CN105095880B - A kind of multi-modal Feature fusion of finger based on LGBP coding - Google Patents
A kind of multi-modal Feature fusion of finger based on LGBP coding Download PDFInfo
- Publication number
- CN105095880B CN105095880B CN201510518806.4A CN201510518806A CN105095880B CN 105095880 B CN105095880 B CN 105095880B CN 201510518806 A CN201510518806 A CN 201510518806A CN 105095880 B CN105095880 B CN 105095880B
- Authority
- CN
- China
- Prior art keywords
- finger
- gray
- image
- value
- feature
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/12—Fingerprints or palmprints
- G06V40/1347—Preprocessing; Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/12—Fingerprints or palmprints
- G06V40/1365—Matching; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
A kind of multi-modal Feature fusion of finger based on LGBP coding.It carries out Gabor filtering to three mode ROI image of finger including the use of Gabor filter, obtains amplitude characteristic image；Image is encoded, feature coding image is formed；Piecemeal is carried out to image；It regards the pixel of block image as feature point extraction its gray feature, forms gray feature vector；Gray feature vector is superimposed, forms gray feature histogram, then connect gray feature histogram to form the gray feature histogram of block image；Finger single mode gray feature histogram fused in tandem is formed into three mode gray feature histogram of finger；The gray feature histogram intersection coefficient of two three mode ROI images of finger to be matched is calculated to judge whether the two matches.The present invention efficiently solves the problems, such as the finger gesture mutability in finger-image collection process, and finger multimodal recognition arithmetic speed is high, discrimination is high.
Description
Technical field
The invention belongs to image identification technical fields, more particularly to a kind of multi-modal feature of finger based on LGBP coding
Fusion method.
Background technique
Currently, single mode living things feature recognition has some limitations in the application, therefore people are unable to satisfy to height
The demand of precision identification.To enable three modal characteristics of finger that fusion is effectively performed, robust features analysis, which becomes, is ground
Critical problem in studying carefully.Due in acquisition three mode ROI of finger (region of interest, area-of-interest) image
There is a problem of finger gesture mutability in the process, and most of finger robust features extracting method is by rotational invariance
Limitation, therefore this problem cannot be efficiently solved.
Summary of the invention
To solve the above-mentioned problems, the purpose of the present invention is to provide a kind of multi-modal features of finger based on LGBP coding
Fusion method.
In order to achieve the above object, the finger multi-modal Feature fusion provided by the invention based on LGBP coding includes
The following steps carried out in order:
1) Gabor is carried out using finger three mode ROI image of the different Gabor filter of scale parameter to different postures
Filtering, respectively obtain 8 directions, i.e., 0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and 157.5 ° of fingerprint, refer to it is quiet
The amplitude characteristic image of arteries and veins and phalangeal configurations；
2) it is encoded respectively using finger three mode amplitude characteristic image of the LBP to above-mentioned 8 directions, is consequently formed 8
The three mode LGBP feature coding image of finger in a direction；
3) piecemeal is carried out to the three mode LGBP feature coding image of finger in above-mentioned 8 directions；
4) it regards the pixel of each block image as characteristic point and extracts its gray feature, gray scale is consequently formed
Feature vector, process are as follows:
Step 1: gray scale is grouped；Firstly, the gray value of each pixel of each block image is arranged from small to large
Sequence forms the sequence of a pixel；Then, this sequence is divided into k gray scale grouping according to the sum of pixel, forms k
Group gray scale is grouped image；The boundary point of each gray scale grouping is determined with the method to round up later, and obtains the boundary point
Gray value；
Step 2: calculating the gray feature vector of each pixel: with each pixel in each gray scale grouping image
Centered on, compare the gray value size of its symmetrical adjoint point, if the gray value of some pixel is greater than the gray value of its symmetrical adjoint point,
It is then 1；Otherwise it is 0, the gray feature vector of 4 binary codes is consequently formed, then converts 16 for 4 binary code vectors
Position binary code gray feature vector；
5) the gray feature vector of each pixel in above-mentioned each gray scale grouping image is superimposed, forms each gray scale point
The gray feature histogram of group image, then connect the gray feature histogram of each gray scale grouping image to form block image
Gray feature histogram；
6) coefficient identical with LGBP feature coding image block number is generated by dimensional Gaussian model first, it is then right
The gray feature histogram of each above-mentioned block image is weighted, later that the gray scale of the block image after above-mentioned weighting is special
Sign histogram connects to obtain finger single mode gray feature histogram, finally, by above-mentioned finger single mode gray feature histogram
Fused in tandem forms three mode gray feature histogram of finger；
7) method by calculating the gray feature histogram intersection coefficient of two three mode ROI images of finger to be matched
To judge whether this two width finger ROI image matches.
