CN103793919A - Two-dimensional image matching effect evaluation method - Google Patents
Two-dimensional image matching effect evaluation method Download PDFInfo
- Publication number
- CN103793919A CN103793919A CN201410088955.7A CN201410088955A CN103793919A CN 103793919 A CN103793919 A CN 103793919A CN 201410088955 A CN201410088955 A CN 201410088955A CN 103793919 A CN103793919 A CN 103793919A
- Authority
- CN
- China
- Prior art keywords
- value
- main peak
- image
- measure function
- correlated measure
- 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.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the field of optical measurement and target recognition, and particularly relates to a two-dimensional image matching effect evaluation method. The method is an evaluation index system containing six parameters, and the larger a simple signal to noise ratio is, the lower the possibility that detection of a main peak is missed is; the larger the cross-sectional area of the bottom of the main peak is, the easier the peak can be found; high matching accuracy can be obtained conveniently through appropriate main peak standard summit cross-sectional area and a larger main peak smoothing coefficient, the shorter the average computing time is, the quicker matching is, and the smaller the average value of matched absolute error vector models is, the higher the matching accuracy is. Due to the fact that six evaluation indexes are used flexibly, an optimal related measure function, an optimal search algorithm, an optimal sub-pixel interpolation method and the like can be determined by using small-range experimental data, in an auxiliary mode, low-efficiency matching operation is reduced to the maximum, and the matching quality is optimized.
Description
Technical field
The present invention relates to Image-matching technical field, particularly relate to a kind of matching effect evaluation method of two dimensional image.
Background technology
Images match also claims image recognition, and this technology is a study hotspot of engineering field in recent decades always, because it is basis and the gordian technique in a lot of fields.There are missile guidance, aerial image analysis, streamline monitoring, traffic administration, word identification, stereoscopic vision, the 3D characteristic recovery, Medical Image Processing, Non-Destructive Testing etc. of map, Experimental Mechanics (optical measurement), scene automatically in the field of its application.
Images match is in fact in transformation space, to find a kind of (class) conversion, makes spatially to realize mapping from two width or the multiple image of the Same Scene of different time or different sensors or different visual angles.Matching algorithm can be divided into the coupling based on gray scale, coupling and the transform domain method etc. based on feature.
Image generally all has some higher characteristics of image, and as gray scale variation, profile, texture, angle point etc., these features are called " object ".If template image f has at least one unique object or object composition, in target image T by the subdomain g of the content that comprises f
(u, v)identify is to be relatively easy to.But in the image that sometimes needs to compare, there is no obvious object or because image too fuzzy (or template is too small) makes cannot define in template the object that is enough to identification, in example certain Rock Mechanics Test And as shown in Figure 1, the surface image of test specimen forms (being commonly referred to as speckle pattern) by some random gray scale spots.When the speckle pattern that two width are contained to identical content mates, because independent speckle does not contain the quantity of information that is enough to resolution, cannot regard object searched for as, only have by increase and compare window, speckle in window is increased, could realize coupling until it has the quantity of information that is enough to differentiate.This coupling is a kind of non-object identification, because adopted image mostly is gray level image, so be commonly referred to as the coupling based on area grayscale.In Experimental Mechanics (optical measurement) field, images match function generally adopts certain correlated measure to calculate, so this image matching method is called to Digital Speckle Correlation Method-DSCM (Digital Speckle Correlation Method).
