CN102236792A - Method for selecting and extracting characteristic points of selenographic image - Google Patents

Abstract

The invention relates to a method for selecting and extracting characteristic points of a selenographic image, which is used for realizing the automatic three-dimensional reconstruction of a full selenograph. In the method, five kinds of characteristic points, namely a center of a large-area shadow region, a center of a large-area highlight area, a shadow center, a luminance center and a connection line center of the shadow center and the luminance center are selected according to the characteristic of a lunar crater, and are extracted from the selenographic image. The method comprises the following steps of: comparing the luminance of each pixel of the image with a set shadow and highlight luminance threshold value, and extracting the center of the large-area shadow region and the center of the large-area highlight area; scanning the selenographic image, extracting a pixel point of which the luminance in the directions of columns and rows is the minimum value or the maximum value and using the pixel point as the shadow center or the luminance center; and pairing the shadow center and the luminance center according to the incident direction of sunlight, and extracting the connection line center of the shadow center and the luminance center. The method has the advantages that the natural characteristic of the selenograph is utilized skillfully, the calculation is simple and quick, and the extracted characteristic points are high in density.

Description

The unique point of lunar surface image is selected and extracting method

Technical field

The invention belongs to the technical field of computer vision and Geographic Information System, relate to the feature point extraction of image, particularly relate in the reconstructing three-dimensional model process, boat (defending) sheet is carried out the method that unique point is selected and extracted.

Background technology

The successful launch of " Chang'e I " satellite has caused showing great attention to of International Politics, military affairs and academia, and one of its task is exactly to obtain the 3-dimensional image figure of lunar surface.

Feature point extraction is the prerequisite and the basis of corresponding image points coupling, the corresponding image points coupling then is to realize being made up automatically by image the key of lunar surface 3-dimensional image figure, the quantity of feature point extraction, stability and accuracy directly influence the corresponding image points matching rate, and further determine resolution and the precision of the lunar surface 3-dimensional image figure of structure automatically.This is because generation 3-dimensional image figure need obtain the corresponding image points on the different images, and calculate the three-dimensional coordinate of each pixel according to principle of stereoscopic vision, classic method needs more manual intervention, and will utilize the detection data of laser altimeter to replenish and merge.If can from bidimensional image, extract the unique point of sufficient density automatically, and to the CCD stereoscopic camera captured face, the unique point of forward sight and backsight three width of cloth different angles images mates, obtain corresponding image points, and then calculate the three-dimensional coordinate of this point, so just can avoid a large amount of manual interventions, improve one-tenth figure speed and the quality of lunar surface 3-dimensional image figure.This shows, extract lunar surface image feature point and should possess high as far as possible density, and satisfy higher accuracy simultaneously.The extraction of unique point is at present adopted various extraction operators normally at different characteristics of image and different application backgrounds.Moravec operator, Forstner operator are as traditional feature point extraction operator, and principle is directly perceived, algorithm is simple, convenience of calculation, but for complex image, extraction of low quality.Simple, the accurate positioning of Harris operator algorithm is not subjected to the influence of image rotation, grey scale change, but nonshrinkly put, anti-affine deformation function a little less than.The SIFT operator is not subjected to the influence of graphical rule and rotation, and light variation, noise, affine deformation are all had robustness.The SUNSAN operator is compared the effective of rim detection to the detection of angle point, is applicable to the image registration based on corners Matching.But above-mentioned these methods commonly used can not adapt to the image feature of lunar surface fully, and can't satisfy the automatic demand that makes up quantity, stability and the accuracy of unique point of lunar surface 3-dimensional image figure.

Summary of the invention

The objective of the invention is to solve in the automatic building process of moon 3-dimensional image figure, the problem of from the lunar surface bidimensional image, extracting which type of feature and how extracting, the unique point that proposes a kind of lunar surface image is selected and extracting method.

The unique point selection and the extracting method implementation of lunar surface image of the present invention are as follows:

1. according to the characteristics of menology lunar crater, five category features points have been designed: the line center that is respectively center, large tracts of land shadow region, large tracts of land highlight regions center, shade center, brightness center, shade center and brightness center.

2. the selection and the extracting method of above-mentioned 5 category feature points may further comprise the steps:

Step 1: determine shade and highlighted luminance threshold respectively,, calculate center, large tracts of land shadow region and large tracts of land highlight regions center, and calculate parameter---the radius that characterizes feature sizes according to the magnitude relationship of each pixel intensity of lunar surface image and threshold value.Wherein, the center (x, y) and their characteristic radius (radius) determine by following formula:

$\left\{\begin{array}{c}x=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{c}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ y=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{r}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ \mathrm{radius}=\sqrt{n/\mathrm{\π}}\end{array}\right.$

In the formula, n is the pixel number in large tracts of land shade or the large tracts of land highlight regions, (c _i, r _i) be the pixel coordinate of i shadow spots or high bright spot, lum[i] be the brightness of i shadow spots or high bright spot, weight[lum] be that gray scale is the weights of the pixel of lum.

