CN102073996B - Image-correlation-evaluation-based method for determining image segmentation threshold - Google Patents

Abstract

The invention provides an image-correlation-evaluation-based method for determining an image segmentation threshold. On the basis of an imaging theory, a signal processing theory and other traditional theories, correlations between image effective spots and background light and stray light are considered, namely the covariance (between-class variance) between an effective spot part image and a background stray light image is analyzed, wherein the maximum coefficient of the covariance refers to the minimum correlation between the effective spot part image and the background stray light image, and the threshold segmentation is best in the state, so the background light, the stray light and the effective light source image can be separated. The method can effectively inhibit the interference of the stray light to the effective spots, and the spots are accurately extracted.

Description

Image segmentation threshold value based on the image correlation assessment is confirmed method

Technical field

The present invention relates to a kind of Technique in Rendezvous and Docking with the image processing method of optical imagery sensor under complex background.

Background technology

Increasing along with the space mission value volume and range of product, the especially development of space station makes the Spacecraft Rendezvous technology obtain significant progress.The closely optical measuring apparatus that CCD optical imagery sensor docks with the target aircraft intersection as tracker, its image quality to the successful intersection of spacecraft to being connected to very big influence.

According to the requirement of intersection docking mission, the blip device is fixedly mounted on predetermined measured zone on the target aircraft according to certain mode usually.CCD optical imagery sensor forms images to this measured zone, and the image that obtains is carried out luminous point extract and discern, and calculates and also exports relative position and the relative attitude angle of optical sensor coordinate system with respect to blip device coordinate system.Effective luminous point that therefore how from CCD optical imagery sensor image, will constitute the blip device extracts from bias light the normal measurement that directly influences CCD optical imagery sensor, and the cut-point between effective luminous point and the background image is image threshold.

Existing method mainly uses fixed threshold (as add three times of variances etc. with image background; Specifically can be referring to M.Sezgin and B.Sankur; Survey over image thresholding techniques and quantitative performance evaluation; Journal of Electronic Imaging, pp.146-156,2003) image is carried out Threshold Segmentation; These methods can not split effective luminous point from background image in complex background image, thereby influence the normal measurement of CCD optical imagery sensor.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiency of prior art, provide a kind of image segmentation threshold value based on the image correlation assessment to confirm method, can effectively suppress the interference of parasitic light to effective luminous point, accurately luminous point is extracted.

Technical solution of the present invention is: the image segmentation threshold value based on the image correlation assessment is confirmed method, and step is following:

(1) set up the picture noise distribution histogram, find out the noise maximal value and with it as noise gate;

(2) noise gate of confirming with step (1) is a benchmark, uniformly discrete pixel extraction greater than noise gate in the image is come out according to the SI, adds up the number of pixels S of each gray scale _i, i ∈ [hmin, hmax], hmin and hmax are respectively minimum gradation value and maximum gradation value, the background mean value background that calculates image and the number of pixels of each gray scale shared ratio in image;

(3) travel through each gray-scale value from minimal gray hmin successively to maximum gray scale, add up respectively from minimal gray hmin to current gray level n number of pixels sum miu1 and greater than the number of pixels sum miu2 of current gray level n smaller or equal to maximum gray scale hmax,

$\mathrm{miu}1=\frac{\underset{i=h\min }{\overset{n}{\mathrm{\Σ}}}(i\×S_i)}{\mathrm{omega}1},$ $\mathrm{omega}1=\underset{i=h\min }{\overset{n}{\mathrm{\Σ}}}{S}_i,$ $\mathrm{miu}2=\frac{\underset{i=n}{\overset{h\min }{\mathrm{\Σ}}}(i\×S_i)}{\mathrm{omega}2},$ $\mathrm{omega}2=\underset{i=n+1}{\overset{h\max }{\mathrm{\Σ}}}{S}_i;$

(4) travel through each gray-scale value successively; The current gray level value as the separatrix; With image be divided into be higher than marginal part be lower than marginal part, calculate two-part inter-class variance xigma, if the inter-class variance that calculates is greater than the inter-class variance and the predetermined offset sum of previous stage gray-scale value; Then with current gray scale as current theoretical threshold value, the inter-class variance computing formula is:

Xigma=omega1 * (miu1-background) ²+ omega2 * (miu2-background) ²+ predetermined offset;

(5) utilize filter function that the threshold value of current theoretical threshold value and previous frame image is carried out weighted mean, obtain the present image threshold value, computing formula is:

Present image threshold value=current theoretical the threshold value of (1-filter function) *+filter function * previous frame image threshold.

Described predetermined offset is obtained by the statistical average background of the multiple image statistical average threshold value divided by correspondence.

The present invention's advantage compared with prior art is: the inventive method is the basis with the traditional theory of imaging theory and signal Processing etc.; Through considering the correlativity between the effective luminous point of image and bias light and the parasitic light; Promptly analyze the covariance size (inter-class variance) between effective luminous point parts of images and the background miscellaneous light image; Correlativity between maximum effective luminous point parts of images of expression of covariance coefficient and the background miscellaneous light image is minimum; Explain that promptly the Threshold Segmentation under this state is best; Thereby with bias light, parasitic light and efficient light sources image area separately, reduced the influence of sunshine, earth light and target aircraft surface emitting light simultaneously, strengthened the ability to work of CCD optical imagery sensor under the complex background environment CCD optical imagery sensor normal attitude position calculation.Simultaneously; The inventive method has been improved traditional threshold value and has been confirmed the dependence of method to the image background degree of purity; Can guarantee CCD optical imagery sensor operate as normal under complicated optical environment; When satisfying the requirement of intersection butt joint different operating period measuring task, practiced thrift the Flame Image Process time, the development of high robust, high refresh rate intersection being docked sensor has important engineering use value.