Images match based on area grayscale can be made following mathematical description:
If I={I (i, j) | 1≤i≤m, 1≤j≤n} is original image, f={f (i, j) | f (i, j)=I (i, j), m
1≤ i≤m
2, n
1≤ j≤n
2it is truncated picture template wherein.T={T (i ', j ') | 1≤i '≤k, 1≤j '≤l} is target image, g={g (i ', j ') | g (i ', j ')=T (i ', j '), k
1≤ i '≤k
2, l
1≤ j '≤l
2be its subdomain, when two width image taking yardsticks are when identical, get k
2-k
1=m
2-m
1and l
2-l
1=n
2-n
1.When by template and the comparison of target subdomain, the gray-scale value of corresponding point position is respectively f (i, j) and g (i+u, j+V), wherein u=k
1-m
1, v=l
1-n
1, (u, v) is relative displacement vector between the picture of g and f, correspondingly g can be expressed as to g
(u, v).According to certain related algorithm, can set up the correlated measure ρ between image.The correlated measure value of calculating g and f, is denoted as ρ (u, v).Change u and v value, be equivalent to choose different target subdomains in T.Each target subdomain has formed correlated measure function ρ=ρ (u, v) with respect to the correlated measure value of f.Have many algorithms based on estimating of Image gray correlation, Jin Guanchang has listed conventional formula in " area of computer aided optical measurement (the 2nd edition) " (Beijing: publishing house of Tsing-Hua University, 2007) 146 pages.In light of the circumstances, adopt suitable algorithm, can make correlated measure function ρ (u, v) obtain good graphical representation, for example when ρ (u, v) have significantly smooth when unimodal, the corresponding subdomain of its peak value
just comprise identical content with template f.
The key issue of image matching technology is to look for an effective matching algorithm.In engineering practice, due to the variation of shooting time, shooting angle, physical environment and the defect of sensor itself, make the image of shooting not only affected by noise, and there is serious tonal distortion and geometric distortion, to the requirement of image matching algorithm concentrate on following some: the matching precision that a. is higher, b. robustness is good, the operand that c. is fewer.
Jin Guanchang etc. are being published in the performance index that proposed four correlation formulas in the article of " Experimental Mechanics " 21 6 phase of volume 689-702 pages in 2006 " Digital Speckle Correlation Method progress and application ": C
mfor related coefficient maximal value; C
secfor inferior peak facies relationship numerical value; W
50be that main peak is 0.5C in related coefficient
mthe width at place; E
xyfor the absolute error of average displacement measurement.The definition of first three parameter as shown in Figure 4 above.Jin Guanchang
[7]deng thinking: C
secbe worth the less related coefficient maximal value C that more easily determines
m, stable C
secalso be beneficial to determining of main peak threshold value; W
50less, peak is more sharp-pointed, and the probability at misjudgement peak is less, and is conducive to improve search precision and speed.
C
mtheoretical value be generally constant, measured value is also all approximate with theoretical value, and the deviation of measured value and theoretical value mainly causes (as sub-pixel skew of the imaging precision error of photograph/video camera, two width image imaging grids etc.) by uncontrollable factor, and little with related function quality and searching method relation, so only there is statistical significance limited under Conditions of General Samples.C
secrandomness very strong, its stability is than the related coefficient σ of the outer relevant function surface of main peak
can at least low order of magnitude, so practical significance is little.Because each level line of measure function curved surface is not generally the circle of standard in contour plane, so measure W
50time secant position selection operation poor.E
xyin fact be not a simple indicator, but total evaluation index of all working quality, the factor that affects it is a lot, as the algorithm of correlated measure, shape of template and yardstick, impact point search/interpolation algorithm etc., by also improper side by side to itself and other index.To E
xyperhaps than direct application, it is more of practical significance to carry out further detailed factor analysis/decomposition, but multivariate statistics factor analysis needs very large sample and the result of so further calculating in fact to amplify error, and therefore practical significance is also little.E in addition
xycalculating depend on the precise information of displacement field, this is can not be obtainable in practice, so only could calculate in the time doing pure theory and analyze and the in the situation that of making some strict hypothesis.
Summary of the invention
The invention provides a kind of matching effect evaluation method of two dimensional image, the method is an assessment indicator system that comprises 6 parameters, can evaluate the matching effect of two dimensional image, reduce to greatest extent inefficient matching operation, Optimized Matching quality, having solved prior art cannot be accurately, the problem of objective evaluation image matching effect.