Step 2: the lunar surface image is carried out line scanning and column scan respectively, and brightness is all minimizing pixel on extraction column direction and the line direction, is the shade center; Brightness is all the pixel of maximum value on extraction column direction and the line direction, is the brightness center; Calculate parameter---parameter of radius and the characteristic feature intensity of variation---gradient that characterizes feature sizes then.

The calculating that characterizes the characteristic parameter radius (radius) of shade center or brightness center P was divided into for two steps: at first with shade center or brightness center P the characteristic length smaller on line direction and column direction as initial radium radius, be the center of circle then with P, radius is a radius, set up circular window, according to from distance of center circle from trying to achieve AverageLum[i successively], i represents the distance from the center of circle, AverageLum[i] represent apart from the center of circle to be the mean flow rate of all pixels of i, i=radius ..., 1,0, if AverageLum[i]＜AverageLum[i-1], then radius subtracts 1, and the like, obtain final radius.

Characteristic parameter gradient diff computing formula is as follows:

diff＝(AverageLum[0]-AverageLum[radius])/radius

Step 3: shade center and brightness center are matched according to the sunshine incident direction, be extracted into the right shade center and the line center at brightness center, and calculate parameter---parameter of radius and the characteristic feature intensity of variation---gradient that characterizes feature sizes.

The establishment at line center is according to sun almanac data, calculates the incident direction of sunray, is specified to right shade center and brightness center, and paired shade center and brightness center have uniqueness.Wherein line central feature parameter radius (radius) and gradient (diff) are determined by the parameter at its corresponding shade center and brightness center:

radius＝radius _s+radius _b

diff＝(diff _s+diff _b)/2

Wherein, radius _sBe the shade radius centered, radius _bIt is highlighted radius centered; Diff _sBe the gradient at shade center, diff _bIt is the gradient at highlighted center.

The unique point selection of lunar surface image of the present invention and the advantage of extracting method:

1. because utilize the natural characteristic of lunar surface dexterously, select five category features point, thereby broken away from the drawback of a large amount of manual interventions of classic method needs when making up moon 3-dimensional image figure, automaticity height;

2. the extraction algorithm of She Ji 5 class lunar surface image feature points is simply quick, the extraction efficiency height;

3. according to five class lunar surface image feature point extraction algorithms of design, the unique point density height that extracts is the basis that the automatic constructing technology of moon 3-dimensional image figure is realized.

Description of drawings

Fig. 1 is according to unique point is selected and extracting method obtains the center, large tracts of land shadow region (square frame) and large tracts of land highlight regions center (triangle) synoptic diagram of lunar surface image;

Fig. 2 is according to the shade center that unique point is selected and extracting method obtains of lunar surface image and line center (square frame) synoptic diagram at brightness center.

Fig. 3 is according to unique point is selected and extracting method obtains the shade center (square frame) and brightness center (triangle) synoptic diagram of lunar surface image;

Embodiment

Now in conjunction with the accompanying drawings and embodiments the unique point of lunar surface image of the present invention is selected and extracting method is done and described in further detail.The feature point extraction of the 168th rail image of taking with " Chang'e I " satellite among the embodiment is an example.Lunar surface image feature point selection of the present invention and extracting method, be according to the various lunar craters of the menology characteristics that image had under solar light irradiation, design is also selected five category features point, is respectively: the line center at center, large tracts of land shadow region, large tracts of land highlight regions center, shade center, brightness center, relevant shade center and brightness center; And designed the extraction algorithm of 5 category feature points, can from the lunar surface bidimensional image, extract the number of characteristics point apace, Fig. 1, Fig. 2, Fig. 3 are respectively the synoptic diagram of the variety classes unique point that extracts from the 168th rail image, because image size is very big, have only intercepted partial image.The unique point of lunar surface image of the present invention is selected and extracting method is described in detail as follows:

1. a large amount of impact crater of lunar surface has formed some large-area shadow region and highlight bars under solar light irradiation, their features are obvious, the bearing accuracy height, uniqueness and high conformity are so select large tracts of land shadow region and large tracts of land highlight regions center as unique point.

2. shade center and brightness center are image local regional luminance extreme points: be called the brightness center when this point is the brightness maximum point, be called the shade center during for the brightness minimum point, this two category feature is that solar light irradiation is uneven to be formed owing to landform that lunar surface just rises and falls is subjected to, they exist in image in a large number, bearing accuracy is higher, consistance is better, can remedy large tracts of land shadow region and the sparse shortcoming of large tracts of land highlight regions central distribution, therefore select shade center and brightness center as unique point.

