in equation (12):
Figure BDA0003299438270000101
fig. 6 is an encoding example. Assuming that the optimal enhancement direction of the current pixel is 3, the feature code of the current pixel can be obtained by using the formula (9) - (12). Similarly, for encoding in other directions, the same is true for the calculation process.
(2) Non-directional coding method
The above definition of the main direction also applies to the non-directional coding method. It is noted that the non-directional encoding method is also dominated by direction 5. The encoding calculation process is as follows:
Figure BDA0003299438270000102
wherein the content of the first and second substances,
Figure BDA0003299438270000103
unlike the directional coding method, in the non-directional coding method, the binary coding of the main direction 5 is always 1. If TQ > 0, binary code c6Is 1. If TP > 0, binary code c4Is 1. Otherwise the binary code is 0.
Feature code F (theta) of current pixel at this timek) The calculation formula is shown in formula (16):
F(θk)＝c4×23+c5×24+c6×25。 (16)
the main difference between the two coding methods is that the optimal enhancement direction of the directional coding method always corresponds to 8-bit binary coding, while the optimal enhancement direction of the non-directional coding method always aligns to the main direction, thus reducing the problem of imbalance of binary coding caused by direction. Other coding rules of the non-directional coding method are the same as those of the directional coding method.
3) Fusing the finger three-mode feature codes obtained in the step 2) by using a feature code fusion method to obtain a final fused image.
The main idea of the step is as follows: (1) since the 5 × 5 neighborhood of WLDs has 8 directions, 8-bit binary coding ([ c1, c2, c3, c4, c5, c6, c7, c8]) is also used to represent the fused information of the finger trimodality. (2) In the binary encoding process of step 2), the direction 5 is taken as the main direction. This step still uses the direction 5 as the main direction for finger tri-modal feature code fusion. (3) In the feature coding fusion process, according to the rules of permutation and combination, the three modes of finger veins, fingerprints and knuckle prints can generate six fusion orders, namely: FV, FP, FKP; FV, FKP, FP; FP, FV, FKP; FP, FKP, FV; FKP, FV, FP; FKP, FP, FV. Taking the first fusion ordering (FV, FP, FKP) as an example, (c4, c6), (c3, c7), (c2, c8) represent feature codes on the left and right sides of the current pixel of the finger vein, fingerprint, and knuckle print, respectively.
The specific fusion method comprises the following steps: (1) when the directional finger tri-modal feature codes are fused, as shown in fig. 7(a), the first line and the second line are finger tri-modal codes according to the coding method of step 2). In the directional coding method, the direction 1 and the direction 8 are special directions, and one bit needs to be moved to the left or the right. And finally, aligning the optimal enhancement direction of each mode with the main direction in the multi-mode fusion frame, and fusing the codes on the left side and the right side of the multi-mode front pixels by taking the main direction as the center according to the fusion sequencing to obtain a final fusion image.
(2) When the non-directional finger tri-modal feature encoding is fused, as shown in fig. 7(b), in the non-directional encoding method, all the modalities are encoded centering on the principal direction 5. Therefore, during fusion, the main direction of the codes is aligned with the main direction in the multi-modal fusion framework, and the codes on the left side and the right side of the current pixel are fused by taking the main direction as the center according to the fusion sequence to obtain the final fusion image.
