CN103679648A - Moment matching satellite image stripe noise removing method based on spatial segmentation - Google Patents

Abstract

A moment matching satellite image stripe noise removing method based on spatial segmentation. The method segments ground objects with different spectral characteristics based on features of a mean, a median, and a gradient, and each segmented region is processed by a standard moment matching. The method is aimed at a remote sensing image with a large amount of data; and based on a basic image processing theory, an effective image noise removing method is provided. The method can effectively restore the image, improve the image quality, and is quite high in algorithm efficiency.

Description

A kind of square coupling satellite image Strip noise removal method of cutting apart based on space

Technical field

The invention belongs to remote sensing image process field, relate to a kind of square coupling satellite image Strip noise removal method of cutting apart based on space.

Background technology

Band noise is a key factor that affects optical satellite image image quality, and suppressing or removing band noise is one of satellite ground pre-service basic link of carrying out irradiation treatment.The imaging system of optical satellite is due to the interference of internal and external factors, as: the performance of CCD device (charge-coupled image sensor) changes in time, atmospheric interference etc., after carrying out the relative radiant correction of image homogenization, in image, still can remain band noise, the existence of this noise like, greatly reduce the sharpness of image, for the follow-up interpretation of image, process and increased difficulty, therefore must be rejected.

By being analyzed, band noise residual in image finds that such band noise has following characteristics: 1. the locus that noise occurs is random.Noise to around between atural object nonlinearity relevant.

The conventional method of removing band noise is summed up and can be divided into two classes at present: a class is the denoising method proposing for image space characteristic of field; Another kind of is that spatial domain and frequency field are combined, and adopts suitable filter operator to remove the method for band noise.Wherein the typical algorithm of spatial domain denoising has the Strip noise removal method of the multi-load image based on square coupling of people's propositions such as Gadallahd, the method requires gradation of image homogeneous, and band noise image is adjacent the higher correlativity of existence between image, this class algorithm has good effect for processing the local single satellite image of atural object, but when processing the abundant satellite image of view picture atural object, poor effect; So on this basis, the people such as Liu Zhengjun have proposed the improvement square matching process based on average compensation, the method utilizes the multispectral information of high spectrum image to correct the tonal distortion that standard square matching algorithm produces, the method has well been processed the tonal distortion between the spy unit causing when standard square matching algorithm is corrected image, but do not correct the tonal distortion in the spy unit that standard square matching algorithm causes, and the method is for single-range panchromatic image inapplicable; The improvement square that the people such as Liu Yan have proposed based on level set mates random Strip noise removal algorithm, although between the spy unit that this algorithm has suppressed to have caused when standard square matching algorithm is corrected image to a certain extent and visit the tonal distortion in unit, but the method for cutting apart based on image greyscale precision when distinguishing the atural object of different spectral characteristics is not high, can be because atural object produces random tonal distortion for wrong minute when processing high resolution ratio satellite remote-sensing image; The denoise algorithm that spatial domain and frequency field combine is mainly to utilize the time-frequency feature of wavelet transformation at present, by image is carried out to wavelet transformation, the wavelet coefficient rule of conversion of research noise, thereby extract the composition of band noise and it is rejected, but wavelet transformation calculated amount is large, and can cause greater loss to spectral information, for the processing of view picture satellite image, computing velocity and spectrum reservation degree are unsatisfactory.

Summary of the invention

The technical matters that the present invention solves is: overcome the deficiencies in the prior art, for the random band noise existing in panchromatic image, a kind of square coupling satellite image Strip noise removal method of cutting apart based on space is provided, solved the problem that standard square matching process can not be applicable to the inhomogeneous satellite image of atural object.

Technical scheme of the present invention is: a kind of square coupling satellite image Strip noise removal method of cutting apart based on space, comprises the following steps:

1) each pixel (x, y) of pending satellite image is set up to neighborhood scope k * k, wherein k is positive odd number,, centered by pixel (x, y), selects the region L of k * k size; Obtain the mean value f (x, y) of the gray-scale value of all pixels in the L of region, obtain the intermediate value g (x, y) of the gray-scale value of all pixels in the L of region, obtain gradient $G(x,y)=\sqrt{G_x^2+{\mathrm{<figure-callout\; id="2"\; label="G\; y"\; filenames="HDA0000416413470000012.png"\; state="\{\{state\}\}">G</figure-callout>}}_y^{}},$ Wherein $G_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]*L,G_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]*L,$ Wherein * is convolution symbol;

Take average f, the intermediate value g of each pixel, the three-dimensional coordinate that gradient G is this pixel, be mapped to f, g, G and be respectively in the three-dimensional cartesian coordinate system of three axles, build the three-dimensional co-occurrence matrix model of pending satellite image; And segmentation times N is set, wherein N is greater than 1 positive integer;

