CN108389186A - The point spread function number estimation method on arbitrary shape curve sword side - Google Patents
The point spread function number estimation method on arbitrary shape curve sword side Download PDFInfo
- Publication number
- CN108389186A CN108389186A CN201810087171.0A CN201810087171A CN108389186A CN 108389186 A CN108389186 A CN 108389186A CN 201810087171 A CN201810087171 A CN 201810087171A CN 108389186 A CN108389186 A CN 108389186A
- Authority
- CN
- China
- Prior art keywords
- spread function
- edge
- sample
- point
- sword side
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
Abstract
The present invention discloses a kind of point spread function number estimation method on arbitrary shape curve sword side, including：Edge detection is carried out to degeneration blade figure, to obtain the sword side of the degeneration blade figure；Using moving window, the sword side of the degeneration blade figure is sampled, to obtain the edge-spread function sample of described pair of change blade figure；The edge-spread function sample is screened, and to the edge-spread function sample after screening into row interpolation and re-sampling operations；Differential is carried out to the edge-spread function sample after interpolation and re-sampling operations, to obtain line spread function sample；The line spread function sample is intercepted and is carried out Gaussian function fitting, to obtain point spread function numerical example.By means of the invention it is possible to Y-PSNR is improved, and noiseproof feature is good.
Description
Technical field
The present invention relates to the technical field of method of estimation more particularly to a kind of point spread functions on arbitrary shape curve sword side
Method of estimation.
Background technology
Image resolution ratio is a critical index of evaluation image quality, in practical problem, it is desirable to the image of acquisition
It is very naturally with higher resolution ratio, better quality.Since infrared remote sensing imaging has passive work, anti-interference
By force, the features such as target identification ability is strong, all weather operations, in terms of having been widely used with military surveillance, monitoring and guidance.With it is visible
Light remote sensing images are compared, and infrared remote sensing image has that resolution ratio is low, contrast is low, edge blurry, the low, complicated component of signal-to-noise ratio etc.
Disadvantage.
The accurate estimation of the point spread function (point spread function, PSF) of optical system is to improve image light
Learn one of the prerequisite of quality.Under many situations, the point spread functions of remote sensing images can only utilize imaging system parameter or
The image of acquisition recognizes system degradation function to obtain accurate PSF.Often utilize the blade target in remote sensing images, line
Property pulse target and square wave target obtain accurate PSF.But these conventional methods are required to be laid with target in advance, and it is tight
The shape of lattice control target and the angle put.In order to overcome the dependence to target, many methods, such as the blind warp of multichannel
Product, the method based on joint transform, the method based on the blind recovery of full variational regularization are suggested and carry out research test.These sides
Although method time overhead is small, but to noise quite sensitive.
Although proposing a variety of schemes for solving curved edges side and obtaining PSF, the think of of most of method recent years
Road can be attributed to two kinds：Based on ISO12233 and its a series of improved methods；Pixel point gray value and point are arrived into bending
The method being further processed after the distance composition set on sword side.Traditional recognition status is directly by edge-spread function (edge
Spread function, ESF) the obtained line spread function (line spread function, LSF) of family of curves's differential carries out
The alignment and sample resampling of family of curves finally obtain unique LSF curves.But this is that the tangent line each put in curve is full
It is just more accurate when foot ideal inclination angle, otherwise it can equally make abscissa elongation that PSF be caused to survey with the situation under angled straight lines sword side
Amount is inaccurate.Curve-fitting method arrives curved edges side using quadratic polynomial fitting curved edges side, by pixel point gray value and point
Distance value be added in distance-gray scale set, finally use the element fitting Fermi functions in set bent as actual ESF
Line.On the one hand, which has used simulated annealing to cause iterations excessive, and time overhead is big.Next sword side edge is very
Hardly possible is fitted with quadratic function perfection.In fact, using the gray value of pixel point and point to curved edges back gauge as ESF samples
With reference to inherently worth discussion.General sword edge image is divided into left right model (as shown in Figure 1a) and upper mo(u)ld bottom half (as shown in Figure 1 b),
For convenience, acquiescence is discussed with left right model, the sword side of upper mo(u)ld bottom half and so on.
Invention content
The main purpose of the present invention is to provide a kind of point spread function number estimation methods on arbitrary shape curve sword side, with solution
It is certainly of the existing technology greatly to be limited by target and the problem for causing estimate error bigger than normal be stretched because of coordinate.
To solve the above problems, the embodiment of the present invention provides a kind of point spread function estimation side on arbitrary shape curve sword side
Method, including：Edge detection is carried out to degeneration blade figure, to obtain the sword side of the degeneration blade figure；Using moving window, to institute
The sword side for stating degeneration blade figure is sampled, to obtain the edge-spread function sample of described pair of change blade figure；To the edge
Spread function sample is screened, and to the edge-spread function sample after screening into row interpolation and re-sampling operations；It is right
The edge-spread function sample after interpolation and re-sampling operations carries out differential, to obtain line spread function sample；To described
Line spread function sample is intercepted and is carried out Gaussian function fitting, to estimate point spread function.
