CN100458847C - Digital image edge information extracting method - Google Patents

Abstract

The invention relates to a method for extracting the edge information of digit image, which is characterized in that: fixing the gradient image of digit image; shielding the gradient image based on the human visual sensitive character; based on the shield gradient image, extracting the edge information of digit image. The invention can consist the edge extracted result and the analyzed edge condition, to make the edge extracted effect meet the human visual sensitive character, to realize better following application and improve the digit image quality in video communication.

Description

A kind of digital image edge information extracting method

Technical field

The present invention relates to the video technique field, be specifically related to a kind of digital image edge information extracting method.

Background technology

Up-to-date psychology of vision studies show that the mankind mainly concentrate in the sub-fraction key message of entire image information for image observation and understanding, by holding these key messages, just can understand image well.

Because the edge is one of the most basic feature of image, the edge has comprised the image information that is used to discern, it is the key character parameter of people's description, recognition objective and interpretation of images, the edge extensively is present between the object and background in the image, between object and the object, HVS (the Human Visual System mankind's vision systems) is the most responsive for the marginal information in the video image, so image edge information is a very important information in the above-mentioned key message.

At present, multimedia video communication relates to all aspect a lot of that each frame of digital image edge information in the video sequence is extracted and based on the processing of marginal information, as adopting the edge to strengthen (edge enhancement) technology in processing procedure before picture coding and the decoded processing procedure, by when the compressed encoding, the local overabsorption bit many to the edge, and, improve the quality of video digital images to the few allocation bit in few place, edge; As in the compressed encoding of based target (Object-based or shape-based) image/video, accurately extracting marginal information is the basic premise that correctly target is split from background; For another example in an emerging class video quality evaluation method, adopt appraisal procedure based on marginal information, contrast the quality correlativity that diminishes after original image and the compression/transmission between (impaired) image, thereby determine to diminish the quality of image.

It can be said that, the extraction of digital image edge information (edge extraction), be that rim detection (edgedetection) is a basic technology in the multimedia video communication, improve the performance of this technology, will increase for the overall performance in the video communication.

Present edge information extracting method depends on edge extracting operator (operator).In general, the edge extracting operator all is a linear operator, promptly obtain the gradient image of original digital picture, by methods such as thresholding differentiations, determine image edge information then by digital picture being carried out linear discrete convolution algorithm (lineardiscrete convolution).

Existing edge information extracting method based on the edge extracting operator mainly comprises two kinds: based on the edge information extracting method of classical operator with extract the edge information extracting method of operator based on optimal edge.

Be widely used, based on the realization principle of the edge information extracting method of classical operator be: at first, the gradient of computed image then, is determined image edge information according to the size of each pixel gradient value.

In fact, strictly speaking, gradient is the notion that exists for consecutive image, because according to the infinitesimal analysis theory of the multivariate function, consecutive image as volume coordinate (x, binary function f y) (x, y), its gradient is that (x, y) with respect to x, the single order partial derivative of y defines by gradation of image f.For discrete picture,,, can adopt discrete differential (discrete difference) to come the method for approximate replacement partial derivative to define " gradient " of discrete picture so gradient is an approximate concept because there is not partial derivative in discrete picture.Digital picture is the special circumstances of discrete picture, and digital picture not only its volume coordinate is separated into integer i, j, and gradation of image f (x y) also can only get discrete numerical value, and therefore, digital picture also is fit to asks for its " gradient " with discrete differential.

In the application of reality, the Grad of pixel zonule operator commonly used such as convolution template and image convolution are come approximate treatment, and different operators has formed the edge detection operator of multiple classics, i.e. gradient operator.Classical gradient operator comprises: Sobel operator, Kirsch operator, Prewitt operator, Roberts operator, Laplacian operator, some operator, line operator etc.

The pixel coordinate of discrete picture is generally used integer i, and j represents that (i j) locates gradation of image f (i, gradient j) to pixel

For:

$\stackrel{\→}{g}(i,j)={[g_x(i,j),g_y(i,j)]}^T---\left(1\right)$

Wherein: component g _x(i, j), g _y(i, j) difference remarked pixel f (i, j) discrete differential on x direction and y direction.

The difference of different classical gradient operators is g _x(i, j), g _y(i, concrete form difference j).

Be that example is described in detail the edge information extracting method based on classical operator with the most frequently used Sobel operator below.

In the method, at first should pass through formula (2) and obtain g _x(i, j), g _y(i, j):

g _x(i，j)＝(f(i+1，j-1)+2f(i+1，j)+f(i+1，j+1))

-(f(i-1，j-1)+2f(i-1，j)+f(i-1，j+1))

g _y(i，j)＝(f(i-1，j-1)+2f(i，j-1)+f(i+1，j-1)) (2)

-(f(i-1，j+1)+2f(i，j+1)+f(i+1，j+1))

Hence one can see that, calculates g _x(i, j), g _y(i, gradient operator j) is respectively:

$G_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]---\left(3\right)$

$G_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]$

Then, use G _x, G _y(i, j) disperse volume collection obtains gradient image with original image f respectively.