For achieving the above object, the technical solution adopted in the present invention is: a kind of matching effect evaluation method of two dimensional image, and step is as follows:
(1) establish I={I (i, j) | 1≤i≤m, 1≤j≤n} is original image, f={f (i, j) | f (i, j)=I (i, j), m
1≤ i≤m
2, n
1≤ j≤n
2be truncated picture template from described original image, T={T (i ', j ') | 1≤i '≤k, 1≤j '≤
lbe target image, in T, intercept target subdomain g={g (i ', j ') | g (i ', j ')=T (i ', j '), k
1≤ i '≤k
2, l
1≤ j '≤l
2, in the time that described original image is identical with target image shooting yardstick, k
2-k
1=m
2-m
1and l
2-l
1=n
2-n
1;
(2), when described image template f and target subdomain g being compared, the gray-scale value of corresponding point position is respectively f (i, j) and g (i+u, j+v), wherein u=k
1-m1, v=l
1-n
1, (u, v) is relative displacement vector between the picture of g and f, and g is expressed as to g
(u, v), according to corresponding related algorithm, set up the correlated measure C between image, calculate the correlated measure value of g and f, be denoted as C (u, v); Change u and v value, choose exactly different target subdomains in T, each target subdomain has formed correlated measure function C=C (u, v) with respect to the correlated measure value of f;
(3) calculate following six evaluation indexes:
1. simple signal to noise ratio (S/N ratio)
wherein C
mfor correlated measure function maximal value, this correlated measure function maximal value is theoretical value or observation mean value,
for main peak relevant measure function average outward, σ
cfor main peak relevant measure function standard deviation outward;
2. main peak bottom section area S
l, border is taken as and makes
level line;
3. the accurate summit area of section of main peak S
t, border is taken as and makes C=C
t=C
l+ 0.7 (C
m-C
l)=0.7C
m+ 0.3C
llevel line;
4. curved surface main peak smoothing factor
be defined as
wherein
(i=L, M, H, T) is each contour plane normalization circularity, S
ifor i contour plane area of the curved surface of function C, L
ifor i level line girth of the curved surface of function C, F
lthe value of corresponding correlated measure function is C
l, F
tthe value of corresponding correlated measure function is C
t, F
mthe value of corresponding correlated measure function is C
m=C
l+ 0.2 (C
m-C
l)=0.2C
m+ 0.8C
l, F
hthe value of corresponding correlated measure function is C
h=C
l+ 0.4 (C
m-C
l)=0.4C
m+ 0.6C
l;
5. the average computation time
for completing the averaging time of an images match, in the time not considering searching algorithm and interpolation calculation, adopt the averaging time of calculating a measure function value
its unit is taken as CPU time;
6. mate the average of absolute error vector mould
its unit is pixel value or is scaled actual distance value;
Described evaluation index 1.～be 4. about (m
1, m
2, n
1, n
2)
tstochastic variable, to evaluation index 1.～4. be calculated as rule of thumb sample 50 times above calculate average;
(4) evaluation criterion of according to 6 evaluation indexes in described step (3), the matching effect of two dimensional image being evaluated is:
1. SSNR is the bigger the better in its preset range, and larger SSNR means that main peak is more outstanding, easier with threshold method location main peak;
2. S
lin its preset range, be the bigger the better, this index is larger, more easily searches main peak by Monte Carlo analysis, genetic algorithm, is conducive to improve the speed that searches main peak; In the time that target image is gone up deformation rate in any direction and is less than 1% with respect to original image, get a
0=2.5; When target image when deformation rate >1% or signal noise ratio (snr) of image <40dB, is got a with respect to original image at least one direction
0=2.0 screening main peaks;
3. larger SSNR and less S
tmean template identified more reliably, less S
trefer to S
tspan be 6～8 pixels, larger SSNR refers to SSNR>3;
4.
be the bigger the better,
5.
it is the smaller the better,
6.
