CN101980293B - Method for detecting MTF of hyperspectral remote sensing system based on edge image - Google Patents

Abstract

The invention relates to a method for detecting a modulation transfer function (MTF) of a hyperspectral remote sensing system based on an edge image. The method takes the influence of an image noise into consideration and comprises the following steps of: selecting edge areas in the horizontal direction and the vertical direction by calculating the modulation degrees of the edges of each wave band of a hyperspectral remote sensing image; performing differential operation on the edge areas to obtain linear spread functions; performing Gaussian function least squares fitting on the linear spread functions to obtain standard deviations; figuring the linear spread functions in the vertical direction and the horizontal direction out by using a median of the standard deviations of the linear spread functions in the vertical direction and the horizontal direction; performing discrete Fourier transform to obtain MTF curves in the two directions; and establishing a two-dimensional MTF matrix of the system by multiplying. The method can select suitable edge areas in the situation of a plurality of wave bands of the hyperspectral remote sensing image, and realize the extraction of the MTF of the system when the signal-to-noise ratio of the image is not high.

Description

A kind of hyperspectral remote sensing system MTF detection method based on the sword edge image

Technical field

The present invention relates to a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image, belong to hyperspectral remote sensing system evaluation of imaging quality field, the MTF that is particularly suitable for hyperspectral remote sensing system detects.

Background technology

Modulation transfer function (MTF, Modulation Transfer Function) is the key factor of estimating the remote sensing system image quality.In the lab investigation stage, adopt specialized equipment to measure its MTF, be that present precision is the highest, reliability the best way.But this method can't be calculated the MTF of the remote sensor that moving.For the remote sensor that is in operation, comprehensive research situation both at home and abroad, obtaining MTF from remote sensing images information is comparatively general way, i.e. the ground target that utilization is manually laid or the image-forming information of surface mark thing on remote sensing images of choosing nature calculate MTF.At present, the method for testing of remote sensing images MTF mainly contains point source method, impulse method, blade method and sample pairing comparision etc., but relevant work is mostly to launch around common panchromatic remote sensing.

Point source method: after the Luminance Distribution function PSF (x, y) of acquisition point light source picture, it is carried out two-dimensional Fourier transform get final product to get MTF.Therefore, if add the some pulse input (such as artificial reflective mirror group of laying) of a pixel size to imaging system, so, it is exactly PSF that its output pulse diagram looks like.PSF is got final product to get MTF do two-dimensional Fourier transform.

Impulse method: impulse method is take linear ground object (as the bridge in image) as observed object, and this linear ground object is positioned on broad background uniformly, and larger with background reflectance.Its input pulse is the square pulse that width is equivalent to the target developed width, and its output pulse is the matched curve of gray-scale value on image; Input/output pulse all passes through Fourier transform, and being divided by also, normalization obtains MTF.

The blade method: the blade method is the border that selection has two adjacent even bright secretly things of certain contrast on image, by measuring imaging system, the expansion situation on this border is determined the response of system on various spatial frequencys, thereby obtain the MTF curve of this imaging system.Specific algorithm is: at first find " sword limit ", the border that namely has two adjacent even bright secretly things of certain contrast, the edge spread function of computed image on perpendicular to the direction on sword limit, obtain a discrete sword limit, and with continuous edge functions of method match such as cubic spline interpolations, and filtering is done at the edge of match processed.Then to the edge differentiate after level and smooth, obtain line spread function.At last line spread function is carried out Fourier transform and just can obtain modulation transfer function.

The sample pairing comparision: the sample pairing comparision is with the sample image of known MTF and satellite remote sensing images compares and interpretation, thereby determines the method for the MTF of remote sensing satellite.The method requires sample areas is chosen in the area that the earth's surface texture has higher-spatial frequencies.In addition, the acquisition time of the aviation of sample areas or space flight data source requires synchronize with the satellite image data of surveying or accurate synchronous, and need to employing aviation is multispectral that sample area is carried out data acquisition.