After obtaining gradient image, can use a thresholding to judge each pixel grey scale in the gradient image

And the magnitude relationship between this thresholding, if the gray scale of pixel

Be higher than threshold value, determine that then this pixel is an edge pixel, otherwise be non-edge pixel.

The edge information extracting method of other classical operator and Sobel operator is basic identical, and its difference only is G _x, G _yThe matrix difference.

In edge information extracting method based on classical operator, specific operator can only detect the marginal information of specific direction, and it is responsive to noise ratio, therefore, when image is carried out actual treatment, if do not know the edge of image distribution arrangement in advance, so, select when improper at operator, the extraction effect of marginal information will be poor, thereby make the of poor quality of digital picture.And, because the situation of the various piece of image differs greatly, can't judge marginal information with a kind of thresholding uniformly usually, therefore, rely on single thresholding to judge whether certain is put is the method for marginal point, be in actual applications have very much circumscribed.In addition, based on the result of the edge information extracting method of classical operator in subsequent applications, especially with the closely-related subsequent applications of human perception in, visual effect will be poor.

Optimal edge extracts operator and is based on the optimum operator that certain optiaml ciriterion (Optimal Criteria) designs, and the Canny operator is wherein very representational a kind of algorithm.Be that example is described in detail the edge information extracting method that extracts operator based on optimal edge with the Canny operator below.

With respect to classical operator, the Canny edge detection operator has better rim detection performance.According to the validity of rim detection and the reliability of location, the Canny operator based on three optiaml ciriterions be:

1, good signal to noise ratio (S/N ratio), i.e. omission marginal point and that non-marginal point is judged to the probability of marginal point is low.

2, good location, promptly detected marginal point will be as far as possible near the center of actual edge.

3, to unique response at single edge, the probability that promptly single edge produces a plurality of responses is low, and the false edge response will obtain maximum the inhibition.

Concrete steps based on the edge information extracting method of Canny operator are:

Step 1, image is carried out smothing filtering,, is specially to reduce The noise with the 2-d gaussian filters device:

H(i，i)＝f(x，y)*G(i，j) (4)

Wherein: (i j) is original image to f, and (i j) is filtered image to H, and (i j) is Gaussian filter to G.

(i j) is obtained by the discretize for the impulse response function (Impulse response function) of continuous Gaussian wave filter G.This impulse response function is as shown in Equation (5):

$G(x,y)=\frac{1}{2{\mathrm{\&pi;\&sigma;}}^2}\exp (-\frac{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="C20051007706100241.png"\; state="\{\{state\}\}">y</figure-callout>}}^2}{2{\mathrm{\&sigma;}}^2})---\left(5\right)$

G (i j) as a discrete convolution operator, can discretize become 3 * 3 as required, and 5 * 5,7 * 7 or bigger matrix, provide 5 * 5 matrixes commonly used here, as shown in Figure 1.

The gradient of step 2, computed image obtains gradient image

Wherein: $\stackrel{\&RightArrow;}{g}(i,j)={[g_x(i,j),g_y(i,j)]}^T,$ And, $g_x(i,j)\&ap;\frac{\&PartialD;f(x,y)}{\&PartialD;x}\left|\begin{array}{c}x=\mathrm{i\&Delta;}\\ y=\mathrm{j\&Delta;}\end{array}\right.=f(i+1,j)-f(i-1,j)---\left(6\right)$

$g_y(i,j)\&ap;\frac{\&PartialD;f(x,y)}{\&PartialD;y}\left|\begin{array}{c}x=\mathrm{i\&Delta;}\\ y=\mathrm{j\&Delta;}\end{array}\right.=f(i,j+1)-f(i,j-1)---\left(7\right)$

Wherein: Δ is the sampling interval (sampling interval) of consecutive image discretize.

Mould (magnitude) according to formula (8) compute gradient image:

$\left|\stackrel{\&RightArrow;}{g}(i,j)\right|=\sqrt{{\left(g_x(i,j)\right)}^2+{\left(g_y(i,j)\right)}^2}---\left(8\right)$

According to the direction of formula (9) compute gradient image, promptly with the angle of horizontal direction.

Step 3, setting thresholding are determined the edge of image point according to the Grad of image.The Canny algorithm is set two thresholdings of height, the Grad of image is defined as marginal point greater than the point of high threshold, give up in the image Grad less than the point of low threshold, if the pixel of Grad between two thresholdings is communicated with other marginal point in the image, as being communicated with definition according to 8, if have edge pixel in 8 promptly adjacent pixels, then determine that it is marginal point, otherwise give up with this pixel.