In step 1), the expression formula of the Gabor filter are as follows:
Wherein, δ represents the scale of Gabor filter, δ=4, and 5,6；θkIndicate the angle value in k-th of direction.
In step 2), method that the three mode amplitude characteristic image of finger to 8 directions is encoded respectively
It is: firstly, 3 × 3 window of a pixel centered on a certain pixel in a certain finger amplitude characteristic image is defined,
Using the gray value of the central pixel point as threshold value, binaryzation is carried out to remaining 8 neighborhood territory pixel point in the window；If a certain neighborhood
The gray value of pixel is less than the gray value of central pixel point, then is encoded to 0；Otherwise, it is encoded to 1, forms 8 binary systems
Value；Then by binary value right shift b times, summation is weighted to the binary value for often moving to right one and obtains the pixel
8 LBP values；Finally, taking the smallest LBP value as the LBP value of the pixel；
The formula of minimum LBP value are as follows:
Wherein, function ROR (x, b) is indicated binary value x right shift b times,Indicate i-th of central pixel point
LBP value,Definition is as shown in formula (3):
In formula: B (Ii-Ic) indicate binaryzation function, i.e.,IiIndicate center pixel
The gray value of point i, IcIndicate that the gray value of neighborhood territory pixel point, a indicate a of binaryzation function, herein P=8.
In step 4), the gray value formula of the acquisition boundary point are as follows:
Wherein,Indicate every group of boundary point, tiIndicate the boundary of i-th of gray scale grouping
Value, IminAnd ImaxRespectively indicate the minimum gradation value and maximum gradation value of image slices vegetarian refreshments.
In step 4), the gray feature vector by 4 binary codes is converted into the gray scale of 16 binary codes
The formula of feature vector are as follows:
Wherein, i indicates that ith pixel point, m indicate the logarithm of the pixel nearest neighbor point.
In step 6), the formula of the dimensional Gaussian model are as follows:
Wherein, σ indicates the mean square deviation of dimensional Gaussian model, and m and n are the number of the image block of every row and each column respectively,
Mid (i) and mid (j) respectively represent the block image of picture centre in ith row and jth column.
In step 7), the gray feature histogram by calculating two three mode ROI images of finger to be matched
The method of figure intersection coefficient is to judge this whether matched method of two width finger ROI image: first with following intersection system
Number expression formula calculates the intersection coefficient of three mode gray feature histogram of finger in two finger ROI images to be matched, if meter
The intersection coefficient of calculating > similitude decision-making value T, then it represents that this two width finger ROI image is similar, that is, indicates this two width finger
ROI image matching；If it intersects coefficient≤T, determine that this two width finger ROI image mismatches；Similitude decision-making value T is hand
Refer to that false rejection rate is 0 in ROI image matching result, and the threshold point that mistake is corresponding when allowing rate minimum；
Intersect the expression formula of coefficient are as follows:
In formula: m1And m2Respectively indicate two finger ROI images to be matched, Hm1(i) and Hm2(i) two width are respectively represented to wait for
The gray feature histogram of matched three mode ROI image of finger, L indicate the gray feature histogram of three modality images of finger
Dimension.
Finger multi-modal Feature fusion provided by the invention based on LGBP coding is efficiently solved in finger figure
As the problem of finger gesture mutability in collection process, and the arithmetic speed of finger multimodal recognition is high, discrimination is high.