In fig. 8, an original finger tri-modal image, (b) a directional finger tri-modal feature encoding fused image; (c) and encoding the fused image by using the three-modal characteristic of the non-directional finger. Obviously, the texture features of the fused image obtained by the non-directional coding method are relatively clearer. This is because for directional coding, some directions always correspond to high coded bits in binary coding, which corresponds to a larger binary coding. In contrast, other directions always correspond to low coded bits in the binary, which corresponds to a smaller binary. This phenomenon can produce unbalanced binary codes and asymmetric coded images. In addition, unclear image features will be detrimental to biometric identification. Therefore, when the finger three-modal feature coding is fused, the invention preferentially uses a non-directional coding method to carry out binary coding on the optimal enhancement direction of each single-modal enhanced image, and then carries out multi-modal feature fusion on the coded result.
In order to fully demonstrate the feasibility and effectiveness of the method, the inventor carried out a series of experiments using a finger tri-modal database created in a laboratory for evaluating the recognition performance of the method. This database contains 585 classes, and 10 finger vein, fingerprint, and knuckle images are collected in the finger classes as the original finger tri-modal images, i.e., 5850 finger vein images, 5850 fingerprint images, and 5850 knuckle images, respectively. And all the original three-mode images of the fingers in the self-made database are normalized into 80 × 180 pixels, the experimental environment is a PC, and the method is completed in the Matlab R2018a environment.
The recognition capability of the feature coding fusion method provided by the invention applied to the finger single-mode features is compared with several commonly used single-mode feature expression methods (LGS, SLGS, LBP, LLBP, Compcode and SCW-LGS). FIG. 9 shows the results of experiments with different expression methods on a monomodal database (finger vein), EER, AVE (with STD) and time for single feature extraction of the different methods are shown in Table 1.
TABLE 1 identification results of different feature expression methods
Figure BDA0003299438270000121
Figure BDA0003299438270000131
According to the rule of permutation and combination, the finger veins, fingerprints and knuckle prints generate six kinds of fusion ordering in the feature code fusion process: (FV, FP, FKP), (FV, FKP, FP), (FP, FV, FKP), (FP, FKP, FV), (FKP, FV, FP), (FKP, FP, FV). In order to determine the most suitable fusion sorting method to fuse the finger tri-modal feature codes, the recognition performances of the 6 methods are compared, and the ROC curve is shown in fig. 10.
From the ROC curve, the identification performance of the six different fusion sorting methods is the same, which indicates that the identification performance of the fusion of the three-modal feature codes is irrelevant to the fusion sorting method of the single-modal feature codes. Namely, the feature coding fusion method provided by the invention has robustness for finger three-mode feature coding fusion. Therefore, in the finger tri-modal feature code fusion experiment, any feature code sorting method can be used.
The proposed FEF-OBC encoding fusion algorithm is compared with the fusion methods (loop granularity, triangle granularity, GOM, graph _ code _ fusion, MRRID) proposed in the last few years. FIG. 11 shows the results of different fusion methods. Table 2 shows the EER, STD and feature extraction times for individual pictures for different fusion methods. It can be seen that the EER result of the FEF-OBC proposed by the present invention is the lowest among the six fusion methods, and the average accuracy AVE of the FEF-OBC proposed by the present invention is about 99.93% closer to the AVE of the graph _ code _ fusion, but the STD proposed by the present invention is much lower than the graph _ code _ fusion, which indicates that the FEF-OBC coding fusion method proposed by the present invention is more stable. In addition, compared with other methods, the method has the highest identification efficiency, and the time spent on single identification is less. In conclusion, the FEF-OBC multi-modal coding fusion algorithm provided by the invention is more accurate and has the highest recognition efficiency. Therefore, the reliability of the fused feature recognition of the finger tri-modal FV, FP and FKP can be significantly improved by using the proposed algorithm.
TABLE 2 recognition results of different fusion methods
Figure BDA0003299438270000141