the smaller the better, while lacking the theoretical model of image, do not calculate,
(5) according to the evaluation criterion in step (4), the matching effect of two dimensional image is evaluated, the work of carrying out comprises:
1. in the situation that all the other conditions are identical, select different measure functions to carry out Matching Experiment, select most suitable measure function according to the performance of evaluation index;
2. in the situation that all the other conditions are identical, select different templates shape and yardstick to carry out Matching Experiment, determine most suitable shape of template and yardstick according to the performance of evaluation index;
3. in the situation that all the other conditions are identical, select different searching method search optimal match point, according to the performance of evaluation index to selecting most suitable searching method;
4. at least two in above three work are combined and unified.
Further, the isocontour method for drafting of correlated measure function surface is:
(1) higher if carry out the picture quality of registration, draw all level lines with bilinear interpolation, picture quality higher finger signal noise ratio (snr) of image >50dB and target image are here gone up deformation rate <1% in any direction with respect to original image;
(2) lower if carry out the picture quality of registration, corresponding C
tisocontour method for drafting be: using the extreme point of correlated measure function as preliminary registration center, first near the correlated measure Jacobian matrix preliminary registration center is tieed up to cubic spline interpolation, the length of side of interpolation area is value between 5～8, draws corresponding C according to the correlated measure Jacobian matrix after interpolation
tlevel line; All the other each isocontour draftings adopt bilinear interpolation to draw.
Further, calculate main peak relevant measure function average outward
with main peak relevant measure function standard deviation sigma outward
ctime, region within the scope of main peak in correlated measure function surface is removed and calculated, and according to the feature of correlated measure function surface, centered by preliminary registration center, remove a square area in correlated measure Jacobian matrix, the length of side of described square area is value in 10～30 pixels.
The beneficial effect that the present invention reaches: the method for introduction of the present invention can be evaluated the matching effect of two dimensional image more accurately, more objectively, six evaluation indexes of the present invention that applies in a flexible way definition can be utilized the auxiliary definite best correlated measure function of experimental data, best searching algorithm, best sub-pixel interpolation method etc. among a small circle, reduce to greatest extent inefficient matching operation, Optimized Matching quality.
Accompanying drawing explanation
The relation of Fig. 1 source images of the present invention and target image;
Fig. 2 correlated measure function surface of the present invention;
Fig. 3 the present invention removes the measure function curved surface of main peak;
The level line of Fig. 4 measure function curved surface of the present invention;
The isocontour projection of Fig. 5 measure function curved surface of the present invention.
Embodiment
Below in conjunction with accompanying drawing, the present invention is described in further details.
The matching effect evaluation method concrete steps of a kind of two dimensional image of the present invention are as follows:
(1) as shown in Figure 1, establish I={I (i, j) | 1≤i≤m, 1≤j≤n} is original image, f={f (i, j) | f (i, j)=I (i, j), m
1≤ i≤m
2, n
1≤ j≤n
2be truncated picture template wherein, and T={T (i ', j ') | 1≤i '≤k, 1≤j '≤l} is target image, intercepts subdomain g={g (i ', j ') in T | g (i ', j ')=T (i ', j '), k
1≤ i '≤k
2, l
1≤ j '≤l
2, when two width image taking yardsticks are when identical, get k
2-k
1=n
2-n
1and l
2-l
1=n
2-n
1;
(2), when template f and target subdomain g being compared, the gray-scale value of corresponding point position is respectively f (i, j) and g (i+u, j+v), wherein u=k
1-m
1, v=l
1-n
1, (u, v) is relative displacement vector between the picture of g and f, correspondingly g can be expressed as to g
(u, v), according to certain related algorithm, can set up the correlated measure ρ between image, calculate the correlated measure value of g and f, be denoted as ρ (u, v); Change u and v value, be equivalent to choose different target subdomains in T, each target subdomain has formed correlated measure function ρ=ρ (u, v) with respect to the correlated measure value of f; Have many algorithms based on estimating of Image gray correlation, in light of the circumstances, adopt suitable algorithm, can make correlated measure function ρ (u, v) have one obvious unimodal, the corresponding subdomain of its peak value
just comprise identical content with template f, as shown in Figure 2;