According to the technique and scheme of the present invention, by carrying out edge detection to degeneration blade figure, to obtain the degeneration blade
The sword side of figure；Using moving window, the sword side of the degeneration blade figure is sampled, to obtain the side of described pair of change blade figure
Edge spread function sample；The edge-spread function sample is screened, and to the edge-spread function sample after screening
Originally into row interpolation and re-sampling operations；Differential is carried out to the edge-spread function sample after interpolation and re-sampling operations, with
Obtain line spread function sample；The line spread function sample is intercepted and carried out Gaussian function fitting, is expanded with estimation point
Dissipate function.Thus, which Y-PSNR can averagely be improved 10dB or more by the embodiment of the present invention, and noiseproof feature is good.
Description of the drawings
Attached drawing described herein is used to provide further understanding of the present invention, and is constituted part of this application, this hair
Bright illustrative embodiments and their description are not constituted improper limitations of the present invention for explaining the present invention.In the accompanying drawings：
Fig. 1 a are the sword edge graphs on left right model sword side；
Fig. 1 b are the sword edge graphs on upper mo(u)ld bottom half sword side；
Fig. 2 is the calculating schematic diagram of recognition status；
Fig. 3 a are the schematic diagrames on 45 ° of inclination angles sword side before degenerating；
Fig. 3 b are the schematic diagrames on 45 ° of inclination angles sword side after degenerating；
Fig. 4 a are the schematic diagrames that pixel gray value solves after degenerating under discrete conditions；
Fig. 4 b are the schematic diagrames of the theoretical foundation of sciagraphy under discrete conditions；
Fig. 5 a are the schematic diagrames of the value of each sub-pixel locations of PSF of sciagraphy under the conditions of true sub-pix；
Fig. 5 b are the schematic diagrames of the gray value of sub-pix pixel point after the degeneration of sciagraphy under the conditions of true sub-pix；
Fig. 5 c are the schematic diagrames of the theoretical foundation of sciagraphy under the conditions of the sub-pix of sciagraphy under the conditions of true sub-pix；
Fig. 6 a be curve sword while sciagraphy the shortest distance react be this apart from lower straight line sword while the case where；
Fig. 6 b are the schematic diagrames of the sampled point for being suitable as ESF samples of curve sword side sciagraphy；
Fig. 6 c are the schematic diagrames of the point spread function number estimation method on arbitrary shape curve sword side according to the ... of the embodiment of the present invention；
Fig. 7 is the flow chart of the point spread function number estimation method on arbitrary shape curve sword side according to the ... of the embodiment of the present invention；
Fig. 8 is tolerance domain according to the ... of the embodiment of the present invention schematic diagram；
Fig. 9 a are the simulation drawings that sampling line number according to the ... of the embodiment of the present invention is chosen；
Fig. 9 b are the simulation drawings on curve sword according to the ... of the embodiment of the present invention side 1；
Fig. 9 c are the simulation drawings on curve sword according to the ... of the embodiment of the present invention side 2；
Figure 10 a, Figure 10 b, Figure 10 c, Figure 10 d are final LSF curves according to the ... of the embodiment of the present invention and true LSF respectively
The comparison figure of curve；
Figure 11 a, Figure 11 b are the simulation drawing of anti-noise ability test according to the ... of the embodiment of the present invention respectively；
Figure 12 a are the sword border regions of actual scene test according to the ... of the embodiment of the present invention；
Figure 12 b are the estimated value and actual value comparison result of actual scene test according to the ... of the embodiment of the present invention；
Figure 12 c are part gray-scale maps before the Wiener filtering that actual scene according to the ... of the embodiment of the present invention is tested；
Figure 12 d are part gray scales after the use estimated value Wiener filtering that actual scene according to the ... of the embodiment of the present invention is tested
Figure.
Specific implementation mode
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with drawings and the specific embodiments, to this
Invention, which is done, to be further described in detail.
First, if clear image is e (x, y), degraded image is o (x, y), and remote sensing images degradation model can be described as public affairs
Formula (1), as follows：
O (x, y)=e (x, y) * h (x, y)+n (x, y), (1)
Wherein, h (x, y) is point spread function, and n (x, y) is additive noise, and * indicates convolution algorithm.
