CN105957067A - Color image edge detection method based on color difference - Google Patents

Abstract

The invention discloses a color image edge detection method based on color difference and belongs to the field of digital image processing. The method has better robustness to smooth filtering operation, may eliminate a plenty of noise edges and just reserve a real edge reflecting the basic structure of an object. The method comprises steps of: performing smooth filtering de-noising; constituting neighborhood dot pairs: for an image IC (x, y) subjected to smooth filtering, classifying eight pixels adjacent to each pixel (x, y) into four neighborhood dot pairs according to the horizontal direction, the vertical direction, the 45-degree direction, and the 135-degree direction; computing a color deviation result; and computing a color difference map to obtain a color difference graph CDM (x, y) which is a final edge detection result. The method may suppress noise interference, eliminates a plenty of noise edges in which users are not interested, just keeps the real edge reflecting the basic structure of the object. The obtained edge has good robustness to smooth filtering. The method still has good edge responses when a filtering window is large in size and is not liable to loss and fracture of important edges.

Description

A kind of color image edge detection method based on color difference

Technical field

The present invention relates to digital image processing field, particularly relate to a kind of color images edge detection based on color difference Method.

Background technology

Rim detection is the basic fundamental in image processing field, is widely used in image retrieval, target detection, image In the various fields such as segmentation, image classification.The point undergone mutation in image generally corresponds to some important object in image or Person's attribute change, mainly includes discontinuous, the profile of target object in image depth and scene illumination condition change etc..Edge The purpose of detection is the point undergone mutation in detection image, by processing these catastrophe points, original image can be greatly decreased Data volume, and reject the usual unconcerned slowly varying part of people, retain the structure attribute that image is important.

Traditional rim detection generally can be by calculating maximum or the zero-crossing examination of second order gradient of First-order Gradient Realize.Edge detection method based on First-order Gradient maximum mainly has Robert operator, Prewitt operator, Sobel operator Deng.Edge detection method based on second order gradient zero-crossing examination mainly has Laplacian operator, LoG operator, Canny operator Deng.But, traditional edge detection method is mainly directed towards gray level image.When processing coloured image, generally require first by colour Image is converted into gray level image, then carries out rim detection.Coloured image only remains monochrome information after gray processing processes, and neglects Omit the information of each color channel.Therefore, the rim detection problem of coloured image receives the extensive concern of people.

The patent of Application No. CN201110448119.1 discloses a kind of color image edge detection method.The method is first First coloured image is decomposed into redness, green, blueness and 4 channel image of yellow, calculates red yellow antagonism figure green, blue the most respectively Picture, and on antagonism image, it is calculated marginal information distributed image, finally edge is carried out micronization processes.

The patent of Application No. CN201310733556.7 discloses a kind of coloured image selected based on multi-channel information Edge detection method.The method carries out rim detection in RGB color respectively and believes with gradient tri-Color Channels of R, G, B Breath extracts, and the edge detection results then in conjunction with each passage selects the marginal point that confidence level is high, obtains edge detection results.

Although existing color image edge detection method is varied, but in general, existing Color Image Edge Technology yet suffers from following two aspect problems:

(1) in order to reduce noise jamming, before rim detection, smoothing filtering operation, existing rim detection are generally carried out Method is easily affected by smothing filtering, edge Loss often occurs；

(2) existing edge detection method is the most sensitive to noise, easily detects a large amount of people uninterested noise limit Edge, interferes subsequent treatment.

Summary of the invention

For the problem in terms of solution above-mentioned two, the present invention proposes the inspection of a kind of Color Image Edge based on color difference Survey method.The method has higher robustness to smoothing filtering operation, can get rid of substantial amounts of noise edge, only retain reflection The true edge of object basic structure.

For achieving the above object, the present invention is by the following technical solutions:

The flow chart of the color image edge detection method based on color difference that the present invention proposes is as it is shown in figure 1, concrete Comprise the following steps:

Step 1, smothing filtering denoising.

Input color image is carried out smothing filtering, suppresses noise jamming.This process is expressed from the next:

$I^C(x,y)=\underset{\mathrm{\Ω}\left(n\right)}{Smooth}\[H^C(x,y)\]---\left(1\right)$

In formula, H^C(x, y) is input color image, the Color Channel of C representative image, and C={R, G, B}, R, G and B are respectively Represent three kinds of colors of RGB.Represent with the Ω (n) smothing filtering as window size, the kind bag of described smothing filtering Include mean filter, gaussian filtering, medium filtering；I^C(x y) is the image after smothing filtering.

Step 2, constitutes neighborhood point pair.

For the image I after smothing filtering^C(x, y), by each pixel, (x, eight y) adjacent pixels are respectively according to water Flat, vertically, 45 ° and 135 ° of total of four directions be divided into four neighborhood points pair.(x-1, y) with (x to being respectively for these four neighborhood points + 1, y), (x, y-1) and (x, y+1), (x-1, y+1) and (x+1, y-1), (x-1, y-1) and (x+1, y+1).Pixel (x, y) with The position corresponding relation of its adjacent eight territory pixels is as shown in Figure 2.

