Summary of the invention
The objective of the invention is to overcome deficiency of the prior art, a kind of relevant sliding fingerprint sequence seamless joint method of phase place of expanding is provided, make them under different sliding speeds and different elastic deformations, joining method still can obtain high-quality fingerprint image.
The present invention is achieved by the following technical solutions, the present invention at first utilizes the estimation based on piece coupling to obtain pixel level translational movement between the adjacent two frame fingerprint images, be correlated with according to the phase place of expansion then and obtain inferior pixel level translational movement estimation, the image that utilizes the linear elasticity model that splicing is obtained at last carries out the slit self-adaptation and removes.
The present invention includes following concrete steps:
1, the detection of the pixel level translational movement of sliding fingerprint sequence
The position of at first using translation motion (Translational Motion) to be similar between the adjacent two frame fingerprint images concerns that promptly there is following relation: F in the gray-scale value at the corresponding pixel in the overlapping region of adjacent two frame fingerprint images
I+1(x, y)=F
i(x+ Δ x, y+ Δ y), wherein (Δ x, Δ y) is former frame fingerprint image F
iWith back one frame fingerprint image F
I+1Between side-play amount.
In order to try to achieve side-play amount (Δ x, Δ y), the present invention uses the method for estimating based on the piece coupling to detect the pixel level side-play amount.At F
iIn get a center at (x
0, y
0), size is (H
b* W
b) piece B
S, at F
I+1The middle searching and piece B
SThe equirotal piece B that mates most
S', then according to piece B
S' centre coordinate and (x
0, y
0) between corresponding relation try to achieve side-play amount (Δ x, Δ y), concrete steps are as follows:
With B
SSort top to bottom and form a vectorial G={g
i}
I=1,2 ..., N, N=H wherein
b* W
b, obtain the increment sign binary sequence B={b of a correspondence according to the magnitude relationship between adjacent two elements
i}
I=1,2 ..., N-1If: g
i>g
I+1, b then
i=1; Otherwise b
i=0.At F
I+1Appoint get a center (x, y), size is (H
b* W
b) piece B
S', obtain corresponding increment sign binary sequence B '={ b
i'
I=1,2 ..., N-1, compare the increment sign correlativity between B and these two increment sign binary sequences of B '
Wherein~and the expression negate, represents XOR.(x y) travels through once, and (x y) obtains that peaked (x y) is (x to establish corresponding rc to all possible
1, y
1), then the side-play amount that will detect is (Δ x, Δ y)=(x
0-x
1, y
0-y
1).
2, the detection of the inferior pixel level translational movement of sliding fingerprint sequence
To equation F
I+1(x, y)=F
iFourier transform is done on (x+ Δ x, y+ Δ y) both sides, obtains following relational expression: ζ
I+1(u, v)=ζ (u, v) e
2 π j (u Δ x+v Δ y), after the normalization:
Wherein ζ is obtained by the F Fourier transform, and ζ (u, v)
*Be that ((u v) is referred to as normalized phase place correlation matrix to Q to ζ for u, conjugate complex number v).With order polynomial P (x)=p
1X+p
2To strange vector (Singular Vector) v=e
2 π ju Δ xAnd v=e
2 π jv Δ yCorresponding phasing degree (unwrap (∠ is v)) carries out least square fitting respectively, and the side-play amount that can get inferior pixel level is estimated:
Because with maximal increment symbol correlativity is the block matching motion estimation of criterion, the side-play amount that seek out inferior pixel level is estimated the general notion that must use multiresolution, and this has increased calculated amount accordingly.Simultaneously the inferior pixel level side-play amount that obtains based on the expansion phase correlation method is estimated, in adjacent two frame fingerprint image overlapping areas 30% o'clock less than every frame fingerprint image area, side-play amount is estimated just not necessarily accurate.Therefore, the present invention is in conjunction with the relative merits of these two kinds of methods, to estimating that obtaining side-play amount accepts or rejects according to the following rules: when | Δ x-Δ x ' | in the time of>1, the side-play amount on the x direction is estimated as Δ x, otherwise is Δ x '.In like manner the side-play amount on the y direction is done similar processing.If the side-play amount that last estimation obtains is (Δ x
0, Δ y
0).
3, the slit self-adaptation is removed
Because sliding fingerprint sequence is when gathering, the pressure inequality of finger presses Acquisition Instrument surface of contact makes to exist different elastic deformations between the adjacent two frame slip fingerprint images.If just suppose to have translation between the adjacent two frame slip fingerprint images, can there be the slit in the fingerprint image that splicing obtains.In fact, can think that former frame sliding fingerprint image is not have elastically-deformablely, then a frame obtains through elastic deformation.Like this, just meet a linear elasticity deformation model between the adjacent two frame slip fingerprint images, and (x, y, t): R
2* R → R
2, (t) (x is y) in the locus of moment t for pixel of expression for x, y for here.To arbitrary given pixel (x, y), (x
S, y
S)= (x, y, 0) represents the position before its elastic deformation, (x
E, y
E)= (x, y, t
E) represent that it is through the position after the elastic deformation. (x, y, t), 0≤t≤t
ERepresent given pixel (x, y) elastically-deformable track.Though, point often all different in the elastic deformation of each several part, to a zonule R
0(fritter in the finger just), the elastic deformation of its inner each pixel can suppose it is the same, satisfies a rigid body translation.Promptly have: (x, y, t
E)=A (x, y, 0)+b, b=(x here
b, y
b) ∈ R
2Represent a translation vector, A ∈ R
2* R
2Represent a rotation matrix, and A
TA is one 2 * 2 a unit matrix, and the determinant of A is 1.