Detailed description of the invention
Fig. 1 is the three mode amplitude characteristic figure of finger in 8 directions, wherein (a) fingerprint (b) refers to vein (c) phalangeal configurations；
Fig. 2 is the three mode LGBP feature coding figure of finger in 8 directions, wherein (a) fingerprint (b) refers to vein (c) phalangeal configurations；
Fig. 3 is the three mode LGBP feature coding block diagram of finger in 8 directions, wherein (a) fingerprint (b) refers to that vein (c) refers to
Save line；
Fig. 4 is 8 nearest neighbor point schematic diagrames of pixel.
Fig. 5 is the gray feature histogram of LGBP feature coding block image, wherein (a) fingerprint (b) refers to vein (c) finger joint
Line；
Fig. 6 is the weighting cascade process schematic diagram based on dimensional Gaussian model；
Fig. 7 is that the recognition performance of 8 × 8 block image difference gray scales grouping compares；
Fig. 8 is that the recognition performance of different block images compares；
Fig. 9 is that the recognition performance of dimensional Gaussian model difference σ value compares；
Figure 10 is the finger vena ROI image of four width difference postures；
Figure 11 is that the recognition performance of three kinds of feature extracting methods compares.
Specific embodiment
The finger multi-modal feature provided by the invention based on LGBP coding is melted in the following with reference to the drawings and specific embodiments
Conjunction method is described in detail.
It is provided by the invention based on LGBP coding the multi-modal Feature fusion of finger include carry out in order it is following
Step:
1) since the texture of three modality images of finger is different, the present invention utilizes scale parameter different (δ=4,5,6)
Gabor filter carries out Gabor filtering, the expression formula of Gabor filter such as formula to the three mode ROI image of finger of different postures
Shown in 1, respectively obtain 8 directions (0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and 157.5 °) fingerprint, refer to it is quiet
The amplitude characteristic image of arteries and veins and phalangeal configurations, as shown in Figure 1；
Wherein, δ represents the scale of Gabor filter, θkIndicate the angle value in k-th of direction.
2) it is carried out respectively using finger three mode amplitude characteristic image of the LBP (local binary patterns) to above-mentioned 8 directions
Coding, coding method is: firstly, defining a pixel centered on a certain pixel in a certain finger amplitude characteristic image
3 × 3 window two-values are carried out to remaining 8 neighborhood territory pixel point in the window using the gray value of the central pixel point as threshold value
Change.Such as: if the gray value of a certain neighborhood territory pixel point is less than the gray value of central pixel point, it is encoded to 0；Otherwise, it is encoded to
1, form 8 binary values；Then by binary value right shift b times, the binary value for often moving to right one is weighted
Summation obtains 8 LBP values of the pixel；Finally, take the smallest LBP value as the LBP value of the pixel, as shown in formula (2),
The three mode LGBP feature coding image of finger in 8 directions is consequently formed, as shown in Figure 2.
Wherein, function ROR (x, b) is indicated binary value x right shift b times,Indicate i-th of central pixel point
LBP value,Definition is as shown in formula (3):
In formula: B (Ii-Ic) indicate binaryzation function, i.e.,IiIndicate center pixel
The gray value of point i, IcIndicate that the gray value of neighborhood territory pixel point, a indicate a of binaryzation function, herein P=8.
3) since MRRID (more support area rotational invariance features) is only applicable to description topography, this step is to upper
The three mode LGBP feature coding image of finger for stating 8 directions carries out piecemeal, in the present invention, by three mould of finger in 8 directions
State LGBP feature coding image is respectively classified into 8 × 8 pieces, and piecemeal schematic diagram is as shown in Figure 3.
4) since the three mode LGBP feature coding block image of finger in above-mentioned 8 directions is smaller, if finding characteristic point meeting
The loss of detailed information is caused, therefore, the present invention improves MRRID, and the pixel of each block image is seen
Its gray feature is extracted at being characteristic point, gray feature vector is consequently formed, process is as follows:
Step 1: gray scale is grouped.Firstly, the gray value of each pixel of each block image is arranged from small to large
Sequence forms the sequence of a pixel；Then, this sequence is divided into k gray scale grouping according to the sum of pixel, forms k
Group gray scale is grouped image；The boundary point of each gray scale grouping is determined with the method to round up later, and obtains the boundary point
Gray value, as shown in formula (4):
Wherein,Indicate every group of boundary point, tiIndicate the boundary of i-th of gray scale grouping
Value, IminAnd ImaxRespectively indicate the minimum gradation value and maximum gradation value of image slices vegetarian refreshments.