3. a lot of shade centers and brightness center are to occur in pairs, the information that imply at their line center is more than shade center and brightness center, in specific zone, can filter out line center according to certain condition, therefore select the line center at shade center and brightness center as category feature point with uniqueness.

4. the extracting method at center, large tracts of land shadow region is:

1) establishing large tracts of land shadow region luminance threshold is MASS_SILL, and the brightness weights factor is f.

2) get 1 P on the image, its coordinate be (c, r), brightness be I (c, r).If I (c, r)＞MASS_SILL, then the P point does not belong to the shadow region, algorithm is got back to step 2) continuation detection consecutive point; Otherwise, should the point stacked, enter step 3).

3) judge whether stack is empty.Stack top element is popped when the stack non-NULL, as current detection point, enters step 4); Stack enters step 5) when being empty.

4) if current detection point has been inserted the formation of large tracts of land shadow region, then return step 3); Otherwise current detection point is inserted the formation of large tracts of land shadow region, check that then the point and this point that whether have around the current detection point not after testing satisfy I≤MASS_SILL, have and then should put stackedly, and return step 3).

5) from large tracts of land shadow region formation (queue length is n), take out each pixel successively, according to following formula calculate the large tracts of land shadow region the center (x, y) and characterize the parameter radius r adius of provincial characteristics size.

$\left\{\begin{array}{c}x=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{c}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ y=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{r}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ \mathrm{radius}=\sqrt{n/\mathrm{\π}}\end{array}\right.$

In the formula, n is the pixel number in the large tracts of land shadow region, (c _i, r _i) be the pixel coordinate of i shadow spots, lum[i] be the brightness of i shadow spots, weight[lum] be that gray scale is the weights of the pixel of lum, by formula weight[lum]=f ^{(MASS_SILL-lum)}Determine; Large tracts of land shadow region luminance threshold is in the present embodiment

MASS_SILL=32, the brightness weights factor is f=1.5.

1. the extracting method at the extracting method at large tracts of land highlight regions center and center, large tracts of land shadow region is similar, and difference is that check point brightness is considered to this point greater than large tracts of land highlight regions luminance threshold HIGH_SILL and belongs to a large tracts of land highlight regions; Large tracts of land luminance area luminance threshold is HIGH_SILL=110 in the present embodiment, and the brightness weights factor is f=1.5.

2. the extracting method at shade center is:

1) image line is scanned, in structure brightness slope on the line direction (interval that dull rising of brightness or decline pixel constitute in the row or column successive range), find out the slope that aspect rises, getting its left end point is minimum point, gets the parameter l en1 as this minimal value characteristic length size on the sign line direction of smaller in this length of grade of putting left and right two slopes.

2) to image column scanning, using the method identical with previous step to find out brightness is minimizing pixel and characteristic length len2 thereof.

3) find out on the column direction and line direction on brightness be all minimizing pixel, be the shade center.

4) the characteristic attribute parameter at calculating shade center mainly comprises the parameter radius of characteristic feature point size and the parameter d iff of the characteristic feature point gradient.Get len1, len2 smaller value for parameter radius.But at this moment only investigated the four direction at this shade center, the value of parameter radius is inaccurate, must further determine: the shade center is designated as a P, the first step is the center with a P, set up radius for (2 * radius+1) circular window, obtain successively and put P distance for i (the some brightness mean value AverageLum[i of 0≤i≤radius)]; Second step is according to i order from big to small, successively AverageLum[i relatively] and AverageLum[i-1], some P is the shade center, so above-mentioned average brightness satisfies AverageLum[i]＞AverageLum[i-1], if do not satisfy then radius subtracts 1.Determine after the radius that the parameter d iff of the characteristic feature point gradient can be by formula diff=AverageLum[0]-AverageLum[radius] try to achieve.

3. the extracting method at the extracting method at brightness center and shade center is similar, and difference is that the brightness center is the brightness maximum point on line direction and the column direction.

4. the extracting method at the line center at shade center and brightness center is:

1) finds out and the paired brightness center of possibility, shade center.Setting a range threshold R, is being the center of circle with the shade center, and radius is the interior brightness center of seeking of the circle of R.

2) in the brightness that previous step finds suddenly in the heart, determine a brightness center paired according to radiation direction with the shade center.If the shade centre coordinate is (i _s, j _s), the brightness centre coordinate is (i _b, j _b), sunshine incident vector is (dx, d _y, 0), establish d _i=i _s-i _b, d _j=j _s-j _b,

If satisfy condition cosa=(d _i* d _x+ d _j* d _y)/r 〉=cos (π/6) can think that then shade center and this candidate's brightness center are a pair of.If exist a plurality of candidate's brightness center to satisfy above-mentioned condition,,, should abandon extracting this line center to shade center and brightness center for avoiding mismatching in view of paired shade center and brightness center have uniqueness.