Description of drawings

Fig. 1 is the FB(flow block) of the inventive method.

Embodiment

As shown in Figure 1, be the theory diagram of the inventive method, after the input of CCD head integral image, the key step that image threshold is confirmed is following:

1, the horizontal statistical appraisal of gradation of image

The gray scale of all pixels of light spot signal should be greater than certain threshold value.The noise maximal value that employing is confirmed by the histogram of noise profile is as thresholding.At first, add up the number of pixels of each gray level, the number of pixels of background mean value and each gray scale that calculates image shared ratio in image according to the plurality of pixels in the uniformly discrete image of choosing the specific region of SI.

If the number of pixels of each gray level is respectively: S ₀, S ₁, S ₂..., S ₂₅₅

2, image correlation analysis

Successively from the minimal gray level to maximum gray scale the traversal each grey level, the statistics from the minimal gray level to this gray level the number of pixels sum and greater than the number of pixels sum of this gray scale smaller or equal to maximum gray scale.

Number of pixels sum from minimal gray level (hmin) to this gray level (being made as n):

$\mathrm{miu}1=\frac{\underset{i=h\min }{\overset{n}{\mathrm{\Σ}}}(i\×S_i)}{\mathrm{omega}1},$ $\mathrm{omega}1=\underset{i=h\min }{\overset{n}{\mathrm{\Σ}}}{S}_i$

More than or equal to the number of pixels sum of this gray scale (being made as n) smaller or equal to maximum gray scale (hmax):

$\mathrm{miu}2=\frac{\underset{i=n}{\overset{h\min }{\mathrm{\Σ}}}(i\×S_i)}{\mathrm{omega}2},$ $\mathrm{omega}2=\underset{i=n+1}{\overset{h\max }{\mathrm{\Σ}}}{S}_i$

3, calculating of image covariance and optimum variance chooses with definite

Travel through each gray-scale value successively, the current gray level value as the separatrix, with image be divided into be higher than marginal part be lower than marginal part, calculate two-part inter-class variance xigma,

xigma＝omega1×(miu1-background) ²+omega2×(miu2-background) ²；

xigma＝xigma+0.1

Travel through each gray level inter-class variance successively; If greater than the previous stage gray-scale value inter-class variance add side-play amount; Then with current gray level as threshold value, too violent in order to guarantee between image upper level class with current inter-stage covariance variable gradient, increased the inter-stage side-play amount and come correction threshold.After carrying out inter-class variance gray scale traversal calculating completion; Make inter-class variance and predetermined offset sum obtain peaked gray-scale value as current theoretical threshold value; The selection principle that predetermined offset is general is to be predetermined offset to multiple image statistical average background divided by likening to of statistical average threshold value, and this side-play amount gets 0.1 usually.

4, adaptive threshold is confirmed

Current theoretical threshold value and previous frame image threshold carry out weighted mean according to filter function, obtain and find best threshold value, can accurately aiming pip and background miscellaneous light be distinguished, and computing formula is:

Current threshold value=current theoretical the threshold value of (1-filter function) *+filter function * previous frame image threshold.

Filter function is to obtain according to 1000 captured width of cloth image statistics backgrounds and changes of threshold gradient, and directly value is 0.1 at present.

The content of not doing to describe in detail in the instructions of the present invention belongs to those skilled in the art's known technology.

Claims (1)

1. confirm method based on the image segmentation threshold value of image correlation assessment, it is characterized in that step is following:

(1) set up the picture noise distribution histogram, find out the noise maximal value and with it as noise gate;

(2) noise gate of confirming with step (1) is a benchmark, uniformly discrete pixel extraction greater than noise gate in the image is come out according to the SI, adds up the number of pixels S of each gray scale _i, i ∈ [hmin, hmax], hmin and hmax are respectively minimum gradation value and maximum gradation value, the background mean value background that calculates image and the number of pixels of each gray scale shared ratio in image;

(3) travel through each gray-scale value from minimal gray hmin successively to maximum gray scale, add up respectively from minimal gray hmin to current gray level n number of pixels sum miu1 and greater than the number of pixels sum miu2 of current gray level n smaller or equal to maximum gray scale hmax,

(4) travel through each gray-scale value successively; The current gray level value as the separatrix; With image be divided into be higher than marginal part be lower than marginal part, calculate two-part inter-class variance xigma, if the inter-class variance that calculates is greater than the inter-class variance and the predetermined offset sum of previous stage gray-scale value; Then with current gray scale as current theoretical threshold value, the inter-class variance computing formula is:

Xigma=omega1 * (miu1-background) ²+ omega2 * (miu2-background) ²+ predetermined offset, wherein predetermined offset is obtained by the statistical average background of the multiple image statistical average threshold value divided by correspondence;

(5) utilize filter function that the threshold value of current theoretical threshold value and previous frame image is carried out weighted mean, obtain the present image threshold value, computing formula is:

Present image threshold value=current theoretical the threshold value of (1-filter function) *+filter function * previous frame image threshold.

Images