Step 2: calculating the gray feature vector of each pixel.Since each pixel has 8 nearest neighbor points, because
This present invention compares the gray value size of its symmetrical adjoint point centered on each pixel in each gray scale grouping image, than
Such as:WithIt is the symmetrical adjoint point of a pair of pixel i, whereinFor the nearest neighbor point marked as 1 of pixel i,For
The nearest neighbor point marked as 5 of pixel i, as shown in figure 4, ifThe gray value of point is greater thanThe gray value of point, then be 1；
Otherwise it is 0, the gray feature vector of 4 binary codes is consequently formed, is then turned 4 binary code vectors using formula (5)
Turn to 16 binary code gray feature vectors.
Wherein, i indicates that ith pixel point, m indicate the logarithm of the pixel nearest neighbor point.
5) the gray feature vector of each pixel in above-mentioned each gray scale grouping image is superimposed, forms each gray scale point
The gray feature histogram of group image, then connect the gray feature histogram of each gray scale grouping image to form block image
Gray feature histogram indicates the LGIGF feature of block image using the histogram.Assuming that the piecemeal of three modality images of finger
Number N=8 × 8, gray scale are grouped number k=5, then the gray scale of the block image of the corresponding the first row first row of three finger single modes
Feature histogram is as shown in Figure 5.
6) since the LGIGF feature of the LGIGF aspect ratio marginal portion of the central part of three modality images of finger is stablized, because
This, generates coefficient identical with LGBP feature coding image block number by dimensional Gaussian model first, then to above-mentioned every
The gray feature histogram of one block image is weighted, later by the gray feature histogram of the block image after above-mentioned weighting
Figure series connection obtains finger single mode gray feature histogram, shown in dimensional Gaussian model such as formula (6):
Wherein, σ indicates the mean square deviation of dimensional Gaussian model, and m and n are the number of the image block of every row and each column respectively,
Mid (i) and mid (j) respectively represent the block image of picture centre in ith row and jth column.Block image LGIGF feature histogram
It is as shown in Figure 6 that figure weights concatenated schematic diagram.
Finally, above-mentioned finger single mode gray feature histogram fused in tandem is formed three mode gray feature histogram of finger
Figure.
In addition, according to formula (4) it is found that gray scale grouping image number k value it is related with the size of block image, that is, divide
The pixel number for including in block image is different, and the best value of k is also different.Therefore, we (receive characteristic by ROC
Curve) curve determines that optimum gradation is grouped the number k and best piecemeal number N of image, so that LGIGF feature histogram matching essence
Exactness is best.First, it will be assumed that N=8 × 8, then as shown in Figure 7, as k=4, LGIGF feature histogram matching precision is most
It is high；According to previous result, it is assumed that k=4, as shown in Figure 8, and as N=8 × 8, LGIGF feature histogram matching precision highest.
According to result above it is found that working as N=8 × 8, when k=4, LGIGF Feature fusion recognition performance proposed by the present invention is best.
Again according to formula (6) it is found that the stability of finger single mode characteristics of image is related with the shape of dimensional Gaussian model, due to two dimension
The shape of Gauss model depends on the value of meansquaredeviationσ, and therefore, the stability of finger single mode characteristics of image depends on mean square deviation
σ, as shown in Figure 9, as σ=0.15, LGIGF feature histogram matching precision highest.
7) formula (7) are utilized, by the gray feature histogram intersection for calculating two three mode ROI images of finger to be matched
A possibility that method of coefficient judges whether this two width finger ROI image matches, and the intersection coefficient of histogram is bigger, matching is got over
Greatly.
In formula: m1And m2Respectively indicate two finger ROI images to be matched, Hm1(i) and Hm2(i) two width are respectively represented to wait for
The gray feature histogram of matched three mode ROI image of finger, L indicate the gray feature histogram of three modality images of finger
Dimension.