Due in actual remote sensing images, sword limit shape atural object is more common and be easy to obtain than point-like and linear ground object, and affected by noise less, thus blade method among above several method use in remote sensing system MTF detects the most extensive.But in the situation that the hyperspectral remote sensing system signal to noise ratio (S/N ratio) is lower, tradition blade method data handling procedure is affected by noise serious, often cause to obtain correct and stable testing result, compare panchromatic remote sensing and have the numerous characteristics of wave band due to hyperspectral remote sensing system in addition, select which type of sword edge regions could satisfy preferably the testing requirement of each wave band, this problem also can't be well solved in the conventional method, can't satisfy the requirement that hyperspectral remote sensing system MTF is detected.

Summary of the invention

The technical matters that the present invention solves is: numerous for the hyperspectral remote sensing system wave band, blade method MTF surveyed area is difficult to select, the high-spectrum remote sensing signal to noise ratio (S/N ratio) is generally lower, the MTF testing process is problem affected by noise easily, provides a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image.

Technical solution of the present invention is: a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image, calculating by counter blade limit degree of modulation, realized the selection of counter blade edge regions, by line spread function is carried out Gauss curve fitting, realized the MTF of insensitive for noise is extracted.Concrete steps are as follows:

(1) read high spectrum image;

(2) select to be used for horizontal sword edge regions and the vertical sword edge regions that MTF detects from the image that step (1) reads;

(3) signal to noise ratio (S/N ratio) of calculation procedure (1) institute each wave band of reading images;

(4) the selected horizontal sword edge regions of step (2) and vertical sword edge regions are carried out the medium filtering pre-service;

(5) the selected horizontal sword edge regions of calculation procedure (2) and vertically the sword limit degree of modulation of each wave band of sword edge regions;

(6) if the sword limit degree of modulation that calculates in step (5) in the value of each wave band all greater than the threshold value of setting, the horizontal sword edge regions that step (4) is obtained and vertically each row of sword edge regions vertical direction and every delegation do Difference Calculation and obtain line spread function, reselect the sword edge regions otherwise return to step (2);

(7) obtain standard deviation to carrying out respectively the Gaussian function least square fitting by the required line spread function that goes out of each row and every delegation in step (6);

(8) each row and all standard deviations of every delegation of trying to achieve in step (7) are got respectively intermediate value as the final value of remote sensor vertical direction and horizontal direction Gaussian line spread function standard deviation;

(9) line spread function that obtains in step (8) is sampled, obtain the MTF on image vertical direction and horizontal direction of each wave band of hyperspectral remote sensing system through discrete Fourier transform (DFT);

(10) with vertical direction MTF as column vector, horizontal direction MTF is as row vector, both multiply each other and draw the system MTF matrix, that is:

MTF(u，v)＝MTF(u)×MTF(v)

Wherein, in step (3), the method for calculating signal to noise ratio (S/N ratio) is as follows:

Choose zone relatively smooth in whole image, should be divided into 4 * 4,5 * 5 by the every band image in zone ... etc. sub-block, obtain respectively average and the standard deviation of each piece:

$m_i=\underset{j}{\mathrm{\Σ}}{x}_{\mathrm{ij}}/N,S_i=\sqrt{\underset{j}{\mathrm{\Σ}}{(x_{\mathrm{ij}}-m_i)}^2}/N$

X wherein _ijBe the gray-scale value of each pixel in sub-block, N is the number of pixels in block, m _iBe sub-block average, S _iBe the sub-block standard deviation.

It is generally acknowledged the corresponding zone uniformly of most sub-block in image, so, selected suitable grade point, to be divided into equally spaced zone between each sub-block standard deviation maximal value and minimum value, statistics is found out the standard deviation LSD of frequency of occurrences maximum at the poor grey level histogram of these regional Plays _m, as the estimation of noise, with the average gray and the LSD that choose image _mRatio represent signal to noise ratio snr.

Wherein, in step (4), the medium filtering template can be selected 3 * 3 or 5 * 5 sizes.