3) calculated characteristics property parameters.The characteristic parameter of line is definite by shade center relevant with it and brightness center, wherein coordinate (i, j) middle i=(i _s+ i _b)/2, j=(j _s+ j _b)/2, the parameter radius=radius of sign scope _s+ radius _b, the parameter d iff=(diff of the sign gradient _s+ diff _b)/2.

The unique point of the lunar surface image that above-mentioned case realizes is selected and the extracting method test environment is: CPU PIV 2.8G, 1G internal memory, 120G hard disk, operating system Windows XP, monitor resolution 1024 * 768, GEFORCE6200 (128M video memory) video card.The image size is 512 * 32554, and extracting all kinds of unique point numbers is 120,536.

Claims (5)

1. the unique point of a lunar surface image is selected and extracting method, it is characterized in that, this method designs and has selected five category features points: the line center at center, large tracts of land shadow region, large tracts of land highlight regions center, shade center, brightness center, shade center and brightness center according to the characteristics of menology lunar crater; The unique point of lunar surface image is selected and the realization of extracting method may further comprise the steps:

Step 1: determine shade and highlighted luminance threshold respectively,, extract center, large tracts of land shadow region and large tracts of land highlight regions center, and calculate parameter---the radius that characterizes feature sizes according to the magnitude relationship of each pixel intensity of lunar surface image and threshold value;

Step 2: the lunar surface image is carried out line scanning and column scan respectively, and brightness is all minimizing pixel on extraction column direction and the line direction, is the shade center; Brightness is all the pixel of maximum value on extraction column direction and the line direction, is the brightness center; Calculate parameter---parameter of radius and the characteristic feature intensity of variation---gradient that characterizes feature sizes then;

Step 3: shade center and brightness center are matched according to the sunshine incident direction, be extracted into the right shade center and the line center at brightness center, and calculate parameter---parameter of radius and the characteristic feature intensity of variation---gradient that characterizes feature sizes.

2. select and extracting method according to the unique point of the described a kind of lunar surface image of claim 1, it is characterized in that: the center of step 1 center, described large tracts of land shadow region and large tracts of land highlight regions (x, y) and its characteristic parameter radius (radius) determine by following formula:

$\left\{\begin{array}{c}x=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{c}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ y=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}{r}_i\×\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]/\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\mathrm{weight}\left[\mathrm{lum}\right[i\left]\right]\\ \mathrm{radius}=\sqrt{n/\mathrm{\π}}\end{array}\right.$

In the formula, n is the pixel number in large tracts of land shade or the large tracts of land highlight regions, (c _i, r _i) be the pixel coordinate of i shadow spots or high bright spot, lum[i] be the brightness of i shadow spots or high bright spot, weight[lum] be that gray scale is the weights of the pixel of lum.

3. select and extracting method according to the unique point of the described a kind of lunar surface image of claim 1, it is characterized in that the calculating that described step 2 characterizes the characteristic parameter radius (radius) of shade center or brightness center P was divided into for two steps: at first with shade center or brightness center P the characteristic length smaller on line direction and column direction as initial radium radius; Be the center of circle with P then, radius is a radius, sets up circular window, according to from distance of center circle from trying to achieve AverageLum[i successively], i represents the distance from the center of circle, AverageLum[i] expression is the mean flow rate of all pixels of i apart from the center of circle, i=radius, ..., 1,0, if AverageLum[i]＜AverageLum[i-1], then radius subtracts 1, and the like, obtain final radius;

Characteristic parameter gradient diff computing formula is as follows:

diff＝(AverageLum[0]-AverageLum[radius?D/radius。

4. select and extracting method according to the unique point of the described a kind of lunar surface image of claim 1, it is characterized in that: the establishment at described step 3 line center is according to sun almanac data, calculate the incident direction of sunray, be specified to right shade center and brightness center, and paired shade center and brightness center has uniqueness.

5. select and extracting method according to the unique point of the described a kind of lunar surface image of claim 1, it is characterized in that: the line central feature parameter radius (radius) of described step 3 and gradient (diff) are determined by the parameter at its corresponding shade center and brightness center:

radius＝radius _s+radius _b

diff＝(diff _s+diff _b)/2

Wherein, radius _sBe the shade radius centered, radius _bIt is highlighted radius centered; Diff _sBe the gradient at shade center, diff _bIt is the gradient at highlighted center.

Images