During above-mentioned images match, three mode gray scale of finger in two finger ROI images to be matched is calculated first
The intersection coefficient of feature histogram.If calculated intersection coefficient > T (similitude decision-making value), then it represents that this two width finger ROI
Image is similar, that is, indicates that this two width finger ROI image matches；If it intersects coefficient≤T, this two width finger ROI image is determined
It mismatches.Similitude decision-making value T is that false rejection rate is 0 in finger ROI image matching result, and when mistake allows rate minimum
Corresponding threshold point.
The present inventor is based on the above method and has done two groups of experiments.In this two groups of experiments, it is all made of homemade database.This
Database includes 100 Different Individuals, and each individual refers to that vein ROI image and 10 width refer to comprising 10 width fingerprint ROI images, 10 width
Save line ROI image.3000 width finger, three mode ROI image in total.And the posture of the finger single mode image of each individual is respectively not
It is identical.Since the resolution ratio of the finger single mode image in database can have differences, will make by oneself database in fingerprint,
Refer to vein, the resolution ratio of finger joint print image adjusts separately as 152*152,88*200,88*200.Experimental situation is PC machine,
It is completed under Matlab R2010a environment.
In the first set of experiments, four width finger vena ROI images are selected from homemade database, as shown in Figure 10.It should
Four width pictures belong to the same person, and posture is different.
In this experiment, we are verified the present invention and are mentioned using following three kinds of Feature fusions indicated by histogram
The invariable rotary characteristic of LGIGF Feature fusion out.
1.LGBP feature coding: firstly, according to the narration of step 1 and step 2, the LGBP for foring 4 width finger venas is special
Levy coded image；Then, according to the narration of step 3, the LGBP feature coding image of 4 width finger venas is divided into 8 × 8 pieces, then
Each LGBP feature coding block image, traditional histogram expression side are described by traditional grey level histogram representation method
Method is: statistics LGBP feature coding image is superimposed to be formed at each gray value from the number of the pixel of gray value 0 to 255
Straight line constitutes the grey level histogram of the LGBP feature coding block image of 4 width finger venas；Finally by each 4 width finger vena
The grey level histogram of LGBP feature coding block image connect to form the histogram of the LGBP feature coding image of 4 width finger venas
Figure.
2. improved MRRID feature: firstly, 4 width finger vena ROI images are divided into 8 × 8 according to the narration of step 3
Block；Then, according to the narration of step 4, each finger vena ROI block image is described by improved MRRID, is formed each
The improved MRRID feature histogram of finger vena ROI block image；Finally changing each finger vena ROI block image
Into MRRID feature histogram connect to form the improved MRRID feature histogram of 4 width finger vena ROI images.
3.LGIGF feature: firstly, according to the narration of step 1 and step 2, the LGBP feature for foring 4 width finger venas is compiled
Code image；Then, according to the narration of step 3, the LGBP feature coding image of 4 width finger venas is divided into 8 × 8 pieces, then pass through
The improved MRRID that step 4 describes describes each LGBP feature coding block image, forms each LGBP feature coding piecemeal
The improved MRRID feature histogram of image；Finally by the dimensional Gaussian model described in step 5 by each LGBP feature coding
The improved MRRID feature histogram of block image weights and connects to form the improved MRRID spy of LGBP feature coding image
Levy histogram, i.e., LGIGF feature histogram proposed by the present invention.
According to feature histogram forming process as described above, by 4 width refer to the feature histogram of vein ROI image respectively into
Row matching compares it and intersects coefficient, as shown in table 1.Can be seen that LGIGF Feature fusion from the data of table 1 has preferably
Invariable rotary characteristic, this method solves the problems, such as that finger gesture is changeable to a certain extent.
1 histogram similarity factor of table
In the second set of experiments, we compile the LGBP feature described in LGIGF Feature fusion, step 1 and step 2
The recognition performance of the improved MRRID Feature fusion described in code method, step 4 compares, ROC curve such as Figure 11 institute
Show.Table 2 be these three Feature fusions match time and corresponding EER (mistake allows rate equal with false rejection rate
The discrimination at place).When can be seen that LGIGF Feature fusion matches from the match time of the different characteristic fusion method of table 2
Between it is shorter.In conjunction with the experimental result of Figure 11 and table 2 it is found that LGIGF Feature fusion not only solves hand to a certain extent
Refer to the changeable problem of posture, there is good matching effect, and improve matching efficiency, there is certain feasibility.