Under normal circumstances, point spread function (point spread function, PSF) has separability, such as formula
(2) shown in：
H (x, y)=hx(x)hy(y), (2)
Wherein, hx(x) be level sampling direction PSF, hy(y) it is the PSF on Vertical Sampling direction, and hx(x) and hy
(y) it is referred to as line spread function (line spread function, LSF).In addition, edge-spread function (edge spread
Function, ESF) as shown in formula (3)：
G (x)=U (x) hx(x), (3)
Wherein, U (x) is the jump function on sword edge direction, and g (x) is the output letter on sword edge direction
Number, is referred to as edge-spread function, shown in the relationship such as formula (4) with line spread function：
Recognition status obtains PSF and tilts recognition status calculating PSF using ISO12233 under ideal inclination angle, as shown in Figure 2.It is basic to think
Want that differential obtains LSF samples, that is, one-dimensional point spread function after ESF samples are calculated using the obtained gray-scale map of sampling
Number.But the tradition sword side method is suitable only for the ideal sword border region that the sampling interval is a picture element unit.
The present invention is based on sciagraphies to propose a kind of point spread function number estimation method on arbitrary shape curve sword side.Also, this
Invention from angled straight lines analysis sciagraphy in physical significance.The degeneration of image can be expressed as discrete convolution, such as public
Shown in formula (5)：
Assuming that hhLength is 5, hh=[u1,u2,u3,u2,u1].Fig. 3 a are shown at the sword side edge at 45 ° of inclination angle, dark in figure
Part indicates the higher part of gray value.Its pixel after above-mentioned PSF degenerations is as shown in Figure 3b.
Since color discrimination is not high, to clearly show that, the number in Fig. 3 b is that gray value is sorted it from small to large
Afterwards as a result, number it is bigger, it is higher to represent gray value.Straight line in Fig. 3 b is theoretic sword side edge.
Wherein, the gray value of the point of -3 positions is formula (6), as follows：
And the gray value of the point of -4 positions is formula (7), as follows：
Have under discrete conditions to draw a conclusion：
Under discrete conditions, the gray value of pixel is equal to the region for the preceding PSF sizes centered on the point of degenerating after degeneration
Gray value and PSF corresponding position value dot products after and, as shown in fig. 4 a.
The principle of sciagraphy can be understood as under discrete conditions：It keeps the distance at pixel point to sword side edge constant, moves back
The result summed after the gray value and PSF corresponding position value dot products in the region of the PSF sizes before changing centered on the point is constant.More
Further, since degeneration front edge edge both ends gray value is changeless, it can be understood as：Keep pixel point to sword while while
The distance of edge is constant, before degenerating the gray value in the region of PSF sizes centered on the point and remain unchanged.But from
It the theoretical foundation of sciagraphy and is unsatisfactory in the case of dissipating, can intuitively be observed from Fig. 4 b, keep pixel point to sword side
The distance h at edge is constant, and the both ends area at sword side edge is but changed.
Cause this contradictory the reason is that true PSF with a square formation is very inaccurate to describe, and should be
With certain point for circle of the center of circle to external diffusion, and the value of the point on the PSF only distance dependent with the centers this distance PSF, such as
Shown in Fig. 5 a.Under the conditions of true sub-pix, the gray value g of sub-pix pixel point is formula (8) after degeneration, as follows：
Wherein, Ω1、Ω2Before respectively degenerating the circle institute covering position of PSF sizes centered on pixel point by sword while while
Two pieces of regions that edge is divided, as shown in Figure 5 b, g1、g2Ω before respectively degenerating1、Ω2Gray value corresponding to region, F are represented
PSF each position gray value function after normalization.
In the case of straight line sword side, since the straight line of the shortest distance of sword while with to sword must be vertical, so only needing
To pass through the distance of the size of PSF and sub-pix pixel point to sword side, so that it may to calculate the pixel after degeneration by formula (8)
The gray value g of point.Meanwhile no matter sword side rotated into any angle, Ω in the circle of the PSF sizes before degenerating centered on the point
1, the integrated value of 2 parts Ω will not change, as shown in Figure 5 c, that is to say, that the distance of the pixel point to sword side straight line can be with
It is considered the abscissa distance apart from sword side edge under ideal inclination angle.Here it is sciagraphy carry out ESF specimen samples it is theoretical according to
According to.
On this basis consider curved edges side when it finds that：The simple vertical range for finding every bit to curve sword side is made
Will produce huge error for the reference of the abscissa of ESF samples because this distance reaction be only this apart from lower straight line
The case where sword side, as shown in Figure 6 a, the problem of this also demonstrates existing algorithm.And since curvilinear equation is difficult to find that suitably
Expression formula, it is difficult to determine the part of these coverings by calculating.In fact, it is only necessary to accurate part ESF samples are obtained, and
Point after need not each degenerating is the sampled point of ESF samples.If (curvature half more gentle in curve sword side can be selected
Diameter is larger) portion point, as shown in Figure 6 b, with straight line sword while projection replace curve sword while project, at these points, point
Distance to sword side straight line approximate can reflect the part that the circle of the PSF sizes centered on the point covered, further
It can more be accurately obtained preferable ESF samples by interpolation and resampling.