Claims (4)

1. A finger multi-mode image coding and fusion method based on optimal coding bits is characterized in that: the method comprises the following steps performed in sequence:
1) enhancing the original finger three-mode image to obtain a finger three-mode enhanced image;
2) carrying out binary coding on the optimal enhancement direction of the finger three-mode enhanced image by using a direction coding method so as to extract effective texture features of the finger three-mode and obtain finger three-mode feature codes;
3) fusing the finger three-mode feature codes obtained in the step 2) by using a feature code fusion method to obtain a final fused image.
2. The optimal encoding bit-based finger multimodal image encoding and fusion method according to claim 1, wherein: in step 1), the method for enhancing the original finger three-modality image to obtain the finger three-modality enhanced image includes: firstly, carrying out Gabor filtering on an original finger three-mode image by using a multi-scale, directional and tolerable Gabor filter group; then, establishing directional Weber differential excitation for the filtered image on the basis of Weber's law; and finally, under the multi-scale condition, obtaining an image with strongest scale response and strongest direction response, namely the finger three-modal enhanced image.
3. The optimal encoding bit-based finger multimodal image encoding and fusion method according to claim 1, wherein: in step 2), the method for performing binary coding on the optimal enhancement direction of the finger tri-modal enhanced image by using a direction coding method to extract effective texture features of the finger tri-modal, and obtaining the finger tri-modal feature code includes:
first, 8 directions of a current pixel are represented using 8-bit binary coding; then, defining a main direction for the current pixel, and aligning the optimal enhancement direction with the main direction; then, only comparing the neighborhood value in the optimal enhancement direction of the finger three-modality enhanced image with the size of the current pixel value to obtain a binary code; if the neighborhood value is larger than the current pixel value, setting the binary code between two pixels as 1, otherwise, setting the binary code as 0; the binary code of the current pixel is constantly 1, and the values of other binary code bits are all set to be 0, so that the finger tri-modal feature code is obtained.
4. The optimal encoding bit-based finger multimodal image encoding and fusion method according to claim 1, wherein: in step 3), the method for fusing the finger three-mode feature codes obtained in step 2) by using the feature code fusion method to obtain the final fusion image includes: firstly, defining a main direction of finger three-mode feature coding fusion; aligning the main direction in the direction coding method in the step 2) with the main direction of the finger three-modal feature code fusion, and then fusing according to the fusion sequence by taking the main direction of the finger three-modal feature code fusion as the center to obtain the final fusion image.
CN202111186478.4A 2021-10-12 Finger multi-mode feature extraction and fusion method based on optimal coding bits Active CN113963387B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111186478.4A CN113963387B (en) 2021-10-12 Finger multi-mode feature extraction and fusion method based on optimal coding bits

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111186478.4A CN113963387B (en) 2021-10-12 Finger multi-mode feature extraction and fusion method based on optimal coding bits

Publications (2)

Publication Number Publication Date
CN113963387A true CN113963387A (en) 2022-01-21
CN113963387B CN113963387B (en) 2024-04-19

Family

ID=

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4562467A (en) * 1983-04-01 1985-12-31 Itek Corporation Data compression apparatus and method for encoding and decoding multi-line signals
US6741748B1 (en) * 1997-06-19 2004-05-25 Samsung Electronics Co. Scan interleaving method based on context values of neighboring pixels, improved encoding/decoding method of binary shape data and encoding/decoding apparatus suitable for the same
CN1667639A (en) * 2004-03-10 2005-09-14 Lg电子有限公司 Binary coding method of fingerprint image
US20070287218A1 (en) * 2006-06-12 2007-12-13 Stmicroelectronics (Research And Development) Limited Digital filter
US20080205785A1 (en) * 2005-11-23 2008-08-28 Cedara Software Corp. Method and system for enhancing digital images
CN103310411A (en) * 2012-09-25 2013-09-18 中兴通讯股份有限公司 Image local reinforcement method and device
CN109902585A (en) * 2019-01-29 2019-06-18 中国民航大学 A kind of three modality fusion recognition methods of finger based on graph model
CN109993142A (en) * 2019-04-10 2019-07-09 黑龙江大学 Two dimensional code identity identifying method based on finger portion multi-modal biological characteristic
CN111353472A (en) * 2020-03-30 2020-06-30 蚌埠学院 Finger vein identification method based on LBP and Gabor direction Weber local descriptor
CN111507206A (en) * 2020-03-29 2020-08-07 杭州电子科技大学 Finger vein identification method based on multi-scale local feature fusion
US11068693B1 (en) * 2020-12-29 2021-07-20 King Abdulaziz University Liveness detection in fingerprint-based biometric systems