With respect to edge extracting method based on classical operator, three optiaml ciriterions extracting the edge information extracting method institute foundation of operator based on optimal edge all are that the thought that basis signal is handled is set up, though on marginal information extraction performance, improve to some extent, but, what this method still adopted is the thought of gradient-thresholding, therefore, the problem that exists in the edge extracting method based on classical operator, this method also all exists.

In recent years, development along with small echo (wavelet) technology and fuzzy mathematics, also develop and a kind of wavelet transformation is used for the edge information extracting method of detection of Singular Point and based on the theoretical edge information extracting method of fuzzy set (fuzzy set), also has the edge information extracting method based on methods such as mathematical morphology such as nonlinear filterings.These methods, though the method with respect to classical operator class has certain improvement in some aspects, but the greatest problem of its existence remains the result of this method in subsequent applications, especially with the closely-related subsequent applications of human perception in, visual effect is poor.

Summary of the invention

The objective of the invention is to, a kind of method of digital picture edge extracting is provided, by vision perception characteristic being incorporated the edge extracting process, make edge extracting result of the present invention in subsequent applications, especially with the closely-related subsequent applications of human apperceive characteristic in the time, reach good visual effect, thereby, improved digital picture quality in the video communication.

For achieving the above object, the method for a kind of digital picture edge extracting provided by the invention comprises:

A, determine the gradient image of digital picture;

B, described gradient image is covered processing according to the human visual perception characteristic;

C, according to the marginal information of extracting digital picture through described gradient image after covering processing.

Described step a specifically comprises:

Adopt the edge extracting operator to determine the gradient image of digital picture.

Human visual perception characteristic among the described step b is: brightness shielding effect and/or space complexity shielding effect.

Human visual perception characteristic among the described step b is: the brightness shielding effect;

And described step b specifically comprises:

According to the brightness shielding effect described gradient image being carried out the gradient image that brightness covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+h(b(i,j),b_0)};$

Wherein:

For brightness cover handle the back pixel (i, Grad j),

(i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, and h (b (i, j), b ₀) cover function for brightness, and h (b (i, j), b ₀) satisfy following condition:

If b (i, j)=b ₀, then h (b (and i, j), b ₀)=0, otherwise, h (b (i, j), b ₀)＞0; Simultaneously, and h (b (i, j), b ₀) be | b (i, j)-b ₀| monotonically increasing function.

Human visual perception characteristic among the described step b is: the brightness shielding effect;

And described step b specifically comprises:

According to the brightness shielding effect described gradient image being carried out the gradient image that brightness covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+{\left(\frac{|b(i,j)-b_0|}{b_0}\right)}^{{\mathrm{\&gamma;}}_1}};$

Wherein: For brightness cover handle the back pixel (i, Grad j), (i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, γ ₁Be positive number.

Human visual perception characteristic among the described step b is: brightness shielding effect and space complexity shielding effect;

And in the described step b formula

Replace with: the gradient image among the described step a is carried out space complexity cover back pixel (i, Grad j)

Human visual perception characteristic among the described step b is: the space complexity shielding effect;

And described step b specifically comprises:

According to the space complexity shielding effect described gradient image being carried out the gradient image that space complexity covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\stackrel{\&RightArrow;}{g}(i,j)}{k\left(\frac{m(i,j)}{m_0}\right)};$

Wherein:

For space complexity cover the back pixel (i, Grad j),

(i, ((i j) is (i, the j) space complexity of the regional area at place, the m of pixel in the original image to m for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean space complexity of the whole two field picture of original image,

For space complexity is covered function, and this function satisfies following condition:

K (0)＞0, and

Be Increasing function.

Human visual perception characteristic among the described step b is: the space complexity shielding effect;

And described step b specifically comprises:

According to the space complexity shielding effect described gradient image being carried out the gradient image that space complexity covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{a_0+{\left(\frac{m(i,j)}{m_0}\right)}^{{\mathrm{\&gamma;}}_2}};$

Wherein:

For cover through brightness handle and space complexity cover pixel in the gradient image after the processing (i, Grad j),

Be pixel in the original image (i, Grad j), m ₀Be the mean space complexity of the whole two field picture of original image, (i j) is (i, the j) space complexity of the regional area at place, the γ of pixel in the original image to m ₂Be positive number.