The recognition performance of the different descriptors of table 2
Claims (5)
1. a kind of multi-modal Feature fusion of finger based on LGBP coding, it is characterised in that: described to be encoded based on LGBP
The multi-modal Feature fusion of finger include the following steps carried out in order:
1) Gabor filter is carried out using finger three mode ROI image of the different Gabor filter of scale parameter to different postures
Wave obtains 8 directions respectively, i.e., 0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and 157.5 ° of fingerprint refers to vein
With the amplitude characteristic image of phalangeal configurations；
2) it is encoded respectively using finger three mode amplitude characteristic image of the LBP to above-mentioned 8 directions, 8 sides is consequently formed
To three mode LGBP feature coding image of finger；
3) piecemeal is carried out to the three mode LGBP feature coding image of finger in above-mentioned 8 directions；
4) it regards the pixel of each block image as characteristic point and extracts its gray feature, gray feature is consequently formed
Vector, process are as follows:
Step 1: gray scale is grouped；Firstly, the gray value of each pixel of each block image is ranked up from small to large,
Form the sequence of a pixel；Then, this sequence is divided into k gray scale grouping according to the sum of pixel, forms k group ash
Degree grouping image；The boundary point of each gray scale grouping is determined with the method to round up later, and obtains the gray scale of the boundary point
Value；
Step 2: calculate the gray feature vector of each pixel: being with each pixel in each gray scale grouping image
The heart compares the gray value size of its symmetrical adjoint point, if the gray value of some pixel is greater than the gray value of its symmetrical adjoint point, for
1；Otherwise it is 0, the gray feature vector of 4 binary codes is consequently formed, then converts 16 two for 4 binary code vectors
Ary codes gray feature vector；
5) the gray feature vector of each pixel in above-mentioned each gray scale grouping image is superimposed, forms each gray scale packet diagram
The gray feature histogram of picture, then connect the gray feature histogram of each gray scale grouping image to form the gray scale of block image
Feature histogram；
6) coefficient identical with LGBP feature coding image block number is generated by dimensional Gaussian model first, then to above-mentioned
The gray feature histogram of each block image is weighted, later that the gray feature of the block image after above-mentioned weighting is straight
Side's figure series connection obtains finger single mode gray feature histogram, finally, above-mentioned finger single mode gray feature histogram is connected
Fusion forms three mode gray feature histogram of finger；
7) sentenced by the method for the gray feature histogram intersection coefficient of two three mode ROI images of finger to be matched of calculating
Whether this two width finger ROI image that breaks matches.
2. the finger multi-modal Feature fusion according to claim 1 based on LGBP coding, it is characterised in that: in step
It is rapid 2) in, the method encoded respectively to the three mode amplitude characteristic image of finger in 8 directions is: firstly, definition
3 × 3 window of one pixel centered on a certain pixel in a certain finger amplitude characteristic image, with the center pixel
The gray value of point is threshold value, carries out binaryzation to remaining 8 neighborhood territory pixel point in the window；If the gray scale of a certain neighborhood territory pixel point
Value is less than the gray value of central pixel point, then is encoded to 0；Otherwise, it is encoded to 1, forms 8 binary values；Then by two into
Value right shift b times processed, summation is weighted to the binary value for often moving to right one and obtains 8 LBP values of the pixel；Most
Afterwards, take the smallest LBP value as the LBP value of the pixel；
The formula of minimum LBP value are as follows:
Wherein, function ROR (x, b) is indicated binary value x right shift b times,Indicate the LBP of i-th of central pixel point
Value,Definition is as shown in formula (3):
In formula: B (Ii-Ic) indicate binaryzation function, i.e.,IiIndicate central pixel point i's
Gray value, IcIndicate that the gray value of neighborhood territory pixel point, a indicate a of binaryzation function, herein P=8.