In order to satisfy stitching algorithm requirement of processing speed in real time in mobile device, with above-mentioned zonule R
0Be reduced to N
a* (H
b-Δ y
0) little rectangle, this hypothesis under A just be reduced to a unit matrix.Because the elastic deformation amplitude of each little rectangle all should be very little in the same frame slip fingerprint image, might as well suppose that the side-play amount of each little rectangle on vertical and horizontal direction all is no more than Δ S.Be that the slit self-adaptation that the present invention proposes is removed algorithm below: little rectangle is moved m pixel in the x direction, calculate the rectangle and the previous frame sliding fingerprint image that obtain after moving and add up and S at the absolute value of the gray scale difference of overlapping region
m, find out S
mObtain the pairing m of minimum value as x
bIn like manner obtain y
b, obtain translation vector (x at last
b, y
b).
Obtain after the translation vector of each little rectangle correspondence, also will do affined transformation these little rectangles.Because concerning adjacent two the little rectangles in the left and right sides, that limit, the left side of that limit, the right side of the little rectangle in the left side and the little rectangle in the right should be overlapping, but after the displacement of these little rectangles through translation vector separately, may become trapezoidal.Utilize the affined transformation can be with these trapezoidal overlapping quadrilaterals in that limit, the right side and original little rectangle that limit, the right that is converted into.
The present invention utilizes maximal increment symbol correlativity to survey the block matching motion method of estimation of function as standard, because the one-dimensional signal that image is converted into two-value earlier handles, so processing speed is very fast, is easy to realize and quicken by hardware.In the method for estimation of the relevant inferior pixel level side-play amount of the phase place of utilizing expansion, owing to can utilize fast fourier transform to substitute Fourier transform, processing speed is also very fast.
In order to improve the robustness of joining method to noise, suppress the aliasing and the boundary effect of wherein existence, the present invention also introduces the filtering of oval two-value mask and the Kaiser window function improves the accuracy that side-play amount is estimated.Oval mask filtering is defined as follows:
Corresponding oval two-value mask filter function is:
Oval two-value mask filter function ξ is carried out filtering to normalization phase place correlation matrix Q suppress aliasing, that is: Q=Q ο ξ.The Kaiser window function K of one dimension (L, β) definition:
Wherein, L and β are respectively two parameters of window function, the major axis of corresponding elliptic function and minor axis, B
0(x) be 0 rank Bessel function.Be generalized to the Kaiser window function of two dimension: W
k=K (λ
1H, β)
Tο K (λ
2W, β)
T, 0≤λ
1, λ
2≤ 1.λ wherein
1And λ
2Be two weight coefficients.Before image is done Fourier transform, utilize the Kaiser window function of two dimension that image is carried out filtering, suppress boundary effect.
The present invention utilizes based on the block matching motion method of estimation of maximal increment symbol correlativity as matching criterior, can not only in hardware, realize easily and quicken, and very low to the requirement of overlapping area between the adjacent two frame fingerprint images, robustness is very strong.The present invention has utilized the phase correlation method of expansion to detect side-play amount between the adjacent two frame fingerprint images, not only can obtain high-precision side-play amount and estimate, and can replace Fourier transform by FFT, improves the speed of splicing.The present invention has utilized the linear elasticity deformation model, utilizes the slit of this model in can adaptive removal stitching image, improves the quality that splicing obtains fingerprint image.The joining method that the present invention proposes by after adjusting relevant parameters and being provided with, goes for various existing glide finger print acquisition modules.Behind the sliding fingerprint acquiring and concatenation module packing encapsulation with integrated joining method of the present invention, can be applied in all kinds of mobile devices.
Embodiment
In order to understand technical scheme of the present invention better, be further described below in conjunction with accompanying drawing and specific embodiment.
The present invention adopts the sliding fingerprint acquiring instrument of the Sweden FPC1031B of FINGERPRINT CARDS company series, the size of the every two field picture of fingerprint sequence that this sliding fingerprint acquiring instrument collects is 32 * 152 (W=152pixel, H=32pixel), resolution is 363dpi, Fig. 1 is the sliding fingerprint sequence that collects with this Acquisition Instrument, and table 1 is the setting of the parameter of using in the joining method of the present invention.