Human visual perception characteristic among the described step b is: brightness shielding effect and space complexity shielding effect;

And among the described step b

Replace with: the gradient image among the described step a is carried out brightness cover processing back pixel (i, Grad j)

Pixel in the original image among the described step b (i, j) the space complexity m of the regional area at place (i, acquisition methods j) specifically comprises:

B1 ', (i j) determines pixel (i, j) two of the edge, place edges according to gradient direction and predetermined gradient mould value in the regional area at place at the pixel of original image;

B2 ', described regional area is divided into a fringe region and two non-fringe region R1, R2 according to described two edges;

B3 ', determine the space complexity m of described two non-fringe region R1, R2 ₁(i, j) and m ₂(i, j);

B4 ', with the space complexity m of described two non-fringe regions ₁(i, j) and m ₂(i, j) average

B3 ', determine the space complexity m of described two non-fringe region R1, R2 ₁(i, j) and m ₂(i, j);

B4 ', with the space complexity m of described two non-fringe regions ₁(i, j) and m ₂(i, mean value j) be defined as pixel (i, j) the space complexity m of the regional area at place (i, j).

Described step b3 ' specifically comprises: the space complexity m of described non-fringe region _p(i j) is:

$m_p(i,j)={\left(\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{|f(s,t)-\stackrel{\&OverBar;}{f_p}|}^k\right)}^{\frac{1}{k}}$ p＝1，2；

Wherein: M is R _pThe number of interior pixel point,

Be R _pThe average brightness in zone, p=1,2; (s is that (s t) locates the original image gray scale to pixel, and Rp is non-fringe region R1, R2, and k is an index t) to f.

Described step b3 ' specifically comprises: the space complexity m of described non-fringe region _p(i j) is:

$m_p(i,j)=\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}|f(s,t)-\stackrel{\&OverBar;}{f_p}|$ P=1,2; Or

$m_p(i,j)=\sqrt{\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{(f(s,t)-\stackrel{\&OverBar;}{f_p})}^2}$ P=1,2; Or

$m_p(i,j)=\underset{(s,t)\&Element;R}{\max }|f(s,t)-\stackrel{\&OverBar;}{f_p}|$ p＝1，2；

Wherein: M is R _pAnd p=1, the number of 2 interior pixel points,

Be R _pThe average brightness in zone, Rp is non-fringe region R1, R2; (s is that (s t) locates the original image gray scale to pixel t) to f.

Described step c specifically comprises:

According to the described gradient image after the processing, the marginal information that the edge extracting operator is determined described digital picture covered.

Described edge extracting operator is: classics extract operator or extract operator based on optimal edge.

Description by technique scheme as can be known, the present invention is by introducing vision perception characteristic in digital picture edge extracting process, the edge extracting result who makes digital picture more meets the purpose of human visually-perceptible picture element amount with human consistent for the edge situation height in image viewing, the understanding from digital picture edge extracting effect of the present invention.

Description of drawings

Fig. 1 is the impulse response function matrix synoptic diagram of 5 * 5 discrete two-dimensional Gaussian filter;

Fig. 2 is that synoptic diagram is covered in brightness of the present invention;

Fig. 3 is that complexity of the present invention is covered synoptic diagram;

(a) among Fig. 4 is original image one;

(b) among Fig. 4 is the edge image synoptic diagram that extracts based on the marginal information of Sobel operator;

The edge image synoptic diagram that (c) among Fig. 4 extracts in conjunction with the marginal information of Sobel operator for the present invention;

(a) among Fig. 5 is original image two;

(b) among Fig. 5 is the edge image synoptic diagram that extracts based on the marginal information of Canny operator;

The edge image synoptic diagram that (c) among Fig. 5 extracts in conjunction with the marginal information of Canny operator for the present invention.

Embodiment

Because the result behind the digital picture edge extracting is in subsequent applications, with the mankind's vision perception characteristic very confidential relation is arranged, so, if in digital image edge extraction method, take into full account human apperceive characteristic, just can make the edge extracting result in subsequent applications, especially with the closely-related subsequent applications of human visual perception in, the very big visual effect of improving video digital images of degree.

Therefore, core of the present invention is: determine the gradient image of digital picture, according to vision perception characteristic the gradient image of described digital picture is covered processing, extract the marginal information of described digital picture according to the described gradient image of covering after the processing.

Based on core concept of the present invention technical scheme provided by the invention is further described below.

Human vision mainly depends on relative light intensity to the reaction of excitation, i.e. contrast, rather than depend on absolute monochrome information.In the appreciable range of light intensity of vision, vision becomes a kind of nonlinear inverse relation to the perception that brightness in the image changes with background luminance, and this vision perception characteristic is called the brightness shielding effect.

The observability at the edge of digital picture not only is subjected to the influence of zone leveling brightness, but also it is relevant with the spatial texture complexity of its region, promptly for the more level and smooth zone of both sides of edges, the vision observability at edge is strong, otherwise, both sides of edges image-region more complicated, details is more, then the vision poor visibility at edge.This vision perception characteristic is called the space complexity shielding effect.