3. the finger multi-modal Feature fusion according to claim 1 based on LGBP coding, it is characterised in that: in step
It is rapid 4) in, the gray value formula of the described acquisition boundary point are as follows:
Wherein, Indicate every group of boundary point, tiIndicate the boundary value of i-th of gray scale grouping,
IminAnd ImaxRespectively indicate the minimum gradation value and maximum gradation value of image slices vegetarian refreshments.
4. the finger multi-modal Feature fusion according to claim 1 based on LGBP coding, it is characterised in that: in step
It is rapid 4) in, the gray feature vector by 4 binary codes is converted into the public affairs of the gray feature vector of 16 binary codes
Formula are as follows:
Wherein, i indicates that ith pixel point, m indicate the logarithm of the pixel nearest neighbor point.
5. the finger multi-modal Feature fusion according to claim 1 based on LGBP coding, it is characterised in that: in step
It is rapid 7) in, the gray feature histogram intersection coefficient by calculating two three mode ROI images of finger to be matched
Method is to judge this whether matched method of two width finger ROI image: calculating first with following intersection coefficient expressions
The intersection coefficient of three mode gray feature histogram of finger in two finger ROI images to be matched, if calculated intersection is
Number > similitude decision-making value T, then it represents that this two width finger ROI image is similar, that is, indicates that this two width finger ROI image matches；If
It intersects coefficient≤T, then determines that this two width finger ROI image mismatches；Similitude decision-making value T is the matching of finger ROI image
As a result middle false rejection rate is 0, and the threshold point that mistake is corresponding when allowing rate minimum；
Intersect the expression formula of coefficient are as follows:
In formula: m1And m2Two finger ROI images to be matched are respectively indicated,WithRespectively represent two it is to be matched
Three mode ROI image of finger gray feature histogram, L indicate three modality images of finger gray feature histogram dimension
Number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510518806.4A CN105095880B (en) | 2015-08-20 | 2015-08-20 | A kind of multi-modal Feature fusion of finger based on LGBP coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510518806.4A CN105095880B (en) | 2015-08-20 | 2015-08-20 | A kind of multi-modal Feature fusion of finger based on LGBP coding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105095880A CN105095880A (en) | 2015-11-25 |
| CN105095880B true CN105095880B (en) | 2019-01-11 |
Family
ID=54576268
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510518806.4A Expired - Fee Related CN105095880B (en) | 2015-08-20 | 2015-08-20 | A kind of multi-modal Feature fusion of finger based on LGBP coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105095880B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975951A (en) * | 2016-05-27 | 2016-09-28 | 国创科视科技股份有限公司 | Finger vein and fingerprint fusion identification method of middle part of finger |
| CN108388862B (en) * | 2018-02-08 | 2021-09-14 | 西北农林科技大学 | Face recognition method based on LBP (local binary pattern) characteristics and nearest neighbor classifier |
| CN108416814B (en) * | 2018-02-08 | 2020-07-31 | 广州大学 | Method and system for quickly positioning and identifying pineapple head |
| CN108509927B (en) * | 2018-04-09 | 2021-09-07 | 中国民航大学 | Finger vein image identification method based on local symmetrical graph structure |
| CN108596126B (en) * | 2018-04-28 | 2021-09-14 | 中国民航大学 | Finger vein image identification method based on improved LGS weighted coding |
| CN109190566B (en) * | 2018-09-10 | 2021-09-14 | 中国民航大学 | Finger vein recognition method integrating local coding and CNN model |
| CN109308462B (en) * | 2018-09-10 | 2021-03-26 | 中国民航大学 | Finger vein and knuckle print region-of-interest positioning method |
| CN109902585B (en) * | 2019-01-29 | 2023-04-07 | 中国民航大学 | Finger three-mode fusion recognition method based on graph model |
| CN109886220A (en) * | 2019-02-26 | 2019-06-14 | 北京凌云天润智能科技有限公司 | A kind of characteristics extraction and comparison algorithm of finger vein image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070172114A1 (en) * | 2006-01-20 | 2007-07-26 | The Johns Hopkins University | Fusing Multimodal Biometrics with Quality Estimates via a Bayesian Belief Network |