(3) calculate following six evaluation indexes:
1. simple signal to noise ratio (S/N ratio)
wherein C
mfor related coefficient maximal value (theoretical value or observation mean value),
for the outer related coefficient average of main peak, σ
cfor the outer related coefficient standard deviation of main peak,
with σ
ccorresponding system deviation and noise level while being images match, so calculated after must main peak being removed in measure function curved surface, according to the feature of related coefficient curved surface, should be centered by preliminary registration center, remove a square area in correlation matrix, the length of side of this square area is value from 10～30 pixel values, as shown in Figure 3;
2. main peak bottom section area S
l, border is taken as and makes
level line;
3. the accurate summit area of section of main peak S
t, border is taken as and makes C=C
t=C
l+ 0.7 (C
m-C
l)=0.7C
m+ 0.3C
llevel line;
4. curved surface main peak smoothing factor
be defined as
wherein
(i=L, M, H, T) is each contour plane normalization circularity, S
ifor i the contour plane area of curved surface C, L
ifor i the level line girth of curved surface C, F
lcorresponding C value is taken as C
l, F
tcorresponding C value is taken as C
t, definition F
mcorresponding C value is C
m=C
l+ 0.2 (C
m-C
l)=0.2C
m+ 0.8C
l, F
hcorresponding C value is C
h=C
l+ 0.4 (C
m-C
l)=0.4C
m+ 0.6C
l, to different correlation formulas, the number of contour plane and height can be chosen flexibly.
If Fig. 4 is some level lines of certain measure function curved surface, Fig. 5 is these isocontour projections.
5. the average computation time
be defined as the averaging time of an images match, in the time not considering searching algorithm and interpolation calculation, can have adopted the averaging time of calculating a measure function value
its unit is chosen as machine clock number, but for convenience of being generally taken as actual CPU time;
6. mate the average of absolute error vector mould
its unit can be pixel value, also can be scaled actual distance value;
Index 1.～be 4. about (m
1, m
2, n
1, n
2)
tstochastic variable, calculate and there is larger randomness and have little significance by single observed reading, therefore it is in fact all the result of repeatedly observing and getting average, according to applicant's experience, samples 50 times and the average of above calculating has statistical stability.
The isocontour method for drafting of related coefficient curved surface is: if it is higher to carry out the picture quality of registration, available bilinear interpolation is drawn all level lines, and the higher finger precision of images of picture quality is here higher, noise level is lower etc.; If it is lower to carry out the picture quality of registration, except corresponding C
tlevel line outside, all the other level lines adopt bilinear interpolations to draw; Using the extreme point of related coefficient as preliminary registration center, first near the correlation matrix preliminary registration center is carried out to two-dimentional cubic spline value of disclosing, draw corresponding C according to the correlation matrix after interpolation
tlevel line.
(4) utilize above 6 parameters to evaluate the matching effect of figure, evaluation criterion is:
1. SSNR is the bigger the better in preset range, and larger SSNR means that main peak is more outstanding, searches main peak easier;
2. S
lin preset range, be the bigger the better, this index is larger, more easily searches main peak and is conducive to improve the speed that searches main peak; Test piece deformation is hour desirable a
0=2.5; In the time that test piece deformation is large or image resolution ratio is lower, a
0=2.0 just can reach the effect of good screening main peak;
3. larger SSNR and less S
tmean template identified more reliably, but S
tbe not the smaller the better, be advisable with 6～8 pixels;
4.
be the bigger the better;
5.
the smaller the better;
6.
the smaller the better, while lacking the theoretical model of image, can not calculate;
(5) take the circumstances into consideration to choose some or all above indexs and can carry out following work:
1. in the situation that all the other conditions are identical, select different measure functions to carry out Matching Experiment, select most suitable measure function according to the performance of evaluation index;
2. in the situation that all the other conditions are identical, select different templates shape and yardstick to carry out Matching Experiment, determine suitable shape of template and yardstick according to the performance of evaluation index;
3. in the situation that all the other conditions are identical, select different searching method search optimal match point, according to the performance of evaluation index to selecting most suitable searching method;
4. multinomial the combining in above three work unified, for example evaluate different " measure function+template yardstick " combination matching effect and select suitable combination.