2) in the three-dimensional co-occurrence matrix model of setting up in step 1), initial segmentation point s (f is set ₀, g ₀, G ₀); To cross some s (f ₀, g ₀, G ₀) and be parallel to fOg plane, cross a some s (f ₀, g ₀, G ₀) and be parallel to the plane of gOG and cross some s (f ₀, g ₀, G ₀) and the plane that is parallel to fOG three-dimensional co-occurrence matrix is divided into eight sub spaces; Four sub spaces that are defined in fOg plane are lower floor subspace, and all the other four sub spaces are subspace, upper strata; In four sub spaces of upper strata, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be reference zone A ₀and B ₀; In lower floor's four sub spaces, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be zoning C ₀and D ₀;

3) along G axle, move s point, Real-time Obtaining reference zone A in moving process ₀, B ₀in entropy and, get and maximum of points be the optimal partition point s on note G axle _g(f ₀, g ₀, G ₀'), and regain reference zone A _0Gand B _0G; Along f axle, move s _gpoint, Real-time Obtaining reference zone A in moving process _0Gand B _0Gin entropy and, get and maximum of points be designated as the optimal partition point s on f axle _fG(f ₀', g ₀, G ₀'), and regain reference zone A _0fGand B _0fG; Along g axle, move s _fGpoint, Real-time Obtaining reference zone A in moving process _0fGand B _0fGentropy and, get and maximum of points be designated as the optimal partition point s on g axle _fgG(f ₀', g ₀', G ₀'), and regain reference zone A _0fgGand B _0fgG; By optimal partition point s _fgGbe designated as s', with reference to region A _0fgGand B _0fgG, be designated as reference zone A ₁and B ₁, obtain eight sub spaces that the three-dimensional co-occurrence matrix first order is cut apart, and obtain the zoning C that the first order is cut apart ₁and D ₁;

4) the zoning C obtaining in step 3) ₁, D ₁in each region in repeating step 2), step 3), obtain the zoning that four second level are cut apart;

5) repeating step 4) until the iterations N arranging is finished, obtain the zoning of final 2N;

6) calculation procedure 5) in average and the variance of each corresponding grey scale pixel value in zoning of obtaining, adopt standard square matching process, recalculate the gray-scale value of this corresponding pixel in region, finally remove the band noise of pending satellite image.

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

For the large remote sensing panchromatic image of data volume, based on basic treatment theory, provide a kind of effective image noise removal method.The method can be cut apart the atural object of different spectral signatures according to the feature of average, intermediate value, gradient, and each cut zone is adopted to standard square matching treatment, enough reconstructed images effectively, improves the quality of image, and efficiency of algorithm is also very high.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the embodiment of the present invention;

Fig. 2 is three-dimensional co-occurrence matrix quadrant schematic diagram.

Embodiment

According to the basic theories of image processing, can know, the histogrammic different peak values of image have represented the different types of ground objects that comprise in image, although standard square matching process can cause tonal distortion when processing the abundant image of atural object, at image histogram, there is a plurality of peak values or histogram peak curve span larger in the situation that, employing standard square matching process can cause tonal distortion phenomenon, but the curve distribution for the less single peak value of tonal range span, standard square matching process is when carrying out noise removal process to image, can't cause and in image, produce tonal distortion phenomenon.Therefore the single inner imaging data of unit of visiting is distinguished according to the difference of image histogram distribution (being different gray shade scales), then ask for respectively variance and the average of different gray areas imaging datas, the last imaging data to different gray areas, carry out respectively standard square matching treatment, so just can solve standard square coupling because average changes tonal distortion between the spy unit causing and because variance changes the tonal distortion in the spy unit causing.

Below in conjunction with Fig. 1 and specific embodiment, describe technical solution of the present invention in detail.

1) I (x, y) be satellite image pixel (x, y) gray-scale value of locating, establishes this neighborhood of pixel points size k * k (choosing k=3 in the present embodiment) centered by pixel (x, y), select the region L of 3 * 3 sizes, and calculate its average f (x, y) for the mean value of the gray-scale value of all pixels in L, calculate intermediate value g (x, y) intermediate value of the gray-scale value of all pixels in L, gradient

wherein $G_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]*L,$ $G_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]*L,$ * be convolution symbol; Take average f, the intermediate value g of each picture element and the three-dimensional coordinate that gradient G is this picture element, be mapped to f, g and G and be respectively in the three-dimensional cartesian coordinate system of three axles, build the three-dimensional co-occurrence matrix model of pending satellite image; And segmentation times N=3 is set;

2) 1) given initial segmentation point s (f in the three-dimensional co-occurrence matrix model set up ₀, g ₀, G ₀) be, to cross some s (f ₀, g ₀, G ₀) and be parallel to fOg plane, cross a some s (f ₀, g ₀, G ₀) and be parallel to the plane of gOG and cross some s (f ₀, g ₀, G ₀) and the plane that is parallel to fOG three-dimensional co-occurrence matrix is divided into eight sub spaces, see Fig. 2; Four sub spaces that are defined in fOg plane are lower floor subspace, and all the other four sub spaces are subspace, upper strata; In four sub spaces of upper strata, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be reference zone A ₀and B ₀, A in Fig. 2 ₀for region 5, B ₀for region 4; In lower floor's four sub spaces, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be zoning C ₀and D ₀, C in Fig. 2 ₀for region 1, D ₀for area 0;