Based on this thought, the present invention proposes the curve sword side method for solving of moving window projection, chooses every time certain
Row (column) fitting boundary curve be straight line and carry out the sampling of ESF samples using arbitrary angle straight line sword side sciagraphy, then
The row (column) of (right side) moving step length sampling downwards continues the sampling of ESF samples until completing the sampling to all sample rows, such as schemes
Shown in 6c, final ESF samples are obtained after finally removing the sampled point that those significant discomfort cooperations are ESF samples.Again to obtaining
ESF samples be further processed.
Hereinafter, by being illustrated to the point spread function number estimation method on the arbitrary shape curve sword side of the present invention.Fig. 7 is this
The flow chart of the point spread function number estimation method on the arbitrary shape curve sword side of inventive embodiments.
In step S702, edge detection is carried out to degeneration blade figure, to obtain the sword side of the degeneration blade figure.
In step S704, using moving window, the sword side of the degeneration blade figure is sampled, it is described right to obtain
Change edge-spread function (edge spread function, ESF) sample of blade figure.
In the present embodiment, the sampling of edge-spread function sample is arbitrary shape curve sword side proposed by the present invention
The core of point spread function number estimation method.Wherein, moving window includes line number size L, each window that each sampling is chosen
Mouth mobile step-length S, each projection centre C, that is, moving window is provided by choosing L, S, C.
Single part to be treated is exactly the selection line number in moving window, as fig. 6 c.Also, first use minimum
The coordinate that square law chooses the sampling of the moving window marginal point of line number carries out fitting a straight line, recycles arbitrary angle sword
Side sciagraphy samples the edge-spread function sample of selection.For example, if fitting a straight line is with vertical direction inclination angle
θ, as fig. 6 c.The coordinate of certain point of the edge-spread function sample of sampling is (i, j), and wherein i is columns where sampled point,
J is place line number, then the coordinate P of the spot projection position is P=j*cos θ ± i*sin θ (for the straight line of lower-left to upper right, symbol
It number takes negative, straight line symbol left to bottom right is taken just).
Then, after the edge-spread function sample for having sampled single, by moving window, (right side) moving step length is adopted downwards
Sample row continues to sample, that is, is sequentially sampled to the sword side of the degeneration blade figure, until completing to adopt to all sample rows
Sample.When calculating projected position coordinate in each moving window, the point on straight line that acquiescence marginal point is fitted is located at projection
Center, need projection centre being aligned in finally obtained edge-spread function sample, that is, according to the movement
The multiple sample is aligned by the projection centre of window.The projected position coordinate P of all the points is added one thus
Deviation, if edge corresponding coordinate in the first row sword side is (i in moving window every timeo,jo), deviation is C- (jo*cosθ±
io*sinθ).At this point, the projected position coordinate value of certain point (i, j) should be just modified in corresponding moving window：P*=j*cos θ ±
i*sinθ+C-(jo*cosθ±io*sinθ).By P*Sampled value with corresponding points gray value as one group of edge-spread function sample
Storage.It is worth noting that, the data with a line may sample repeatedly according to different inclination angles, but this is bent in knife edge
The ESF sample projected position coordinates obtained with a line in the case that line is continuous and radius of curvature is larger are almost the same.
In step S706, the edge-spread function sample is screened, and to the edge-diffusion after screening
Function sample is into row interpolation and re-sampling operations.
In the present embodiment, there can be the point being not suitable for as sample in the sampled value of edge-spread function sample, therefore,
Being screened to the edge-spread function sample can be by discontinuous or radius of curvature in knife edge curve compared with where dot
The sampled value of sampling row is rejected.Furthermore, it is contemplated that the problem of true environment and equipment, obtained edge-spread function sample
It also can there is a large amount of abnormal points and noise spots in this sampled value.These points deviate considerably from true edge spread function curve.
Before further processing edge-spread function sample, need to reject these abnormal points and noise spot.Edge-spread function is bent
On one minizone of line, can approximation regard straight line as, that is, sampled point is fitted into straight line.
Also, the present embodiment is that equidistant tolerance domain is built in straight line both sides by standard of the distance to fitting a straight line, such as
Shown in Fig. 8.Sampled point other than the range in tolerance domain is considered as abnormal point or noise spot.The specific practice of the present embodiment is first
Edge-spread function sample is segmented on the basis of unit pixel, and every section of edge-spread function sample is extended to both ends
Half unit pixel, then straight line is carried out to the edge-spread function sample in every section of edge-spread function sample and its extended area and is intended
It closes, then judges the noise spot of edge-spread function sample and rejecting in every section of edge-spread function sample and its extended area,
Then the operation of cancelling noise point is re-started to the sample point after rejecting abnormalities point to exclude abnormal point to sample fitting straight line
Influence.Wherein, the extended area can beWherein σ is that the radius of extended area is, △ liIt is i-th
The distance i of sampled point to fitting a straight line is the positive integer more than 0, and n is the number of sampled point.