| CN102521575A (en) * | 2011-12-16 | 2012-06-27 | 北京天诚盛业科技有限公司 | Iris identification method based on multidirectional Gabor and Adaboost |
| CN102629320A (en) * | 2012-03-27 | 2012-08-08 | 中国科学院自动化研究所 | Ordinal measurement statistical description face recognition method based on feature level |
| CN102799870A (en) * | 2012-07-13 | 2012-11-28 | 复旦大学 | Single-training sample face recognition method based on blocking consistency LBP (Local Binary Pattern) and sparse coding |
| CN103679153A (en) * | 2013-12-16 | 2014-03-26 | 中国民航大学 | Finger multi-modal biometric characteristic polarization imaging system |
-
2015
- 2015-08-20 CN CN201510518806.4A patent/CN105095880B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070172114A1 (en) * | 2006-01-20 | 2007-07-26 | The Johns Hopkins University | Fusing Multimodal Biometrics with Quality Estimates via a Bayesian Belief Network |
| CN102521575A (en) * | 2011-12-16 | 2012-06-27 | 北京天诚盛业科技有限公司 | Iris identification method based on multidirectional Gabor and Adaboost |
| CN102629320A (en) * | 2012-03-27 | 2012-08-08 | 中国科学院自动化研究所 | Ordinal measurement statistical description face recognition method based on feature level |
| CN102799870A (en) * | 2012-07-13 | 2012-11-28 | 复旦大学 | Single-training sample face recognition method based on blocking consistency LBP (Local Binary Pattern) and sparse coding |
| CN103679153A (en) * | 2013-12-16 | 2014-03-26 | 中国民航大学 | Finger multi-modal biometric characteristic polarization imaging system |
Non-Patent Citations (1)
| Title |
|---|
| 多模态生物特征识别技术进展综述;王瑜 等;《计算机应用与软件》;20090228;第26卷(第2期);第32-34页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105095880A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105095880B (en) | A kind of multi-modal Feature fusion of finger based on LGBP coding | |
| CN104850825B (en) | A kind of facial image face value calculating method based on convolutional neural networks | |
| CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
| CN105138974B (en) | A kind of multi-modal Feature fusion of finger based on Gabor coding | |
| CN108520216B (en) | Gait image-based identity recognition method | |
| CN102844766B (en) | Human eyes images based multi-feature fusion identification method | |
| CN102682302B (en) | Human body posture identification method based on multi-characteristic fusion of key frame | |
| JP2017157138A (en) | Image recognition device, image recognition method and program | |
| CN109871781A (en) | Dynamic gesture identification method and system based on multi-modal 3D convolutional neural networks | |
| CN109190566A (en) | A kind of fusion local code and CNN model finger vein identification method | |
| Shams et al. | Iris recognition based on LBP and combined LVQ classifier | |
| Arif et al. | Automated body parts estimation and detection using salient maps and Gaussian matrix model | |
| CN105160305B (en) | A kind of multi-modal Feature fusion of finger | |
| Yu et al. | Human action recognition with extremities as semantic posture representation | |
| CN106529504B (en) | A kind of bimodal video feeling recognition methods of compound space-time characteristic | |
| KR101105435B1 (en) | Face detection and recognition method | |
| CN108596126A (en) | A kind of finger venous image recognition methods based on improved LGS weighted codings | |
| CN105469050B (en) | Video behavior recognition methods based on local space time's feature description and pyramid words tree | |
| CN109919137B (en) | Pedestrian structural feature expression method | |
| CN102592150B (en) | Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory | |
| Mohamed et al. | Adaptive extended local ternary pattern (aeltp) for recognizing avatar faces | |
| Arif et al. | Human pose estimation and object interaction for sports behaviour | |
| CN110222568A (en) | A kind of across visual angle gait recognition method based on space-time diagram | |
| Uddin et al. | Human activity recognition using robust spatiotemporal features and convolutional neural network | |
| Li et al. | Pedestrian detection based on clustered poselet models and hierarchical and–or grammar |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190111 Termination date: 20190820 |