3) along G axle, move s point, Real-time Obtaining reference zone A in moving process ₀, B ₀in entropy and, get and maximum of points be the optimal partition point s on note G axle _g(f ₀, g ₀, G ₀'), and regain reference zone A _0Gand B _0G; Along f axle, move s _gpoint, Real-time Obtaining reference zone A in moving process _0Gand B _0Gin entropy and, get and maximum of points be designated as the optimal partition point s on f axle _fG(f ₀', g ₀, G ₀'), and regain reference zone A _0fGand B _0fG; Along g axle, move s _fGpoint, Real-time Obtaining reference zone A in moving process _0fGand B _0fGentropy and, get and maximum of points be designated as the optimal partition point s on g axle _fgG(f ₀', g ₀', G ₀'), and regain reference zone A _0fgGand B _0fgG; By optimal partition point s _fgGbe designated as s', with reference to region A _0fgGand B _0fgG, be designated as reference zone A ₁and B ₁, obtain eight sub spaces that the three-dimensional co-occurrence matrix first order is cut apart, and obtain the zoning C that the first order is cut apart ₁and D ₁;

4) the zoning C obtaining in step 3) ₁, D ₁in each region in repeating step 2), step 3), obtain the zoning that 4 second level are cut apart;

5) repeat 4), obtain 8 final zonings;

6) to 5) in each zoning of obtaining, calculate average and the variance of this corresponding grey scale pixel value in region, adopt standard square matching process, recalculate the gray-scale value of this corresponding pixel in region, finally remove the band noise of pending satellite image.

The content not being described in detail in the present invention belongs to professional and technical personnel in the field's known technology.

Claims (1)

1. a square coupling satellite image Strip noise removal method of cutting apart based on space, is characterized in that step is as follows:

1) each pixel (x, y) of pending satellite image is set up to neighborhood scope k * k, wherein k is positive odd number,, centered by pixel (x, y), selects the region L of k * k size; Obtain the mean value f (x, y) of the gray-scale value of all pixels in the L of region, obtain the intermediate value g (x, y) of the gray-scale value of all pixels in the L of region, obtain gradient $G(x,y)=\sqrt{G_x^2+G_y^2},$ Wherein $G_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]*L,G_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]*L,$ Wherein * is convolution symbol;

Take average f, the intermediate value g of each pixel, the three-dimensional coordinate that gradient G is this pixel, be mapped to f, g, G and be respectively in the three-dimensional cartesian coordinate system of three axles, build the three-dimensional co-occurrence matrix model of pending satellite image; And segmentation times N is set, wherein N is greater than 1 positive integer;

2) in the three-dimensional co-occurrence matrix model of setting up in step 1), initial segmentation point s (f is set ₀, g ₀, G ₀); To cross some s (f ₀, g ₀, G ₀) and be parallel to fOg plane, cross a some s (f ₀, g ₀, G ₀) and be parallel to the plane of gOG and cross some s (f ₀, g ₀, G ₀) and the plane that is parallel to fOG three-dimensional co-occurrence matrix is divided into eight sub spaces; Four sub spaces that are defined in fOg plane are lower floor subspace, and all the other four sub spaces are subspace, upper strata; In four sub spaces of upper strata, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be reference zone A ₀and B ₀; In lower floor's four sub spaces, definition and the crossing subspace of G axle, and with the subspace of this subspace symmetry be zoning C ₀and D ₀;

3) along G axle, move s point, Real-time Obtaining reference zone A in moving process ₀, B ₀in entropy and, get and maximum of points be the optimal partition point s on note G axle _g(f ₀, g ₀, G ₀'), and regain reference zone A _0Gand B _0G; Along f axle, move s _gpoint, Real-time Obtaining reference zone A in moving process _0Gand B _0Gin entropy and, get and maximum of points be designated as the optimal partition point s on f axle _fG(f ₀', g ₀, G ₀'), and regain reference zone A _0fGand B _0fG; Along g axle, move s _fGpoint, Real-time Obtaining reference zone A in moving process _0fGand B _0fGentropy and, get and maximum of points be designated as the optimal partition point s on g axle _fgG(f ₀', g ₀', G ₀'), and regain reference zone A _0fgGand B _0fgG; By optimal partition point s _fgGbe designated as s', with reference to region A _0fgGand B _0fgG, be designated as reference zone A ₁and B ₁, obtain eight sub spaces that the three-dimensional co-occurrence matrix first order is cut apart, and obtain the zoning C that the first order is cut apart ₁and D ₁;

4) the zoning C obtaining in step 3) ₁, D ₁in each region in repeating step 2), step 3), obtain the zoning that four second level are cut apart;

5) repeating step 4) until the iterations N arranging is finished, obtain the zoning of final 2N;

6) calculation procedure 5) in average and the variance of each corresponding grey scale pixel value in zoning of obtaining, adopt standard square matching process, recalculate the gray-scale value of this corresponding pixel in region, finally remove the band noise of pending satellite image.

Images