After to edge-spread function sample excluding gross error, there are error, another party for one side sword side edge detection
It will appear missing values in the edge-spread function sample of face, this will influence further line spread function specimen sample.Therefore, to institute
State operation of the edge-spread function sample after edge-spread function sample is screened into resampling after row interpolation.Specific step
It is rapid as follows：To the edge-spread function sample after excluding gross error, into row interpolation, (purpose of interpolation is in order to by rejecting abnormalities point
It is filled with the missing values of the edge-spread function sample after noise spot).Wherein, first according to projected position coordinate P*Size,
Edge-spread function sample is ranked up, then uses cubic spline interpolation in the interval of projected position coordinate P*
The gray value of edge-spread function sample is into row interpolation, and moving step length is 0.05 pixel.Then, the point that interpolation is obtained and
The edge-spread function sample is arranged according to the value P* of projected position again together (i.e. with former edge-spread function sample)
Sequence obtains new edge-spread function sample.Later, piecewise fitting is carried out to new edge-spread function sample and resampling is grasped
Make, that is, moving window obtains the resampling value of edge-spread function sample, the size of moving window is 1 pixel, in Moving Window
Cubic polynomial fitting is for example carried out to edge-spread function sample in mouthful, wherein using match value at moving window center as weight
Sampled value, and moving step length is 0.05 pixel.
In step S708, differential is carried out to the edge-spread function sample after interpolation and re-sampling operations, to obtain
Line taking spread function sample.
By formula (4) it is found that line spread function sample can be obtained come direct differentiation to edge-spread function sample.Separately
Outside, for discrete sample, line spread function sample can get formula (9), following institute by edge-spread function sample difference
Show：
Wherein, LSFiFor i-th of line spread function sample, ESFiFor i-th of edge-spread function sample, ESFi-1It is i-th-
1 edge-spread function sample, △ x are sample point interval.
In step S710, the line spread function sample is intercepted and carries out Gaussian function fitting, with estimation point
Spread function.
In the present embodiment, before being fitted processing to line spread function sample, line spread function sample is first estimated
The size of size.Also, the data handled in the algorithm that the present embodiment proposes are the differentiated line spread function sample samples of projection
Originally, the problem of therefore being not in deviation.Noise can interfere the accuracy of profile value (line spread function sample-size), therefore
The present embodiment sets a threshold value and is limited.To determine profile size, the line that acquisition is begun stepping through from left end (right end) expands
Dissipate function sample (sampling interval is 0.05 unit length).When certain point is both greater than some threshold value together with its adjacent 2 points
When, it is endpoint 1 (endpoint 2) by its coordinate record.Threshold value is usually set to 0.1 depending on noise size.By 2 coordinate of endpoint
It subtracts 1 coordinate of endpoint and downward rounding obtains profile value size.
In addition, carrying out tail portion interception to line spread function sample to weaken line spread function sample tail portion effect of jitter.For
Ensure that the accuracy of line spread function sample fitting, the final length for intercepting part should be less than the big of the profile value obtained before
It is small.Since the shape of line spread function sample is similar to Gaussian, in order to reduce the error of point spread function estimation, to what is obtained
Line spread function sample carries out one-dimensional Gaussian function fitting, and model of fit can be expressed as formula (10), as follows：
, h (x)=Aexp [- (x-u)2/2σ2], (10)
Wherein, h (x) is point spread function, and A is the peak value of function, and x is LSF samples, and u it is expected for normal distribution, and σ is just
The standard deviation of state distribution.
Also, above-mentioned model can be solved according to the principle of least square.Finally using the central point of fitting result as
The size of the center interception profile value of interception obtains final line spread function vector.By formula (2) formula, point spread function matrix number
It can be obtained by the line spread function vector product of both direction, be exactly final point spread function after normalization.
It is above-mentioned it is stated that the arbitrary shape curve sword side of the present invention point spread function number estimation method, provided below one
A little examples verify the applicability and feasibility of the above method.
Experiment test is the center constant term C of projection, the line number sampled every time respectively it needs to be determined that three hyper parameters
The L and step-length S of frame movement.The constant term C selections of projection centre are relatively random, it is only necessary to after ensureing each sample projection
Coordinate is just.
When relatively sword side algorithm performance, commonly uses Y-PSNR (Peak Signal To Noise Ratio) and make
For the objective evaluation index of algorithm performance.The curved edges side simulation drawing of use, as shown in Fig. 9 b and Fig. 9 c.The line number chosen every time
Size L and the size of picture and PSF have a very large relationship.The present embodiment is first by taking the simulation drawing of 128*128 as an example, to choose
Line number be independent variable, using the Y-PSNR of estimated PSF and actual value as dependent variable, finally obtained result such as Fig. 9 a institutes
Show.In fig. 9 a, three curves are using 2 as step-length PSF sizes respectively for 5*5, every sub-sampling line number of 9*9 and 15*15
With the relational graph of Y-PSNR.
As can be seen that optimal selections line number is unrelated with the size of PSF in Fig. 9 a, every time sampling line number 10 or so when
Waiting Y-PSNR can be more stable, but sampling line number is too small or big city excessively has an impact result.Sample line number mistake
Small to be changed greatly in some edges or discrete point when easy tos produce prodigious error, sampling line number is excessive then
Error can be generated during algorithm curve fitting a straight line influences the precision of experiment.Determine the line number sampled every time and PSF
Size it is unrelated after, the present embodiment is similarly tested with the simulation drawing of 64*64 again, and optimal selection line number is 5 or so.By
This, optimal selection line number is usually no more than 1/10th of dimension of picture.
For verify the present embodiment arbitrary shape curve sword side point spread function number estimation method feasibility in theory,
PSF curves are estimated to the curve sword side simulation drawing after following obscure is added with the method for the present embodiment.As shown in figure 9b, to curve
It is 15*15, the gaussian filtering that variance is 1 that sword side 1, which carries out size,；It is 9*9, the height that variance is 0.5 to carry out size to curve sword side 1
This filtering；It is 15*15, the gaussian filtering that variance is 0.5 to carry out size to curve sword side 1.As is shown in fig. 9 c, to curve sword side 2
Progress size is 15*15, and the gaussian filtering that variance is 1, the comparison figure of final LSF curves and true LSF curves is respectively as schemed
10a, Figure 10 b, Figure 10 c, analog result shown in Figure 10 d.As shown in figure 9b, the LSF that the method that the present embodiment proposes is estimated
Curve and actual value are completely superposed, and in varied situations, and the method that the present embodiment proposes can preferably estimate out LSF
Curve.
In addition, different curve sword sides and method progress contrast experiment are chosen, respectively with the curve sword side of two angles
Figure (such as Fig. 9 b and Fig. 9 c) tests the method that traditional recognition status, twice curve fitting method and the present embodiment propose, final
Experimental result is recorded in table 1.For two kinds of curve sword sides, traditional recognition status has a more stable effect, but curve matching
Method but occurs significantly fluctuating very much in measurement result, even will appear under some cases and measures serious inaccurate feelings
Condition.For example for curve sword side 1, in the case where 15*15 size variances are 1, peak value letter is dry than there was only 11.52dB；For curve
Sword side 2, in the case where 15*15 size variances are 1, peak value letter is dry than there was only 9.36dB.It follows that the present embodiment proposed
Method will be far superior to other two kinds of algorithms in all cases, and PSNR can averagely be improved on the basis of optimal effectiveness
10dB or so.
The Y-PSNR of table 1 PSF estimated values and actual value
In addition, being influence of the test noise for algorithm, the PSF for being respectively 1 with size 15*15 variances on curve sword side 2
Gaussian noise and random noise are added in simulation drawing after gaussian filtering, carrys out the anti-noise ability of testing algorithm, as a result respectively such as
Shown in Figure 11 a and Figure 11 b.After tested, although measurement accuracy is declined, the method that the present embodiment proposes remains to relatively accurately
PSF estimated values are obtained, and anti-noise ability is good.
In addition, tested for actual scene, scene after size is the gaussian filtering that 9*9 variances are 1, cut
The part of intermediate 64*64 sizes is taken to be tested using the point spread function number estimation method on arbitrary shape curve sword side, such as Figure 12 a
It is shown, the LSF curves finally measured and true LSF curve comparison figures, as shown in Figure 12b.Figure 12 c and Figure 12 d are to degenerate respectively
Figure carries out the result of the front and back same part intercepted of Wiener filtering using the PSF of algorithm estimation.It can be seen that under real scene,
Image can be obtained sword border region edge and often have discrete point, in this case, the method that the present embodiment proposes
True PSF still can be preferably estimated, and obtains good recovery effect.
In conclusion according to the technique and scheme of the present invention, by carrying out edge detection to degeneration blade figure, described in acquisition
The sword side of degeneration blade figure；Using moving window, the sword side of the degeneration blade figure is sampled, to obtain described pair of change knife
The edge-spread function sample of sword figure；The edge-spread function sample is screened, and the edge after screening is expanded
Function sample is dissipated into row interpolation and re-sampling operations；The edge-spread function sample after interpolation and re-sampling operations is carried out
Differential, to obtain line spread function sample；The line spread function sample is intercepted and is carried out Gaussian function fitting, to obtain
Take point spread function numerical example.Thus, which Y-PSNR can averagely be improved 10dB or more, and anti-noise by the embodiment of the present invention
It is functional.