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4562467A (en) * 1983-04-01 1985-12-31 Itek Corporation Data compression apparatus and method for encoding and decoding multi-line signals
US6741748B1 (en) * 1997-06-19 2004-05-25 Samsung Electronics Co. Scan interleaving method based on context values of neighboring pixels, improved encoding/decoding method of binary shape data and encoding/decoding apparatus suitable for the same
CN1667639A (en) * 2004-03-10 2005-09-14 Lg电子有限公司 Binary coding method of fingerprint image
US20080205785A1 (en) * 2005-11-23 2008-08-28 Cedara Software Corp. Method and system for enhancing digital images
US20070287218A1 (en) * 2006-06-12 2007-12-13 Stmicroelectronics (Research And Development) Limited Digital filter
CN103310411A (en) * 2012-09-25 2013-09-18 中兴通讯股份有限公司 Image local reinforcement method and device
CN109902585A (en) * 2019-01-29 2019-06-18 中国民航大学 A kind of three modality fusion recognition methods of finger based on graph model
CN109993142A (en) * 2019-04-10 2019-07-09 黑龙江大学 Two dimensional code identity identifying method based on finger portion multi-modal biological characteristic
CN111507206A (en) * 2020-03-29 2020-08-07 杭州电子科技大学 Finger vein identification method based on multi-scale local feature fusion
CN111353472A (en) * 2020-03-30 2020-06-30 蚌埠学院 Finger vein identification method based on LBP and Gabor direction Weber local descriptor
US11068693B1 (en) * 2020-12-29 2021-07-20 King Abdulaziz University Liveness detection in fingerprint-based biometric systems

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
杨玉清: "手指多模态图像方向性增强和编码方法研究", 17 February 2022 (2022-02-17) *
杨玉清等: ""融合Weber定律与Gabor滤波的手指静脉图像增强方法"", 《激光与光电子学进展》, vol. 58, no. 4, 31 July 2021 (2021-07-31), pages 1410003 - 1 *
王华彬等: "基于双Gabor方向韦伯局部描述子的掌纹识别", 电子与信息学报, no. 04, 15 April 2018 (2018-04-15), pages 936 - 943 *

Similar Documents

Publication Publication Date Title
Xin et al. Multimodal feature-level fusion for biometrics identification system on IoMT platform
Pietikäinen et al. Two decades of local binary patterns: A survey
Pflug et al. Ear biometrics: a survey of detection, feature extraction and recognition methods
Sotoodeh et al. A novel adaptive LBP-based descriptor for color image retrieval
Li et al. Local discriminant coding based convolutional feature representation for multimodal finger recognition
Bakshi et al. A novel phase-intensive local pattern for periocular recognition under visible spectrum
Shu et al. Multiple channels local binary pattern for color texture representation and classification
CN108596126A (en) A kind of finger venous image recognition methods based on improved LGS weighted codings
CN108509927A (en) A kind of finger venous image recognition methods based on Local Symmetric graph structure
Fei et al. Complete binary representation for 3-D palmprint recognition
He et al. Iris feature extraction method based on LBP and chunked encoding
Yang et al. Copy-move forgery detection based on adaptive keypoints extraction and matching
Shah et al. Image splicing forgery detection using DCT coefficients with multi-scale LBP
Zhou et al. Fast minutiae extractor using neural network
Patel et al. An improved approach in fingerprint recognition algorithm
El-Abed et al. Quality assessment of image-based biometric information
CN109977909B (en) Finger vein identification method and system based on minutia area matching
Kumar et al. Non-overlapped blockwise interpolated local binary pattern as periocular feature
Rana et al. MSRD-CNN: Multi-scale residual deep CNN for general-purpose image manipulation detection
Yee et al. Face recognition using Laplacian completed local ternary pattern (LapCLTP)
Nachar et al. Hybrid minutiae and edge corners feature points for increased fingerprint recognition performance
Shao et al. A deep biometric hash learning framework for three advanced hand‐based biometrics
CN113963387B (en) Finger multi-mode feature extraction and fusion method based on optimal coding bits
CN113963387A (en) Finger multi-modal feature extraction and fusion method based on optimal coding bit
CN113963164A (en) Texture feature extraction method based on grouping neighborhood intensity difference coding

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant