CN104537646A - Multi-angle automatic MTF estimation method of remote sensing image - Google Patents

Abstract

The invention provides a multi-angle automatic MTF estimation method of a remote sensing image. The method comprises the two processes of selecting image blocks and estimating an MTF based on the edge method. The process of selecting the image blocks comprises the steps of rotating an original image by multiple angles to obtain a rotated image, carrying out edge point extraction on the rotated image and the original image to obtain four edge point images, and selecting the image blocks for estimating the MTF with the four edge point images as the reference. The process of estimating the MTF based on the edge method comprises the steps of carrying out edge detection on the obtained image blocks to obtain the positions of edge points, namely, the positions of sub-pixel points of the edge points, fitting the edges of the image blocks through the least square method, carrying out interpolation and average edge spread function extraction on the fit edges of the images blocks, carrying out simple differencing on an average edge spread function to obtain a line spread function, fitting the line spread function through Gaussian distribution, carrying out discrete Fourier transformation on the fit line spread function, and carrying out model extraction on the result to obtain a final MTF series.

Description

The automatic MTF method of estimation of multi-angle of remote sensing images

Technical field

The invention belongs to remote sensing application field, the automatic MTF method of estimation of the multi-angle in particular to a kind of remote sensing images.

Background technology

In remote sensing application field, modulation transfer function (MTF) is the important comprehensive evaluation index of remote sensing optical imaging system.The height of imaging system MTF directly has influence on the quality of image quality: the details such as MTF is lower, the Edge texture of the remote sensing images obtained will be fuzzyyer.For satellite in orbit remote sensor, due to the impact by the severe factor of the repeatedly pose adjustment in satellite launch, track keep-process, cosmic space radiation, day and night temperature impact etc., the imaging performance of remote sensor can decline gradually, and Remote Sensing Image Quality can be deteriorated gradually.Detect the MTF of satellite in orbit remote sensor, the application for satellite in orbit remote sensor is extremely important.

Before remote sensing satellite heaven, in the ground experiment room stage, adopting its MTF of specialized equipment measurements and calculations, is that current precision is the highest, reliability the best way.But this method can only be carried out under laboratory conditions, cannot calculate the MTF of satellite remote sensor in orbit.

Utilize the ground target laid and the image-forming information of the surface mark thing chosen on remote sensing images to calculate MTF, for the monitoring in-orbit of remote sensing satellite, comparatively for convenience of feasible.Wherein, representationally two class ways are mainly contained: the land-based target calibration method that the use that a Lei Shi U.S., France etc. adopt manually is laid.Another kind of is natural feature on a map method, be with the U.S. be representative directly utilize surface mark thing, as the ground target selecting bridge, airport etc. large-scale, from containing these targets remote sensing images, directly calculating MTF method.

Mainly have employed in prior art " sample pairing comparision " and " terrestrial reference mensuration ".Sample pairing comparision, is with the sample image of known MTF and satellite remote sensing images compares and interpretation, thus determines the method for the MTF of remote sensing satellite.Terrestrial reference mensuration is the same with land-based target calibration method mentioned above, the mainly method of foreign.

Summary of the invention

The present invention is directed to the situation manually choosing image block in prior art, propose a kind of automatic MTF method of estimation of multi-angle of remote sensing images.

Above-mentioned purpose of the present invention is realized by the technical characteristic of independent claims, and dependent claims develops the technical characteristic of independent claims with alternative or favourable mode.

For reaching above-mentioned purpose, the technical solution adopted in the present invention is as follows:

The automatic MTF method of estimation of multi-angle of remote sensing images, comprises image block and chooses and estimate MTF two parts based on recognition status, wherein:

Aforementioned image block is chosen and is comprised the following steps:

1-1) original image is carried out multi-angle rotary, obtain postrotational image;

1-2) respectively marginal point extraction is carried out to postrotational image and original image, obtain 4 breadths edge dot image;

1-3) utilizing 4 width edge images for referencial use respectively, choosing the image block for estimating MTF;

Aforementioned based on recognition status estimation MTF comprise the following steps:

2-1) for the image block obtained, make rim detection, obtain the position of marginal point, namely obtain the sub-pix point position of marginal point;

2-2) with the edge in least square fitting image block;

2-3) interpolation carried out to the edge in the image block of matching and extract average edge spread function;

2-4) simple differencing is done to average spread function, obtain line spread function, and by Gaussian distribution, matching is carried out to line spread function;

2-5) discrete Fourier transformation is carried out to the line spread function after matching, and delivery is carried out to result, obtain final MTF sequence.

In further embodiment, abovementioned steps 1-1) in original image is carried out multi-angle rotary, its anglec of rotation is respectively: 0 degree, 45 degree, 90 degree and 135 degree.

In further embodiment, abovementioned steps 1-2) middle row marginal point extraction, choose the extraction that marginal point is carried out in Canny (Tuscany) conversion, obtain the marginal point image of entire image.

In further embodiment, abovementioned steps 1-3) in utilize the 4 width edge images image block chosen for estimating MTF for referencial use, its realization comprises:

31) image block of each angle is chosen the process that process is expanded as an image block width, for each point of edge image, if this point is marginal point, then what start image block chooses process, comprising:

311) height H setting image block is 5, width w self-adaptation, width is determined by the peak width of edge line width and edge line both sides, edge line width is 1, its initial value of the peak width of edge line left and right sides W1=W2=1, both sides relation is W=W1+W2+1, and initial image block width is W=3, and what namely comprise edge line and its both sides faces antermarginal each one-row pixels point;

312) increase width W 1, the W2 of edge line both sides, each step-length increased is 1, then again increases as long as the region of both sides is all smooth region width simultaneously, then meets W while increasing width and be less than or equal to 15;

32) for the image block chosen, be more than or equal to 9 and be less than or equal to 15 meeting width, when the gray average difference of edge line both sides is more than or equal to 66, this image block selected is as the image block calculated for MTF, otherwise give up, then return and turn 31), until the point in edge image travels through completely, then exit traversal.

In further embodiment, abovementioned steps 2-1) in rim detection, namely the position obtaining marginal point obtains the sub-pix point position of marginal point, and its realization comprises:

41) simple differencing computing is carried out to each row of image block, choose maximum of points or minimum point, the i.e. flex point of every a line after difference, as the pixel edge point of this line;

42) selected pixels level marginal point and the right and left totally 4 pixels calculate the position of sub-pix point, calculate and adopt the method for cubic polynomial curve to carry out matching to these 4 data, 0 position of matched curve is the sub-pixel edge point position of this row.

In further embodiment, abovementioned steps 2-2) in, comprise with least square fitting edge and fitting a straight line is carried out to sub-pix point, move on to all enforceable for all marginal points extracted in abovementioned steps on straight line, thus make the center of every a line be marginal point.

In further embodiment, abovementioned steps 2-3) in, described step 6) in interpolation is carried out to the edge in the image block of matching and extracts the process of average edge spread function, wherein:

Interpolation processing comprises carries out cubic spline interpolation to each line number strong point of image block, the interpolated resolution adopted is 0.05, namely between every two data points, 20 interpolation points are inserted, so make every a line of image become an approximately continuous line, its grey value profile is exactly the edge spread function of this row;

Extract average edge spread function to comprise and the edge spread function of all row added up and gets its mean value, obtain average edge spread function.

In further embodiment, abovementioned steps 2-4) in, do simple differencing to average spread function, obtain line spread function, and carry out matching by Gaussian distribution to line spread function, its specific implementation comprises:

In order to make the noise factor of the smooth region of both sides of edges, the impact of MTF result of calculation is reduced as far as possible, first suitable intercepting is carried out in the region of edge both sides, namely intercept each 100 data values of both sides of edges, utilize Gaussian distribution to carry out matching to it to the line spread function after intercepting.

As long as should be appreciated that aforementioned concepts and all combinations of extra design described in further detail below can be regarded as a part for subject matter of the present disclosure when such design is not conflicting.In addition, all combinations of theme required for protection are all regarded as a part for subject matter of the present disclosure.

The foregoing and other aspect of the present invention's instruction, embodiment and feature can be understood by reference to the accompanying drawings from the following description more all sidedly.Feature and/or the beneficial effect of other additional aspect of the present invention such as illustrative embodiments will be obvious in the following description, or by learning in the practice of the embodiment according to the present invention's instruction.

Accompanying drawing explanation

Accompanying drawing is not intended to draw in proportion.In the accompanying drawings, each identical or approximately uniform ingredient illustrated in each figure can represent with identical label.For clarity, in each figure, not each ingredient is all labeled.Now, the embodiment of various aspects of the present invention also will be described with reference to accompanying drawing by example, wherein:

Fig. 1-4,5-8 exemplarily illustrates remote sensing images and the marginal point image schematic diagram thereof of 4 angles, wherein Fig. 1, Fig. 2, Fig. 3, Fig. 4 carry out 0 degree, 45 degree, 90 degree and 135 degree to original image to rotate the remote sensing images obtained, and Fig. 5, Fig. 6, Fig. 7, Fig. 8 are the marginal point image that the extraction utilizing Canny operator to carry out marginal point to Fig. 1, Fig. 2, Fig. 3, Fig. 4 respectively obtains four angles respectively.

Fig. 9 is that image block chooses process flow diagram.

Figure 10-15 is image block edge line and fitting result schematic diagram, Figure 10 is image block and its corresponding gray-scale value schematic diagram, Figure 11 is a line gray-scale value schematic diagram in image block, Figure 12 is the differentiated result schematic diagram of a line gray-scale value in image block, Figure 13 is the intensity profile schematic diagram of single row of pixels, the difference profile figure of Figure 14 single row of pixels, Figure 15 are the result schematic diagram of rim detection.

Figure 16 is the schematic diagram of image block single file interpolation.

Figure 17-18 is average edge spread function (ESF) acquisition process schematic diagram, and Figure 17 is the ESF curve synoptic diagram of all row in image block (5 row), and Figure 18 is the average ESF curve synoptic diagram of band sampled point.

Figure 19-21 is that line spread function (LSF) obtains and fit procedure schematic diagram, Figure 19 calculates line spread function (LSF) curve synoptic diagram, Figure 20 is the LSF curve synoptic diagram after intercepting, and Figure 21 is that Gaussian distribution is to the LSF fitting result schematic diagram after intercepting.

Figure 22 is MTF sequence chart schematic diagram.

Figure 23 is the mtf value schematic diagram that all image blocks calculate.

Figure 24 is the exemplary process diagram of the multi-angle braking MTF method of estimation of the remote sensing images that the present invention proposes.

Embodiment

In order to more understand technology contents of the present invention, institute's accompanying drawings is coordinated to be described as follows especially exemplified by specific embodiment.

Each side with reference to the accompanying drawings to describe the present invention in the disclosure, shown in the drawings of the embodiment of many explanations.Embodiment of the present disclosure must not be intended to comprise all aspects of the present invention.Be to be understood that, multiple design presented hereinbefore and embodiment, and those designs described in more detail below and embodiment can in many ways in any one is implemented, this should be design disclosed in this invention and embodiment is not limited to any embodiment.In addition, aspects more disclosed by the invention can be used alone, or otherwisely anyly appropriately combinedly to use with disclosed by the invention.

In conjunction with the flow process shown in Figure 24, the automatic MTF method of estimation of multi-angle of the remote sensing images that the present invention proposes, it realizes comprising image block and chooses and estimate MTF two parts based on recognition status, wherein:

Aforementioned image block is chosen and is comprised the following steps:

1-1) original image is carried out multi-angle rotary, obtain postrotational image;

1-2) respectively marginal point extraction is carried out to postrotational image and original image, obtain 4 breadths edge dot image;

1-3) utilizing 4 width edge images for referencial use respectively, choosing the image block for estimating MTF;

Aforementioned based on recognition status estimation MTF comprise the following steps:

2-1) for the image block obtained, make rim detection, obtain the position of marginal point, namely obtain the sub-pix point position of marginal point;

2-2) with the edge in least square fitting image block;

2-3) interpolation carried out to the edge in the image block of matching and extract average edge spread function;

2-4) simple differencing is done to average spread function, obtain line spread function, and by Gaussian distribution, matching is carried out to line spread function;

2-5) discrete Fourier transformation is carried out to the line spread function after matching, and delivery is carried out to result, obtain final MTF sequence.

Below in conjunction with Fig. 1-Figure 23, and the signal of the multi-angle of the remote sensing images shown in Figure 24 automatic MTF method of estimation realization flow, illustrate the exemplary implementation of the method.

For remote sensing images to be assessed, first original image is carried out the rotation of 0 degree, 45 degree, 90 degree and 135 degree 3 angles, thus obtain the remote sensing images of other 3 angles, as shown in Figure 1, Figure 2, Figure 3, Figure 4.

The extraction utilizing Canny (Tuscany) operator to carry out marginal point to original image and other 3 images respectively obtains the marginal point image of four angles, as shown in Fig. 5, Fig. 6, Fig. 7, Fig. 8.

Next the edge image of each angle is utilized to choose the image block that this angle can be used for MTF calculating.For each width edge image, each pixel in detected image, if this pixel is marginal point, then what start image block chooses work, otherwise judges next point.Choosing of image block is not carried out on edge image, but on the remote sensing images of correspondence.

In this example, abovementioned steps 1-3) in utilize the 4 width edge images image block chosen for estimating MTF for referencial use, its realization comprises:

31) image block of each angle is chosen the process that process is expanded as an image block width, for each point of edge image, if this point is marginal point, then what start image block chooses process, comprising:

311) height H setting image block is 5, width w self-adaptation, width is determined by the peak width of edge line width and edge line both sides, edge line width is 1, its initial value of the peak width of edge line left and right sides W1=W2=1, both sides relation is W=W1+W2+1, and initial image block width is W=3, and what namely comprise edge line and its both sides faces antermarginal each one-row pixels point;

312) increase width W 1, the W2 of edge line both sides, each step-length increased is 1, then again increases as long as the region of both sides is all smooth region width simultaneously, then meets W while increasing width and be less than or equal to 15;

32) for the image block chosen, be more than or equal to 9 and be less than or equal to 15 meeting width, when the gray average difference of edge line both sides is more than or equal to 66, this image block selected is as the image block calculated for MTF, otherwise give up, then return and turn 31), until the point in edge image travels through completely, then exit traversal.

Shown in composition graphs 9, exemplarily give the flow process that image block is chosen, its realization comprises:

Step 1) first height H=5 of image block are set, the width of the edge line left and right sides is respectively W1=W2=1, and the width of image block is W=3;

Step 2) judge whether the region of edge line both sides is smooth region, if be all smooth region, W1 and W2 increased 1, W and increase 2 accordingly and turn 3), otherwise turn 4);

Step 3) judge whether W is less than or equal to 15, turn 2 if satisfy condition), otherwise turn 4);

Step 4) W1 and W2 all subtract that 1, W is corresponding subtracts 2, judge whether W is more than or equal to 9, if satisfy condition, this image block meets MTF design conditions, is usable image block, otherwise gives up this image block.

Through above image block selecting step, the image block that 4 directions can be used for MTF calculating will be out selected, will adopt the evaluation work carrying out MTF in the following method for each width image block.

There is a marginal point each provisional capital of image block, and these marginal points constitute the edge line of image block, and first step work is exactly the position obtaining marginal point.

In this example, abovementioned steps 2-1) in rim detection, namely the position obtaining marginal point obtains the sub-pix point position of marginal point, and its realization comprises:

41) simple differencing computing is carried out to each row of image block, choose maximum of points or minimum point, the i.e. flex point of every a line after difference, as the pixel edge point of this line;

42) selected pixels level marginal point and the right and left totally 4 pixels calculate the position of sub-pix point, calculate and adopt the method for cubic polynomial curve to carry out matching to these 4 data, 0 position of matched curve is the sub-pixel edge point position of this row.

As shown in fig. 10-15 for obtaining the process of data line marginal point, Figure 10 is an image block and its corresponding gray-scale value, and the height and width of this image block are respectively 5 and 15.If Figure 11 is the determination for every a line edge point position of the data of a line gray-scale value and the distribution situation (by 0 ~ 255 span) of gray scale, as preferred scheme in this example, its realization comprises:

Step 1) simple differencing is done to the every a line gray-scale value in data block, namely x [n]-x [n-1] is calculated, as a line gray-scale value that Figure 11 is in image block, Figure 12 is differentiated result, and get max (x [n]-x [n-1)], the position of maximum slope point just can be detected, as Figure 13,14 is respectively intensity profile and the difference profile figure of single row of pixels;

In theory, above-mentioned maximum slope point, namely flex point position is the position of marginal point, but this result is Pixel-level position, can not meet the demands.In order to determine the accurate location of the sub-pixel of this row marginal point further, carry out step 2).

Step 2) adopt the value of 4 points near flex point, the data point of the overstriking of 4 shown in Figure 11, do cubic polynomial curve matching, fitting formula is as follows:

a(x)＝a ₁+a ₂x ²+a ₃x+a ₄

Because second derivative crosses 0 point, that is:

a″(x)＝6a ₁x+2a ₂＝0

Solve:

x＝-(2a ₂)/(6a ₁)

X is the marginal point sub-pixel position of this required row, each provisional capital in Figure 10 is carried out to the process of as above step, just can obtain the position of the sub-pixel of the marginal point of every a line.Be the result of rim detection shown in Figure 15, the position being the marginal point sub-pixel detected by every one-row pixels of red small circle mark, in Figure 15 tables of data bracket, be the sub-pixel edge point position of often going.

After obtaining the position of marginal point, further work is exactly the matching carrying out edge.A basic assumption of blade method is that the edge of blade is positioned on straight line.All can have an impact to the final MTF result calculated relative to the departing from of any point point of this straight line.So, consider based on this point, by all enforceable on straight line for all marginal points extracted, therefore, edge testing result must carry out linear fit above.The method that matching adopts is least square method.

In this example, abovementioned steps 2-2) in, comprise with least square fitting edge and fitting a straight line is carried out to sub-pix point, move on to all enforceable for all marginal points extracted in abovementioned steps on straight line, thus make the center of every a line be marginal point.

Suppose straight-line equation:

y＝ax+b

Wherein

$a=\frac{m\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i{y}_i\right)-\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i\right)\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{y}_i\right)}{m\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i^2\right)-{\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i\right)}^2}$

$b=\frac{\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i^2\right)\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{y}_i\right)-\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i{y}_i\right)\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i\right)}{m\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i^2\right)-{\left(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i\right)}^2}$

Wherein m is marginal point number, x _ifor the height of image block, y _ifor sub-pixel edge point position.At this moment adjust the relative position of every one-row pixels, the standard of adjustment is that the position of each row sub-pixel edge point that the first step is detected all can accurately be dropped on fitted straight lines of edges, and each row of data is expanded to both sides centered by marginal point.The object done like this is, conveniently asks the average of each row edge spread function, reduces the impact of error.

After adjusting each row data, next step is exactly carry out cubic spline interpolation to each line number strong point of image.The interpolated resolution adopted is 0.05, namely between every two data points, inserts 20 interpolation points.Like this, every a line of image just becomes an approximately continuous line, and its intensity profile is exactly the edge spread function of this row.Figure 16 shows that the result of single file data being carried out to cubic spline interpolation.Because interpolation is carried out in each provisional capital of image, therefore, a nearly continuous print ESF curve can be drawn in its each provisional capital, is added up by the ESF of all row and gets its mean value, just can obtain average edge spread function ESF.Figure 17 shows that the ESF of all row in image block (5 row), Figure 18 is the average ESF curve of band sampled point.

In this example, abovementioned steps 2-3) in interpolation is carried out to the edge in the image block of matching and extracts the process of average edge spread function, wherein:

Interpolation processing comprises carries out cubic spline interpolation to each line number strong point of image block, the interpolated resolution adopted is 0.05, namely between every two data points, 20 interpolation points are inserted, so make every a line of image become an approximately continuous line, its grey value profile is exactly the edge spread function of this row;

Extract average edge spread function to comprise and the edge spread function of all row added up and gets its mean value, obtain average edge spread function.

Once obtain average edge spread function (ESF), next step calculates line spread function (LSF) exactly, and only need do simple differencing to average ESF, difference formula is as follows, and acquired results is as Figure 19:

LSF(n)＝ESF(n)-ESF(n-1)

The image block calculating MTF requires that blade both sides do not have too many noise or the interference of other target objects, and namely the intensity profile of the gray scale smooth region of knife edge both sides is relatively even.Reduce to make the dosage that affects of the noise factor of the smooth region in knife edge both sides on MTF result of calculation, here suitable intercepting is carried out in the region of edge both sides, namely intercept both sides of edges each 100 data values, only leave the suitable region of the crest of edge and both sides.Figure 20 is the LSF curve after intercepting.Then in, Gaussian distribution carries out matching to the LSF after intercepting, and the fitting result obtained as shown in figure 21.

In this example, abovementioned steps 2-4) in, do simple differencing to average spread function, obtain line spread function, and carry out matching by Gaussian distribution to line spread function, its specific implementation comprises:

In order to make the noise factor of the smooth region of both sides of edges, the impact of MTF result of calculation is reduced as far as possible, first suitable intercepting is carried out in the region of edge both sides, namely intercept each 100 data values of both sides of edges, utilize Gaussian distribution to carry out matching to it to the line spread function after intercepting.

After obtaining the line spread function LSF after matching, discrete Fourier transform (DFT) is carried out to it, get the mould of the rear each component of conversion, and for benchmark, do normalized with DC component, i.e. first mtf value after conversion, just obtain desired MTF sequence.Formula (1) expression carries out Fourier transform to line spread function, and formula (2) represents the DC component to the result of Fourier transform, and namely real part does normalized.What formula (3) represented is give Nyquist frequency computation part formula, and Nyquist frequency is the half of cutoff frequency.

MTF(n)＝|DFT(LSF(n))| (1)

norm_MTF(n)＝MTF(n)/MTF(1) (2)

Nyquist_frequency＝(whole_data_size×resolution)/2+1

＝(Number_of_trimmed_pixel)/2+1 (3)

Wherein DFT () represents discrete Fourier transform (DFT), norm_MTF (n) is normalization mtf value, Nyquist_frequency is Nyquist Frequency point, whole_data_size is LSF sub-pixel width after the intercepting comprising all interpolation points, resolution is interpolated resolution, and Number_of_trimmed_pixel is the LSF Pixel-level width after intercepting.Figure 22 is depicted as the final MTF calculated, and is 0.5301 in the mtf value result of calculation at Nyquist frequency place.

Figure 23 is the mtf value that all image blocks (35 image blocks) calculate, and all utilizes above-mentioned method to calculate MTF for each image block.

Although the present invention with preferred embodiment disclose as above, so itself and be not used to limit the present invention.Persond having ordinary knowledge in the technical field of the present invention, without departing from the spirit and scope of the present invention, when being used for a variety of modifications and variations.Therefore, protection scope of the present invention is when being as the criterion depending on those as defined in claim.

Claims (8)

1. the automatic MTF method of estimation of the multi-angle of remote sensing images, is characterized in that, the automatic MTF method of estimation of this multi-angle comprises image block and chooses and estimate MTF two parts based on recognition status, wherein:

Aforementioned image block is chosen and is comprised the following steps:

1-1) original image is carried out multi-angle rotary, obtain postrotational image;

1-2) respectively marginal point extraction is carried out to postrotational image and original image, obtain 4 breadths edge dot image;

1-3) utilizing 4 width edge images for referencial use respectively, choosing the image block for estimating MTF;

Aforementioned based on recognition status estimation MTF comprise the following steps:

2-1) for the image block obtained, make rim detection, obtain the position of marginal point, namely obtain the sub-pix point position of marginal point;

2-2) with the edge in least square fitting image block;

2-3) interpolation carried out to the edge in the image block of matching and extract average edge spread function;

2-4) simple differencing is done to average spread function, obtain line spread function, and by Gaussian distribution, matching is carried out to line spread function;

2-5) discrete Fourier transformation is carried out to the line spread function after matching, and delivery is carried out to result, obtain final MTF sequence.

2. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 1, is characterized in that, abovementioned steps 1-1) in original image is carried out multi-angle rotary, its anglec of rotation is respectively: 0 degree, 45 degree, 90 degree and 135 degree.

3. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 1, is characterized in that, abovementioned steps 1-2) middle row marginal point extraction, choose the extraction that marginal point is carried out in Canny conversion, obtain the marginal point image of entire image.

4. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 1, is characterized in that, abovementioned steps 1-3) in utilize the 4 width edge images image block chosen for estimating MTF for referencial use, its realization comprises:

31) image block of each angle is chosen the process that process is expanded as an image block width, for each point of edge image, if this point is marginal point, then what start image block chooses process, comprising:

311) height H setting image block is 5, width W self-adaptation, width is determined by the peak width of edge line width and edge line both sides, edge line width is 1, its initial value of the peak width of edge line left and right sides W1=W2=1, both sides relation is W=W1+W2+1, and initial image block width is W=3, and what namely comprise edge line and its both sides faces antermarginal each one-row pixels point;

312) increase width W 1, the W2 of edge line both sides, each step-length increased is 1, then again increases as long as the region of both sides is all smooth region width simultaneously, then meets W while increasing width and be less than or equal to 15;

32) for the image block chosen, be more than or equal to 9 and be less than or equal to 15 meeting width, when the gray average difference of edge line both sides is more than or equal to 66, this image block selected is as the image block calculated for MTF, otherwise give up, then return and turn 31), until the point in edge image travels through completely, then exit traversal.

5. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 1, is characterized in that, abovementioned steps 2-1) in rim detection, obtain the sub-pix point position that namely position of marginal point obtains marginal point, its realization comprises:

41) simple differencing computing is carried out to each row of image block, choose maximum of points or minimum point, the i.e. flex point of every a line after difference, as the pixel edge point of this line;

42) selected pixels level marginal point and the right and left totally 4 pixels calculate the position of sub-pix point, calculate and adopt the method for cubic polynomial curve to carry out matching to these 4 data, 0 position of matched curve is the sub-pixel edge point position of this row.

6. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 5, it is characterized in that, abovementioned steps 2-2) in, comprise with least square fitting edge and fitting a straight line is carried out to sub-pix point, move on to all enforceable for all marginal points extracted in abovementioned steps on straight line, thus make the center of every a line be marginal point.

7. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 6, it is characterized in that, abovementioned steps 2-3) in, described step 6) in interpolation is carried out to the edge in the image block of matching and extracts the process of average edge spread function, wherein:

Interpolation processing comprises carries out cubic spline interpolation to each line number strong point of image block, the interpolated resolution adopted is 0.05, namely between every two data points, 20 interpolation points are inserted, so make every a line of image become an approximately continuous line, its grey value profile is exactly the edge spread function of this row;

Extract average edge spread function to comprise and the edge spread function of all row added up and gets its mean value, obtain average edge spread function.

8. the automatic MTF method of estimation of the multi-angle of remote sensing images according to claim 7, is characterized in that, abovementioned steps 2-4) in, simple differencing is done to average spread function, obtain line spread function, and carry out matching by Gaussian distribution to line spread function, its specific implementation comprises:

In order to make the noise factor of the smooth region of both sides of edges, the impact of MTF result of calculation is reduced as far as possible, first suitable intercepting is carried out in the region of edge both sides, namely intercept each 100 data values of both sides of edges, utilize Gaussian distribution to carry out matching to it to the line spread function after intercepting.