Method of the present invention is an assessment indicator system that comprises 6 parameters, and what the simple larger main peak of signal to noise ratio (S/N ratio) was missed may be less; Main peak bottom section area more greatly more easily searches main peak; The accurate summit area of section of moderate main peak and larger main peak smoothing factor are conducive to obtain higher matching precision, and the average computation time, less coupling was faster, and the precision of the less coupling of average of coupling absolute error vector mould is higher.Six evaluation indexes of applying in a flexible way can be utilized the auxiliary definite best correlated measure function of experimental data, best searching algorithm, best sub-pixel interpolation method etc. among a small circle, reduce to greatest extent inefficient matching operation, Optimized Matching quality.
Claims (3)
1. a matching effect evaluation method for two dimensional image, is characterized in that: step is as follows:
(1) establish I={I (i, j) | 1≤i≤m, 1≤j≤n} is original image, f={f (i, j) | f (i, j)=I (i, j), m
1≤ i≤m
2, n
1≤ j≤n
2be truncated picture template from described original image, T={T (i ', j ') | 1≤i '≤k, 1≤j '≤l} is target image, intercepts target subdomain g={g (i ', j ') in T | g (i ', j ') and=T (i ', j '), k
1≤ i '≤k
2, l
1≤ j '≤l
2, in the time that described original image is identical with target image shooting yardstick, k
2-k
1=m
2-m
1and l
2-l
1=n
2-n
1;
(2), when described image template f and target subdomain g being compared, the gray-scale value of corresponding point position is respectively f (i, j) and g (i+u, j+v), wherein u=k
1-m
1, v=l
1-n
1, (u, v) is relative displacement vector between the picture of g and f, and g is expressed as to g
(u, v), according to corresponding related algorithm, set up the correlated measure C between image, calculate the correlated measure value of g and f, be denoted as C (u, v); Change u and v value, choose exactly different target subdomains in T, each target subdomain has formed correlated measure function C=C (u, v) with respect to the correlated measure value of f;
(3) calculate following six evaluation indexes:
1. simple signal to noise ratio (S/N ratio)
wherein C
mfor correlated measure function maximal value, this correlated measure function maximal value is theoretical value or observation mean value,
for main peak relevant measure function average outward, σ
cfor main peak relevant measure function standard deviation outward;
2. main peak bottom section area S
l, border is taken as and makes
level line;
3. the accurate summit area of section of main peak S
t, border is taken as and makes C=C
t=C
l+ 0.7 (C
m-C
l)=0.7C
m+ 0.3C
llevel line;
4. curved surface main peak smoothing factor
be defined as
wherein
(i=L, M, H, T) is each contour plane normalization circularity, S
ifor i contour plane area of the curved surface of function C, L
ifor i level line girth of the curved surface of function C, F
lthe value of corresponding correlated measure function is C
l, F
tthe value of corresponding correlated measure function is C
t, F
mthe value of corresponding correlated measure function is C
m=C
l+ 0.2 (C
m-C
l)=0.2C
m+ 0.8C
l, F
hthe value of corresponding correlated measure function is C
h=C
l+ 0.4 (C
m-C
l)=0.4C
m+ 0.6C
l;
5. the average computation time
for completing the averaging time of an images match, in the time not considering searching algorithm and interpolation calculation, adopt the averaging time of calculating a measure function value
its unit is taken as CPU time;
6. mate the average of absolute error vector mould
its unit is pixel value or is scaled actual distance value;
Described evaluation index 1.～be 4. about (m
1, m
2, n
1, n
2)
tstochastic variable, to evaluation index 1.～4. be calculated as rule of thumb sample 50 times above calculate average;
(4) evaluation criterion of according to 6 evaluation indexes in described step (3), the matching effect of two dimensional image being evaluated is:
1. SSNR is the bigger the better in its preset range, and larger SSNR means that main peak is more outstanding, easier with threshold method location main peak;
2. S
lin its preset range, be the bigger the better, this index is larger, more easily searches main peak by Monte Carlo analysis, genetic algorithm, is conducive to improve the speed that searches main peak; In the time that target image is gone up deformation rate in any direction and is less than 1% with respect to original image, get a
0=2.5; When target image when deformation rate >1% or signal noise ratio (snr) of image <40dB, is got a with respect to original image at least one direction
0=2.0 screening main peaks;
3. larger SSNR and less S
tmean template identified more reliably, less S
trefer to S
tspan be 6～8 pixels, larger SSNR refers to SSNR>3;
4.