The present invention is directed to the characteristics and the needs of digital picture edge extracting, fully in conjunction with brightness shielding effect in the vision perception characteristic and space complexity shielding effect, gradient image done cover processing, promptly the Grad of video digital images is carried out brightness and cover with space complexity and cover, brightness shielding effect, complexity shielding effect are played an important role in the marginal information leaching process.Thereby, make marginal information of the present invention extract the result and have higher consistance with human vision understanding to image, it is good to make marginal information extract accuracy.

Technical scheme provided by the invention can be in conjunction with the whole bag of tricks of present widely used gradient-thresholding class, be that the present invention can be in the gradient-thresholding edge extracting class methods of routine, processing is covered in fusion brightness and space complexity is covered processing links, make way of realization of the present invention varied, have wide range of applications.

Realization principle of the present invention mainly comprises following four steps:

Step 1, at first obtains original image f (i, gradient image j) according to the edge extracting algorithm based on gradient-thresholding thought of certain routine at present Above-mentioned routine based on the edge extracting algorithm of gradient-thresholding thought can for any based on classical operator the edge extracting algorithm or based on the edge extracting algorithm of optimum operator, also can comprise the edge extracting algorithm of gradient image for any intermediate result.

Step 2, according to the gradient image of human visual perception characteristic to obtaining in the step 1

Carry out brightness and cover processing.

Step 3, according to the human visual perception characteristic gradient image after the processing is covered in the brightness that obtains in the step 2 and carried out space complexity and cover processing.

Step 4, cover gradient image after the processing through space complexity, determine the image border according to the edge locating rule of the edge extracting algorithm of selecting in the step 1 to what obtain in the step 3.If promptly the edge extracting algorithm of selecting in step 1 that is based on classical operator then needs to unify thresholding according to the overall situation and judges the image border; If the edge extracting algorithm of selecting in step 1 that is based on the Canny operator then needs the thought according to the Canny operator, adopt two thresholdings of height to judge the image border; Equally, if what select in step 1 is other edge extracting algorithms, then need to judge edge of image in conjunction with the technical scheme of this edge extracting algorithm.

Certainly, in above-mentioned realization principle, can there be step 3, promptly only the gradient image in the step 1 is carried out brightness and cover processing, like this, in step 4, cover gradient image after the processing to what obtain in the step 2 through brightness, determine the image border according to the edge locating rule of the edge extracting algorithm of selecting in the step 1; Equally, in the above-mentioned realization principle, can not have step 2 yet, promptly only the gradient image in the step 1 is carried out space complexity and cover processing; In addition, the order of step 3 and step 2 also can swap round, promptly earlier the gradient image in the step 1 is carried out space complexity and cover processing, then, carry out brightness and cover processing cover gradient image after the processing through space complexity again, like this, in step 4, to earlier through space complexity cover processings, after cover the gradient image of processing through brightness, determine the image border according to the edge locating rule of the edge extracting algorithm of selection in the step 1.

Below to the present invention at first to original image f (i, gradient image j) Carry out brightness and cover processing, then, again gradient image after the processing is covered in brightness and carried out space complexity and cover the method for processing and be described in detail.

In conjunction with the adaptive ability of vision system to background luminance, the present invention to the gradient image that obtains based on present predetermined edge extraction algorithm carry out each pixel in the gradient image after processing is covered in brightness (i, Grad j) can pass through formula (10) acquisition:

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+h(b(i,j),b_0)};---\left(10\right)$

Wherein:

For brightness cover handle the back pixel (i, Grad j),

(i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, and h (b (i, j), b ₀) cover function for brightness, and h (b (i, j), b ₀) satisfy following condition:

Function h (b (i, j), b ₀) be one about | b (i, j)-b ₀| increasing function, and:

H (b (i, j), b ₀)=0 when b (i, j)=b ₀

H (b (i, j), b ₀)＞0 is in other any situations.(11)

A specific embodiment of formula (10) is as shown in Equation (12):

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+{\left(\frac{|b(i,j)-b_0|}{b_0}\right)}^{{\mathrm{\&gamma;}}_1}}---\left(12\right)$

Wherein:

For brightness cover handle the back pixel (i, Grad j), (i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, γ ₁Be positive number.

(i, j) regional area at place as shown in Figure 2 for pixel.Fig. 2 is that (i j) is the center, and size is the zone of N * N, and wherein, N is an odd number with pixel.

Pixel among Fig. 2 (i, j) the mean flow rate b of region (i, j) can pass through formula (13) and obtain:

$b(i,j)=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="C20051007706100241.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{(p,q)\&Element;B}{\mathrm{\&Sigma;}}f(p,q)---\left(13\right)$

Wherein: p, q are the pixel value in N * n-quadrant.