Table 1 initial parameter
Embodiment
1, the detection of the pixel level translational movement of sliding fingerprint sequence
Fig. 2 shown based on block matching motion estimation principles figure, and various Block Matching Algorithm exists difference aspect following three: 1) matching criterior; 2) search strategy; 3) selection of block size.The present invention adopts maximal increment symbol correlativity as matching criterior, owing to wherein relate generally to Binary Operation, is convenient to realize on hardware.Because finger is consistent in the direction that sliding fingerprint acquiring instrument face slides, and might as well suppose to point from top to bottom to slide.Like this, the lower zone of former frame fingerprint image is overlapping with the upper area of back one frame fingerprint image.In view of the above, the present invention is according to (the x that sets in the table 1
0, y
0) and (H
b* W
b) value detects the value of side-play amount.Fig. 3 is the increment sign correlativity 3-D display that has translation between two width of cloth images, as can be seen, has a tangible crest at the side-play amount correspondence position.
2, the detection of the inferior pixel level translational movement of sliding fingerprint sequence
Estimate to obtain the accuracy of side-play amount in order to verify joining method of the present invention, our intercepting artificial from same image two two field pictures, and in image artificial adding white Gaussian noise, side-play amount between them is (x, y), estimate that according to block matching motion the side-play amount that obtains is (Δ x, Δ y), the side-play amount that the combined block matched motion is estimated and the detection of expansion phase correlation method obtains is (Δ x
0, Δ y
0), and introducing variable γ illustrates the robustness of joining method of the present invention to adjacent two frame fingerprint image overlapping areas.Here γ represents the ratio of overlapping area and every frame fingerprint image area, wherein realizes by down-sampling (downsample) less than 1 side-play amount.
The emulation experiment that table 2 side-play amount detects
(x,y) | ??(Δx,Δy) | (Δx
0,Δy
0)
| ????γ |
(0.0000,0.0000) | ????(0,0) | (0,-1.2673e-016) | ????1 |
(0.0000,0.5000) | ????(0,0) | (0.0000,0.5156) | ????0.9967 |
(0.2500,0.2500) | ????(0,0) | (0.2502,0.2496) | ????0.9906 |
(0.5000,0.2500) | ????(1,0) | (0.4952,0.2398) | ????0.9828 |
(0.5000,0.5000) | ????(1,0) | (0.5166,0.4782) | ????0.9811 |
(5.0000,5.0000) | ????(5,5) | (5.0213,5.0090) | ????0.8160 |
(10.0000,5.0000) | ????(10,5) | (10.0127,4.9882) | ????0.6649 |
(14.0000,5.0000) | ????(14,5) | (14.0116,4.9665) | ????0.5440 |
(20.0000,5.0000) | ????(20,5) | (20.0236,5.0106) | ????0.3627 |
(24.0000,5.0000) | ????(24,5) | (24.0275,5.0078) | ????0.2418 |
(28.0000,5.0000) | ????(28,5) | (28.0000,5.0186) | ????0.1209 |
(30.0000,5.0000) | ????(30,5) | (30.0000,5.0163) | ????0.0604 |
As can be seen from Table 2, estimate to detect the side-play amount that obtains inferior pixel level precision,, estimate (the Δ x that obtains even in γ≤0.1 with the relevant joining method of expansion phase place based on block matching motion
0, Δ y
0) still accurately.
3, remove in the slit
As shown in Figure 4, the figure on the left side does not do the effect that remove in the slit, and the figure on the right is the effect behind the removal slit, and as can be seen, the fingerprint image after remove in the slit is obviously smooth a lot.Fig. 5 is the fingerprint image that obtains according to joining method splicing proposed by the invention.
Present embodiment is based on the sliding fingerprint seamless joint method of expansion phase correlation method, and unique point and similarity comparative result between the fingerprint image that fingerprint image that splicing obtains and U.are.U4000 optical fingerprint acquisition instrument collect see Table 3.Whether the quality of splicing effect except can be subjective whether have the stitching image obvious gaps exists from visually analyzing, produces pseudo-characteristic point or causes real unique point to lose, and also is a very important weighing criteria.The value of similarity is between 0 to 1, and it is similar more to be worth big more expression two width of cloth fingerprint images.As known from Table 3, the image that image that splicing obtains and U.are.U4000 collect, corresponding similarity can judge at this time that all greater than 0.150 two width of cloth images are that the match is successful.
Table 3 unique point and similarity are relatively
The sample of Xuan Zeing at random | Count from the detected feature of stitching image | Count from the detected feature of U.are.U4000 fingerprint image | The unique point that is complementary is right | Similarity |
????1 | ????25 | ????37 | ????22 | ??0.178 |
????2 | ????23 | ????29 | ????18 | ??0.445 |
????3 | ????22 | ????35 | ????18 | ??0.195 |
????4 | ????38 | ????35 | ????28 | ??0.188 |
????5 | ????30 | ????34 | ????23 | ??0.365 |
????6 | ????19 | ????20 | ????14 | ??0.320 |
????7 | ????25 | ????36 | ????18 | ??0.275 |
????8 | ????46 | ????42 | ????27 | ??0.370 |
????9 | ????37 | ????40 | ????26 | ??0.560 |
????10 | ????25 | ????23 | ????15 | ??0.235 |