be the bigger the better,
5.
it is the smaller the better,
6.
the smaller the better, while lacking the theoretical model of image, do not calculate,
(5) according to the evaluation criterion in step (4), the matching effect of two dimensional image is evaluated, the work of carrying out comprises:
1. in the situation that all the other conditions are identical, select different measure functions to carry out Matching Experiment, select most suitable measure function according to the performance of evaluation index;
2. in the situation that all the other conditions are identical, select different templates shape and yardstick to carry out Matching Experiment, determine most suitable shape of template and yardstick according to the performance of evaluation index;
3. in the situation that all the other conditions are identical, select different searching method search optimal match point, according to the performance of evaluation index to selecting most suitable searching method;
4. at least two in above three work are combined and unified.
2. the matching effect evaluation method of two dimensional image according to claim 1, is characterized in that, the isocontour method for drafting of correlated measure function surface is:
(1) higher if carry out the picture quality of registration, draw all level lines with bilinear interpolation, picture quality higher finger signal noise ratio (snr) of image >50dB and target image are here gone up deformation rate <1% in any direction with respect to original image;
(2) lower if carry out the picture quality of registration, corresponding C
tisocontour method for drafting be: using the extreme point of correlated measure function as preliminary registration center, first near the correlated measure Jacobian matrix preliminary registration center is carried out to two-dimentional cubic spline interpolation, the length of side of interpolation area is value between 5～8, draws corresponding C according to the correlated measure Jacobian matrix after interpolation
tlevel line; All the other each isocontour draftings adopt bilinear interpolation to draw.
3. the matching effect evaluation method of two dimensional image according to claim 1, is characterized in that, calculates main peak relevant measure function average outward
with main peak relevant measure function standard deviation sigma outward
ctime, region within the scope of main peak in correlated measure function surface is removed and calculated, and according to the feature of correlated measure function surface, centered by preliminary registration center, remove a square area in correlated measure Jacobian matrix, the length of side of described square area is value in 10～30 pixels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088955.7A CN103793919A (en) | 2014-03-12 | 2014-03-12 | Two-dimensional image matching effect evaluation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410088955.7A CN103793919A (en) | 2014-03-12 | 2014-03-12 | Two-dimensional image matching effect evaluation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103793919A true CN103793919A (en) | 2014-05-14 |
Family
ID=50669542
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410088955.7A Pending CN103793919A (en) | 2014-03-12 | 2014-03-12 | Two-dimensional image matching effect evaluation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103793919A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110134909A (en) * | 2019-05-23 | 2019-08-16 | 武汉轻工大学 | Surface Rendering approach, equipment, storage medium and device |
| CN111738349A (en) * | 2020-06-29 | 2020-10-02 | 重庆紫光华山智安科技有限公司 | Detection effect evaluation method and device of target detection algorithm, storage medium and equipment |
| WO2023272516A1 (en) * | 2021-06-29 | 2023-01-05 | Siemens Aktiengesellschaft | A method, device, system and storage medium for scheme recommendation |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH1021393A (en) * | 1996-06-28 | 1998-01-23 | Mitsubishi Electric Corp | Image processor |
| CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
-
2014
- 2014-03-12 CN CN201410088955.7A patent/CN103793919A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH1021393A (en) * | 1996-06-28 | 1998-01-23 | Mitsubishi Electric Corp | Image processor |
| CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
Non-Patent Citations (3)
| Title |
|---|
| G.C.JIN 等: "Digital Speckle Correlation Method and Its Application to Micro-Mechanics", 《PROCEEDING OF SPIE》 * |
| L.B. MENG 等: "Application of iteration and finite element smoothing technique for displacement and strain measurement of digital speckle correlation", 《OPTICS & LASERS IN ENGINEERING》 * |
| 金观昌 等: "数字散斑相关技术进展及应用", 《实验力学》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110134909A (en) * | 2019-05-23 | 2019-08-16 | 武汉轻工大学 | Surface Rendering approach, equipment, storage medium and device |
| CN110134909B (en) * | 2019-05-23 | 2023-05-12 | 武汉轻工大学 | Curved surface drawing method, equipment, storage medium and device |
| CN111738349A (en) * | 2020-06-29 | 2020-10-02 | 重庆紫光华山智安科技有限公司 | Detection effect evaluation method and device of target detection algorithm, storage medium and equipment |
| CN111738349B (en) * | 2020-06-29 | 2023-05-02 | 重庆紫光华山智安科技有限公司 | Detection effect evaluation method and device of target detection algorithm, storage medium and equipment |
| WO2023272516A1 (en) * | 2021-06-29 | 2023-01-05 | Siemens Aktiengesellschaft | A method, device, system and storage medium for scheme recommendation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108549873B (en) | Three-dimensional face recognition method and three-dimensional face recognition system | |
| CN102236794B (en) | Recognition and pose determination of 3D objects in 3D scenes | |
| CN106530347B (en) | Stable high-performance circle feature detection method | |
| US10424078B2 (en) | Height measuring system and method | |
| JP6114052B2 (en) | Point cloud analysis processing device and point cloud analysis processing program | |
| CN110672020A (en) | Stand tree height measuring method based on monocular vision | |
| CN105354841B (en) | A kind of rapid remote sensing image matching method and system | |
| WO2015149712A1 (en) | Pointing interaction method, device and system | |
| WO2019062843A1 (en) | Ultrasound image processing method and system | |
| CN108765475B (en) | Building three-dimensional point cloud registration method based on deep learning | |
| CN110223355B (en) | Feature mark point matching method based on dual epipolar constraint | |
| CN114972459B (en) | Point cloud registration method based on low-dimensional point cloud local feature descriptor | |
| CN103295232A (en) | SAR (specific absorption rate) image registration method based on straight lines and area | |
| Shi et al. | A visual circle based image registration algorithm for optical and SAR imagery | |
| CN107463873B (en) | Real-time gesture analysis and evaluation method and system based on RGBD depth sensor | |
| CN104268880A (en) | Depth information obtaining method based on combination of features and region matching | |
| Tian et al. | An experimental evaluation method for the performance of a laser line scanning system with multiple sensors | |
| CN103218809A (en) | Image measuring method of pearl length parameter | |
| Jun et al. | Point cloud registration algorithm based on NDT with variable size voxel | |
| CN103793919A (en) | Two-dimensional image matching effect evaluation method | |
| WO2021114026A1 (en) | 3d shape matching method and apparatus based on local reference frame | |
| Yang et al. | Auto-recognition Method for Pointer-type Meter Based on Binocular Vision. | |
| CN102201060B (en) | Method for tracking and evaluating nonparametric outline based on shape semanteme | |
| CN112508885B (en) | Method and system for detecting three-dimensional central axis of bent pipe | |
| JP6171781B2 (en) | Gaze analysis system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140514 |