Wherein, in step (5), the formula of calculating sword limit degree of modulation is:

$M_{\mathrm{SNR}}=\frac{L_a-L_b-\frac{2\·L_v}{\mathrm{SNR}}}{L_a+L_b}$

L wherein _aBe the bright side average gray value in sword limit, L _bBe the dark side average gray value in sword limit, L _vBe sword edge image average gray value, SNR is signal to noise ratio (S/N ratio).

Wherein, the sword limit degree of modulation selection criterion in step (6) is:

$M_{\mathrm{SNR}}=\frac{L_a-L_b-\frac{2\·L_v}{\mathrm{SNR}}}{L_a+L_b}>M_{\mathrm{th}}$

M wherein _thBe sword limit threshold modulation.

Wherein, the method with Gaussian function fitting in step (7) is found the solution line spread function, and selected Gaussian function expression formula is:

$f\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}}{e}^{-\frac{x^2}{{2\mathrm{\σ}}^2}}$

Wherein σ is the Gaussian function standard deviation.

The present invention's advantage compared with prior art is:

(1) the present invention provides a kind of system of selection of high-spectrum remote sensing MTF surveyed area, in conjunction with each wave band signal to noise ratio (S/N ratio), judges by calculating sword limit degree of modulation whether selected areas is suitable for the detection of MTF.

(2) the present invention has provided in a kind of MTF testing process the method with the Gaussian function fitting line spread function, has reduced in the data handling procedure susceptibility of noise is applicable to the lower situation of high spectrum image signal to noise ratio (S/N ratio).

Description of drawings

Fig. 1 is implementation method process flow diagram of the present invention.

Embodiment

As shown in Figure 1, specific implementation method of the present invention is as follows:

1, read high spectrum image;

2, select to be used for horizontal sword edge regions and the vertical sword edge regions that MTF detects from the image that step 1 reads, require selected areas that adjacent homogeneous bulky atural object is arranged, there is certain contrast in brightness, and the border is the straight line of certain-length;

3, the signal to noise ratio (S/N ratio) of 1 each wave band of reading images of calculation procedure, computing method are as follows:

Choose zone relatively smooth in whole image, should be divided into 4 * 4,5 * 5 by the every band image in zone ... etc. sub-block, obtain respectively average and the standard deviation of each piece:

$m_i=\underset{j}{\mathrm{\Σ}}{x}_{\mathrm{ij}}/N,S_i=\sqrt{\underset{j}{\mathrm{\Σ}}{(x_{\mathrm{ij}}-m_i)}^2}/N---\left(1\right)$

X wherein _ijBe the gray-scale value of each pixel in sub-block, N is the number of pixels in block, m _iBe sub-block average, S _iBe the sub-block standard deviation.

It is generally acknowledged the corresponding zone uniformly of most sub-block in image, so, selected suitable grade point, to be divided into equally spaced zone between each sub-block standard deviation maximal value and minimum value, statistics is found out the standard deviation LSD of frequency of occurrences maximum at the poor grey level histogram of these regional Plays _m, as the estimation of noise, with the average gray and the LSD that choose image _mRatio represent signal to noise ratio snr;

4, the selected horizontal sword edge regions of step 2 and vertical sword edge regions are carried out the medium filtering pre-service, the medium filtering template can be selected 3 * 3 or 5 * 5 sizes;

5, the selected horizontal sword edge regions of calculation procedure 2 and vertically the sword edge regions is in the sword limit of each wave band degree of modulation, establishing certain bright side average gray value in wave band sword limit is L _a, dark side average gray value is L _b, the average gray value of view picture figure is L _v, signal to noise ratio (S/N ratio) is SNR, its computing formula is:

$M_{\mathrm{SNR}}=\frac{L_a-L_b-\frac{2\·L_v}{\mathrm{SNR}}}{L_a+L_b}---\left(2\right)$

If the sword limit degree of modulation that calculates in 6 steps 5 in the value of each wave band all greater than the threshold value M that sets _th(M _thBe empirical value, about general desirable 0.1-0.2) the line spread function that Difference Calculation obtains correspondence direction is done by each row and every delegation of horizontal sword edge regions and vertical sword edge regions vertical direction, reselect corresponding sword edge regions otherwise return to step 2;

7, to obtain series of standards poor to carrying out respectively Gaussian function fitting by the required line spread function that goes out of each row and every delegation in step 6, and selected Gaussian function expression formula is:

$f\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}}{e}^{-\frac{x^2}{2{\mathrm{\σ}}^2}}---\left(3\right)$

Wherein σ is the Gaussian function standard deviation.