When technical scheme of the present invention is that advanced row space complexity is covered processing, carry out brightness again and cover when handling, in above-mentioned formula (10) and the formula (12)

Should replace with

Promptly (i, the gradient image based on pre-definite operator j) carry out space complexity and cover back pixel (i, Grad j) to original image f

At this moment, variable being changed to of formula (10): $\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|}{1+h(b(i,j),b_0)};$ Variable being changed to of formula (12): $\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|}{1+{\left(\frac{b(i,j)-b_0}{b_0}\right)}^{{\mathrm{\&gamma;}}_1}}.$

The present invention needs that also gradient image after the processing is covered in brightness and carries out space complexity and cover processing after carrying out brightness and covering processing.

The present invention to brightness cover gradient image after the processing carry out each pixel in the gradient image after space complexity is covered processing (i, Grad j) can pass through formula (14) and obtain:

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|}{k\left(\frac{m(i,j)}{m_0}\right)}---\left(14\right)$

Wherein:

For cover back pixel (i, Grad j), m through brightness ₀Be the mean space complexity of the whole two field picture of original image, m (i, j) be pixel in the original image (i, the j) space complexity of the regional area at place,

For space complexity is covered function, and this function should satisfy following condition:

k(0)＞0 (15)

And

Be

Increasing function.

A specific embodiment of formula (14) is as shown in Equation (16):

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|}{a_0+{\left(\frac{m(i,j)}{m_0}\right)}^{{\mathrm{\&gamma;}}_2}}---\left(16\right)$

Wherein:

For cover through brightness handle and space complexity cover pixel in the gradient image after the processing (i, Grad j),

For cover back pixel (i, Grad j), m through brightness ₀Be the mean space complexity of the whole two field picture of original image, (i j) is (i, the j) space complexity of the regional area at place, the γ of pixel in the original image to m ₂Be positive number.

(i, j) regional area at place as shown in Figure 3 for pixel in the original image.Fig. 3 is that (i j) is the center, and size is the zone of N * N, and wherein, N is an odd number with pixel.

As can be seen from Figure 3, (i is that the N * n-quadrant at center is made up of three parts j), i.e. the subregion R1 and the R2 of edge, both sides of edges with pixel.The space complexity of setting subregion R1 and R2 is respectively: m ₁(x, y), m ₂(x, y), then pixel among Fig. 3 (i, j) the space complexity m of the regional area at place (i, j) can pass through formula (17) and obtain:

$m=(i,j)=\frac{m_1(i,j)+m_2(i,j)}{2}---\left(17\right)$

The space complexity m of above-mentioned both sides of edges subregion _i(i j) may be defined as the function of original image gray scale and this zone leveling gray scale in this zone, and its general form can embody by formula (18):

$m_p(i,j)={\left(\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{|f(s,t)-\stackrel{\&OverBar;}{f_p}|}^k\right)}^{\frac{1}{k}}$ (p＝1，2) (18)

Wherein: M is R _pThe number of interior pixel point,

Be R _pThe average brightness in zone, (s is that (s t) locates the original image gray scale to pixel t) to f.

Obtain the space complexity m of both sides of edges subregion _i(i, three specific embodiments j) are shown in formula (19), (20), (21):

$m_p(i,j)=\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}|f(s,t)-\stackrel{\&OverBar;}{f_p}|$ (p＝1，2) (19)

$m_p(i,j)=\sqrt{\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{(f(s,t)-\stackrel{\&OverBar;}{f_p})}^2}$ (p＝1，2) (20)

$m_p(i,j)=\underset{(s,t)\&Element;R}{\max }|f(s,t)-\stackrel{\&OverBar;}{f_p}|$ (p＝1，2) (21)

Wherein: M is R _pThe number of (p=1,2) interior pixel point,

Be R _pThe average brightness in zone.

The scope of determining both sides of edges subregion R1 and R2 is as follows:

Usually there is certain width at the edge, therefore, is carrying out before space complexity covers, and at first should determine the scope at edge, i.e. the width at edge, thereby, also just determined R1, the scope of R2.

Set pixel (i, j) distance apart from its two edges, edge, place is respectively E1, E2, the method for calculating E1, E2 is: from pixel (i j) sets out, along the direction search of gradient, when the mould value of gradient drops to λ | grad _b(i, j) | the time, then think at this moment, just to have obtained E1 in an edge that has arrived the edge; Then, from pixel (i j) sets out, and promptly becomes 180 ° direction search with gradient along the reverse direction of gradient, when Grad drops to λ | grad _b(i, j) | the time, then think at this moment, just to have obtained E2 in another edge that has arrived the edge.The width at edge has just been determined in E1 and E2 addition.Above-mentioned λ is horizontal controlling elements.

Because the zone of N * N is a neighborhood (neighbourhood) among a small circle, therefore, can think roughly that the width at edge is constant in this zone.Horizontal controlling elements λ can adjust in concrete implementation procedure, to reach better effect.