Obviously, those skilled in the art should be understood that the three dimensional anisotropic evanescent wave field simulation of the above-mentioned present invention
Method and each step can be realized with general computing device, they can be concentrated on a single computing device, Huo Zhefen
Cloth is on network constituted by multiple computing devices, and optionally, they can be with the program code that computing device can perform come real
It is existing, it is performed by computing device it is thus possible to be stored in storage device.In this way, the present invention is not limited to any spies
Fixed hardware and software combines.Storage device is nonvolatile memory, such as：ROM/RAM, flash memory, magnetic disc, CD etc..
Example the above is only the implementation of the present invention is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.Any modification made by all within the spirits and principles of the present invention,
Equivalent replacement, improvement etc., should be included within scope of the presently claimed invention.
Claims (7)
1. a kind of point spread function number estimation method on arbitrary shape curve sword side, which is characterized in that include the following steps：
Edge detection is carried out to degeneration blade figure, to obtain the sword side of the degeneration blade figure；
Using moving window, the sword side of the degeneration blade figure is sampled, the edge to obtain described pair of change blade figure expands
Dissipate function sample；
The edge-spread function sample is screened, and to the edge-spread function sample after screening into row interpolation and
Re-sampling operations；
Differential is carried out to the edge-spread function sample after interpolation and re-sampling operations, to obtain line spread function sample；
The line spread function sample is intercepted and is carried out Gaussian function fitting, to estimate point spread function.
2. the point spread function number estimation method on arbitrary shape curve sword side according to claim 1, which is characterized in that described
Moving window includes that line number, the step-length and projection centre of window movement are chosen in sampling.
3. the point spread function number estimation method on arbitrary shape curve sword side according to claim 2, which is characterized in that described
Using moving window, the sword side of the degeneration blade figure is sampled, to obtain the edge-diffusion letter of described pair of change blade figure
The step of numerical example includes：
The coordinate for choosing the marginal point of line number to the sampling of the moving window using least square method carries out fitting a straight line；
Sequentially the sword side of the degeneration blade figure is sampled using the inclination recognition status of arbitrary angle, to obtain multiple samplings
Sample；
According to the projection centre of the moving window, the multiple sample is aligned, to obtain the degeneration blade
The edge-spread function sample of figure.
4. the point spread function number estimation method on arbitrary shape curve sword side according to claim 3, which is characterized in that institute
It includes that discontinuous or radius of curvature is adopted compared with dot place eventually by knife edge curve to state edge-spread function sample and carry out screening
The sampled value of sample row is rejected.
5. the point spread function number estimation method on arbitrary shape curve sword side according to claim 4, which is characterized in that institute
It states edge-spread function sample and screen and further include：
The edge-spread function sample of the degeneration blade figure is segmented on the basis of unit pixel, and every section of edge expands
It dissipates function sample and extends half unit pixel to both ends, then the edge in every section of edge-spread function sample and its extended area is expanded
It dissipates function sample and carries out fitting a straight line, judge abnormal point and noise spot and reject.
6. the point spread function number estimation method on arbitrary shape curve sword side according to claim 5, which is characterized in that described
Extended area meets following formula：
Wherein, σ is the radius of extended area, △ liFor the distance of ith sample point to fitting a straight line, i is the positive integer more than 0,
N is the number of sampled point.