8, all standard deviations of each row of trying to achieve in step 7 and every delegation are got respectively intermediate value as the value of σ in remote sensor vertical direction and horizontal direction line spread function expression formula (3);

9, by required resolution, the continuous lines spread function curve that obtains in step 8 is sampled, the process discrete Fourier transform (DFT) is obtained the MTF on each band image vertical direction of hyperspectral remote sensing system and horizontal direction;

10, with vertical direction MTF column vector and the capable multiplication of vectors of horizontal direction MTF as the system MTF matrix, that is:

MTF(u，v)＝MTF(u)×MTF(v)。

Claims (4)

1. hyperspectral remote sensing system MTF detection method based on the sword edge image comprises the following steps:

(1) read high spectrum image;

(2) select to be used for horizontal sword edge regions and the vertical sword edge regions that MTF detects from the image that step (1) reads;

(3) signal to noise ratio (S/N ratio) of calculation procedure (1) institute each wave band of reading images;

(4) the selected horizontal sword edge regions of step (2) and vertical sword edge regions are carried out the medium filtering pre-service;

(5) the selected horizontal sword edge regions of calculation procedure (2) and vertically the sword limit degree of modulation of each wave band of sword edge regions;

(6) if the sword limit degree of modulation that calculates in step (5) in the value of each wave band all greater than the threshold value of setting, the horizontal sword edge regions that step (4) is obtained and vertically each row of sword edge regions vertical direction and every delegation do Difference Calculation and obtain line spread function, reselect the sword edge regions otherwise return to step (2);

(7) obtain standard deviation to carrying out respectively the Gaussian function least square fitting by the required line spread function that goes out of each row and every delegation in step (6);

(8) each row and all standard deviations of every delegation of trying to achieve in step (7) are got respectively intermediate value as the final value of remote sensor vertical direction and horizontal direction Gaussian line spread function standard deviation;

(9) line spread function that obtains in step (8) is sampled, obtain the MTF on image vertical direction and horizontal direction of each wave band of hyperspectral remote sensing system through discrete Fourier transform (DFT);

(10) with vertical direction MTF as column vector MTF (u), horizontal direction MTF as the row vector M TF (v), both multiply each other and draw system MTF matrix M TF (u, v), that is:

MTF(u,v)=MTF(u)×MTF(v)。

2. a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image according to claim 1 is characterized in that: the computing formula of the sword limit degree of modulation described in step (5) is:

$M_{\mathrm{SNR}}=\frac{L_a-L_b-\frac{2\·L_v}{\mathrm{SNR}}}{L_a+L_b}$

L wherein _aBe the bright side average gray value in sword limit, L _bBe the dark side average gray value in sword limit, L _vBe sword edge image average gray value, SNR is signal to noise ratio (S/N ratio).

3. a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image according to claim 1 is characterized in that: the sword limit degree of modulation described in step (6) selects criterion to be:

$M_{\mathrm{SNR}}=\frac{L_a-L_b-\frac{2\·L_v}{\mathrm{SNR}}}{L_a+L_b}>M_{\mathrm{th}}$

M wherein _thFor sword limit threshold modulation wherein, L _aBe the bright side average gray value in sword limit, L _bBe the dark side average gray value in sword limit, L _vBe sword edge image average gray value, SNR is signal to noise ratio (S/N ratio).

4. a kind of hyperspectral remote sensing system MTF detection method based on the sword edge image according to claim 1, it is characterized in that: the method with Gaussian function fitting described in step (7) is found the solution line spread function, and the expression formula of selected Gaussian function is:

$f\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}}{e}^{-\frac{x^2}{2{\mathrm{\σ}}^2}}$

Wherein σ is the Gaussian function standard deviation.

Images