In technical scheme of the present invention is that advanced row space complexity is covered processing, carry out brightness again and cover when handling, in above-mentioned formula (14) and the formula (16)

Replace with:

Promptly earlier to original image f (i, the gradient image based on pre-definite operator j) carries out space complexity and covers processing, at this moment, formula (14) is replaceable to be: $\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{k{\left(\frac{m(i,j)}{m_0}\right)}^{{\mathrm{\&gamma;}}_2}};$ Formula (16) is replaceable to be: $\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{a_0+{\left(\frac{m(i,j)}{m_0}\right)}^{{\mathrm{\&gamma;}}_2}}.$

Description by technique scheme as can be known, edge extracting method of the present invention is the marginal information of orientation image accurately, like this, in the video communication process, by marginal information is carried out enhancement process, can reach the details sharpening that makes image, make image have sharpening (sharp) effect, thereby, improved the video communication competitiveness of product in market.

The present invention can also become and H.265 waits in emerging video coding technique and the international standard, carries out the basic technology of video compression coding pre-treatment based on object.By technical scheme of the present invention is applied in the object extraction technology, then can improve international standard effect in actual applications, thereby, improve the competitive power of emerging video product.

The concrete experiment effect that adopts edge extracting method of the present invention is shown in accompanying drawing 4 and accompanying drawing 5.

(a) figure among Fig. 4 is the original image of basket, (b) figure among Fig. 4 is the edge image that the edge information extracting method based on the Sobel operator extracts, the edge image that (c) among Fig. 4 extracts in conjunction with the edge information extracting method of Sobel operator for the present invention.

(a) figure among Fig. 5 is the original image of basket, (b) figure among Fig. 5 is the edge image that the edge information extracting method based on the Canny operator extracts, the edge image that to be the present invention extract in conjunction with the edge information extracting method of Canny operator of (c) figure among Fig. 5.

From the experimental result shown in Fig. 4 and Fig. 5 as can be seen, in the prior art result, have much should not be that the thing at edge has become the edge in human intelligible, such as the meadow around the basket etc., the human in fact edge of paying close attention to when observing this image should be the texture edge on the basket, and prior art is carrying out having missed important marginal information when marginal information is extracted.In the result of the present invention, more focus on the texture edge on the basket, not only extract the texture marginal information on a large amount of baskets accurately, and, reduced extraction in a large number to part edge information in meadow in the image, thereby, make the marginal information of extracting more near human visual characteristic.

Though described the present invention by embodiment, those of ordinary skills know, the present invention has many distortion and variation and do not break away from spirit of the present invention, and the claim of application documents of the present invention comprises these distortion and variation.

Claims (13)

1, a kind of digital image edge information extracting method is characterized in that, comprises step:

A, employing edge extracting operator are determined the gradient image of digital picture;

B, described gradient image is covered processing according to the human visual perception characteristic;

C, according to the marginal information of extracting digital picture through described gradient image after covering processing.

2, a kind of digital image edge information extracting method as claimed in claim 1 is characterized in that, the human visual perception characteristic among the described step b is: brightness shielding effect and/or space complexity shielding effect.

3, a kind of digital image edge information extracting method as claimed in claim 2 is characterized in that, the human visual perception characteristic among the described step b is: the brightness shielding effect;

And described step b specifically comprises step:

According to the brightness shielding effect described gradient image being carried out the gradient image that brightness covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+h(b(i,j),b_0)};$

Wherein: For brightness cover handle the back pixel (i, Grad j), (i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, and h (b (i, j), b ₀) cover function for brightness, and h (b (i, j), b ₀) satisfy following condition:

If b (i, j)=b ₀, then h (b (and i, j), b ₀)=0, otherwise, h (b (i, j), b ₀)＞0; Simultaneously, and h (b (i, j), b ₀) be | b (i, j)-b ₀| monotonically increasing function.

4, a kind of digital image edge information extracting method as claimed in claim 3 is characterized in that, the human visual perception characteristic among the described step b is: the brightness shielding effect;

And described step b specifically comprises:

According to the brightness shielding effect described gradient image being carried out the gradient image that brightness covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_b(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{1+{\left(\frac{|b(i,j)-b_0|}{b_0}\right)}^{r_1}};$

Wherein:

For brightness cover handle the back pixel (i, Grad j),

(i, ((i j) is (i, the j) mean flow rate of the regional area at place, the b of pixel in the original image to b for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean flow rate of the whole two field picture of original image, γ ₁Be positive number.