7. the point spread function number estimation method on arbitrary shape curve sword side according to claim 6, which is characterized in that described
Include into row interpolation and re-sampling operations to the edge-spread function sample after screening：
To the edge-spread function sample after screening into row interpolation, and the point that interpolation is obtained and the edge-spread function
Sample is ranked up again according to the value of projected position and carries out piecewise fitting and re-sampling operations together.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810087171.0A CN108389186A (en) | 2018-01-30 | 2018-01-30 | The point spread function number estimation method on arbitrary shape curve sword side |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810087171.0A CN108389186A (en) | 2018-01-30 | 2018-01-30 | The point spread function number estimation method on arbitrary shape curve sword side |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108389186A true CN108389186A (en) | 2018-08-10 |
Family
ID=63074093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810087171.0A Pending CN108389186A (en) | 2018-01-30 | 2018-01-30 | The point spread function number estimation method on arbitrary shape curve sword side |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108389186A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007268647A (en) * | 2006-03-31 | 2007-10-18 | Mitsubishi Materials Kobe Tools Corp | End mill |
| CN102692273A (en) * | 2012-05-31 | 2012-09-26 | 中国资源卫星应用中心 | Method of on-track detection for MTF (modulation transfer function) of interference hyperspectral imager |
| CN102779333A (en) * | 2012-07-10 | 2012-11-14 | 武汉大学 | Optical image restoration method based on Kalman filter |
| CN104933713A (en) * | 2015-06-12 | 2015-09-23 | 杭州电子科技大学 | Image MTF (Modulation Transfer Function) estimation method using edge analysis |
| CN105069313A (en) * | 2015-08-24 | 2015-11-18 | 北京理工大学 | Phase nonlinear resampling and knife-edge fitting based in-orbit MTF (Modulation Transfer Function) estimation method |
| US9295431B2 (en) * | 2011-10-28 | 2016-03-29 | New York University | Constructing a 3-dimensional image from a 2-dimensional image and compressing a 3-dimensional image to a 2-dimensional image |
| CN205538161U (en) * | 2016-02-03 | 2016-08-31 | 上海仪万光电科技有限公司 | Optical lens's modulation transfer function's device is measured to unlimited conjugation light path |
-
2018
- 2018-01-30 CN CN201810087171.0A patent/CN108389186A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007268647A (en) * | 2006-03-31 | 2007-10-18 | Mitsubishi Materials Kobe Tools Corp | End mill |
| US9295431B2 (en) * | 2011-10-28 | 2016-03-29 | New York University | Constructing a 3-dimensional image from a 2-dimensional image and compressing a 3-dimensional image to a 2-dimensional image |
| CN102692273A (en) * | 2012-05-31 | 2012-09-26 | 中国资源卫星应用中心 | Method of on-track detection for MTF (modulation transfer function) of interference hyperspectral imager |
| CN102779333A (en) * | 2012-07-10 | 2012-11-14 | 武汉大学 | Optical image restoration method based on Kalman filter |
| CN104933713A (en) * | 2015-06-12 | 2015-09-23 | 杭州电子科技大学 | Image MTF (Modulation Transfer Function) estimation method using edge analysis |
| CN105069313A (en) * | 2015-08-24 | 2015-11-18 | 北京理工大学 | Phase nonlinear resampling and knife-edge fitting based in-orbit MTF (Modulation Transfer Function) estimation method |
| CN205538161U (en) * | 2016-02-03 | 2016-08-31 | 上海仪万光电科技有限公司 | Optical lens's modulation transfer function's device is measured to unlimited conjugation light path |
Non-Patent Citations (2)
| Title |
|---|
| 曲梦雅等: "一种优化的基于倾斜刃边的点扩散函数重建方法", 《测绘科学技术学报》 * |
| 范冲等: "点扩散函数的改进倾斜刃边重建的高精度估计", 《测绘学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2137519B1 (en) | Method and apparatus for detecting defects in optical components | |
| US8180154B2 (en) | Method and apparatus for region-based segmentation image processing using region mapping | |
| CN111083365B (en) | Method and device for rapidly detecting optimal focal plane position | |
| CN111968144B (en) | Image edge point acquisition method and device | |
| EP3534334B1 (en) | Method for identification of characteristic points of a calibration pattern within a set of candidate points derived from an image of the calibration pattern | |
| JPH0847481A (en) | Optical measuring method | |
| CN108564085A (en) | A kind of method of automatic reading pointer type instrument reading | |
| CN113160084B (en) | Denoising method and device for quantum dot fluorescence image on surface of porous silicon biosensor | |
| CN112991287B (en) | Automatic indentation measurement method based on full convolution neural network | |
| CN109060290A (en) | The method that wind-tunnel density field is measured based on video and Sub-pixel Technique | |
| CN108335310B (en) | Portable grain shape and granularity detection method and system | |
| CN109448059A (en) | A kind of quick X angular-point sub-pixel detection method | |
| CN114005108A (en) | Pointer instrument degree identification method based on coordinate transformation | |
| CN106017328A (en) | Method and device for measuring various types of line widths | |
| CN108320799A (en) | Image analysis and recognition method for lateral flow paper strip disease diagnosis | |
| CN106022354B (en) | Image MTF measurement methods based on SVM | |
| CN116385440B (en) | Visual detection method for arc-shaped blade | |
| CN113375555A (en) | Power line clamp measuring method and system based on mobile phone image | |
| US8934735B2 (en) | Oriented, spatio-spectral illumination constraints for use in an image progress | |
| CN108389186A (en) | The point spread function number estimation method on arbitrary shape curve sword side | |
| CN1667351A (en) | Method for automatically extracting image feature points of workpiece with rough grain under the strong reflection background | |
| CN116563298A (en) | Cross line center sub-pixel detection method based on Gaussian fitting | |
| US20220383506A1 (en) | Line detection method | |
| CN107845077A (en) | A kind of method and storage medium of correction DMD micro mirrors member imaging Asymmetrical deformation | |
| CN112819823A (en) | Furniture board-oriented circular hole detection method, system and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180810 |
|
| RJ01 | Rejection of invention patent application after publication |