5, as claim 3 or 4 described a kind of digital image edge information extracting methods, it is characterized in that the human visual perception characteristic among the described step b is: brightness shielding effect and space complexity shielding effect;

And in the described step b formula

Replace with: the gradient image among the described step a is carried out space complexity cover back pixel (i, Grad j)

6, a kind of digital image edge information extracting method as claimed in claim 2 is characterized in that, the human visual perception characteristic among the described step b is: the space complexity shielding effect;

And described step b specifically comprises:

According to the space complexity shielding effect described gradient image being carried out the gradient image that space complexity covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\stackrel{\&RightArrow;}{g}(i,j)}{k\left(\frac{m(i,j)}{m_0}\right)};$

Wherein:

For space complexity cover the back pixel (i, Grad j),

(i, ((i j) is (i, the j) space complexity of the regional area at place, the m of pixel in the original image to m for i, Grad j) j) to locate gradation of image f for pixel in the original image ₀Be the mean space complexity of the whole two field picture of original image,

For space complexity is covered function, and this function satisfies following condition:

K (0)＞0, and

Be

Increasing function.

7, a kind of digital image edge information extracting method as claimed in claim 6 is characterized in that, the human visual perception characteristic among the described step b is: the space complexity shielding effect;

And described step b specifically comprises:

According to the space complexity shielding effect described gradient image being carried out the gradient image that space complexity covers after the processing is:

$\left|{\stackrel{\&RightArrow;}{g}}_m(i,j)\right|=\frac{\left|\stackrel{\&RightArrow;}{g}(i,j)\right|}{a_0+{\left(\frac{m(i,j)}{m_0}\right)}^{{\mathrm{\&gamma;}}_2}};$

Wherein:

For cover through brightness handle and space complexity cover pixel in the gradient image after the processing (i, Grad j),

Be pixel in the original image (i, Grad j), m ₀Be the mean space complexity of the whole two field picture of original image, (i j) is (i, the j) space complexity of the regional area at place, the γ of pixel in the original image to m ₂Be positive number.

8, as claim 6 or 7 described a kind of digital image edge information extracting methods, it is characterized in that the human visual perception characteristic among the described step b is: brightness shielding effect and space complexity shielding effect;

And among the described step b

Replace with: the gradient image among the described step a is carried out brightness cover processing back pixel (i, Grad j)

9, as claim 6 or 7 described a kind of digital image edge information extracting methods, it is characterized in that, pixel in the original image among the described step b (i, j) the space complexity m of the regional area at place (i, acquisition methods j) specifically comprises:

B1 ', (i j) determines pixel (i, j) two of the edge, place edges according to gradient direction and predetermined gradient mould value in the regional area at place at the pixel of original image;

B2 ', described regional area is divided into a fringe region and two non-fringe region R1, R2 according to described two edges;

B3 ', determine the space complexity m of described two non-fringe region R1, R2 ₁(i, j) and m ₂(i, j);

B4 ', with the space complexity m of described two non-fringe regions ₁(i, j) and m ₂(i, mean value j) be defined as pixel (i, j) the space complexity m of the regional area at place (i, j).

10, a kind of digital image edge information extracting method as claimed in claim 9 is characterized in that, described step b3 ' specifically comprises:

The space complexity m of described non-fringe region _p(i j) is:

$m_p(i,j)={\left(\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{|f(s,t)-\stackrel{\&OverBar;}{f_p})|}^k\right)}^{\frac{1}{k}},p=\mathrm{1,2};$

Wherein: M is R _pThe number of interior pixel point, f _pBe R _pThe average brightness in zone, p=1,2; (s is that (s t) locates the original image gray scale to pixel, and Rp is non-fringe region R1, R2, and k is an index t) to f.

11, a kind of digital image edge information extracting method as claimed in claim 10 is characterized in that, described step b3 ' specifically comprises:

The space complexity m of described non-fringe region _p(i j) is:

$m_p(i,j)=\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}|f(s,t)-\stackrel{\&OverBar;}{f_p}|,p=\mathrm{1,2};$ Or

$m_p(i,j)=\sqrt{\frac{1}{M}\underset{(s,t)\&Element;R_p}{\mathrm{\&Sigma;}}{(f(s,t)-\stackrel{\&OverBar;}{f_p})}^2},p=\mathrm{1,2};$ Or

$m_p(i,j)=\underset{(s,t)\&Element;R}{\max }|f(s,t)-\stackrel{\&OverBar;}{f_p}|,p=\mathrm{1,2};$

Wherein: M is R _pAnd p=1, the number of 2 interior pixel points, f _pBe R _pThe average brightness in zone, Rp is non-fringe region R1, R2; (s is that (s t) locates the original image gray scale to pixel t) to f.

12, a kind of digital image edge information extracting method as claimed in claim 2 is characterized in that, described step c is specially:

According to the described gradient image after the processing, the marginal information that the edge extracting operator is determined described digital picture covered.

13, as claim 1 or 12 described a kind of digital image edge information extracting methods, it is characterized in that described edge extracting operator is: classics extract operator or extract operator based on optimal edge.

Images