Step 3, calculates color from result.

(x, the color of 4 neighborhood points pair y) is from result, the color of these four neighborhood points pair to calculate each pixel respectively Carry out from computing respectively, be specifically expressed as follows:

I^C(x-1,y)ΘI^C(x+1,y) (2)

I^C(x,y-1)ΘI^C(x,y+1) (3)

I^C(x-1,y+1)ΘI^C(x+1,y-1) (4)

I^C(x-1,y-1)ΘI^C(x+1,y+1) (5)

Wherein, Θ represents that color is a kind of binocular computing from computing, color from computing, is respectively (x for coordinate₁, y₁) and (x₂,y₂) two pixels, its color is as follows from operation definition:

I^C(x₁,y₁)ΘI^C(x₂,y₂)=DR ∪ DG ∪ DB (6)

In formula, DR, DG, DB are respectively

$DR=\{\begin{array}{cc}1,& I^R(x_1,y_1)-I^R(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^R(x_1,y_1)-I^R(x_2,y_2)<TH\end{array}---\left(7\right)$

$DG=\left\{\begin{array}{cc}1,& I^G(x_1,y_1)-I^G(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^G(x_1,y_1)-I^G(x_2,y_2)<TH\end{array}\right.---\left(8\right)$

$DB=\left\{\begin{array}{cc}1,& I^B(x_1,y_1)-I^B(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^B(x_1,y_1)-I^B(x_2,y_2)<TH\end{array}\right.---\left(9\right)$

Wherein, I^R(x,y)、I^G(x, y) and I^B(x, y) represents the pixel value of tri-Color Channels of R, G, B respectively, and TH is threshold Value.

Step 4, calculates colour-difference component.

According to each pixel (x, the color of four neighborhood points pair y) from result, calculate colour-difference component CDM (x, y), Computational methods are as follows:

$CDM(x,y)=\left\{\begin{array}{cc}255,& CD=1\\ 0,& CD=0\end{array}\right.---\left(10\right)$

In formula, the computing formula of CD is

$\begin{array}{c}CD=I^C(x-1,y-1){\mathrm{\&Theta;I}}^C(x+1,y+1)\&cup;I^C(x+1,y-1){\mathrm{\&Theta;I}}^C(x-1,y+1)\\ \&cup;I^C(x-1,y){\mathrm{\&Theta;I}}^C(x+1,y)\&cup;I^C(x,y-1){\mathrm{\&Theta;I}}^C(x,y+1)\end{array}---\left(11\right)$

In formula, color is higher than or computing " ∪ " from the priority of computing " Θ ".

(x y) is final edge detection results to calculated colour-difference component CDM.

Ω (n) needs to take 3x3,5x5,7x7 equidimension according to the noise level of input picture.

Compared with prior art, the present invention has following obvious advantage and a beneficial effect:

1. can effectively suppress noise jamming, get rid of the uninterested noise edge of a large amount of people, only retain reflection object The true edge of basic structure change；

2. the edge obtained has preferable robustness for smothing filtering, remains able to when filter window size is bigger There is stronger skirt response, do not easily cause important edges and lose and fracture.

Accompanying drawing explanation

Fig. 1 is the flow chart of color image edge detection method involved in the present invention；

Fig. 2 is that (x, y) with the position relationship of its 8-neighborhood territory pixel for pixel；

Fig. 3 is the testing result contrast of edge detection method involved in the present invention and existing edge detection method, and (a) is Input color image in embodiment, (b) is the testing result of Sobel operator, and (c) is the testing result of Laplacian operator, D () is the testing result of Canny operator, (e) is the testing result of VG operator, and (f) is the inspection of the method using the present invention to propose Survey result.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the present invention will be further described.

S1 smothing filtering denoising

First, the coloured image of input is carried out mean filter, suppresses noise jamming.In the present embodiment, smothing filtering Kind elects mean filter as, and window size Ω (n) of filtering is taken as 5x5.

S2 constitutes neighborhood point pair

By pixel each in the image after mean filter (x, adjacent 8 pixels y) respectively according to level, vertically, 45 ° and 135 ° altogether 4 directions constitute 4 neighborhood points pair.These 4 neighborhood points to be respectively (x-1, y) with (x+1, y), (x, y-1) with (x, y+1), (x-1, y+1) and (x+1, y-1), (x-1, y-1) and (x+1, y+1).

S3 calculates color from result

(x, the color of 4 neighborhood points pair y) is from result to calculate each pixel respectively according to formula (2)-(4).This reality Executing in example, color threshold value TH in computing takes 30.

S4 calculates colour-difference component

According to each pixel, (x, the color of four neighborhood points pair y), from result, calculates colour-difference according to formula (10) Component, obtains final edge detection results.

Shown in Fig. 3 is the testing result pair of edge detection method involved in the present invention and existing edge detection method Ratio, (a) is the input color image in embodiment, and (b) is the testing result of Sobel operator, and (c) is Laplacian operator Testing result, (d) is the testing result of Canny operator, and (e) is the testing result of VG operator, and (f) proposes for using the present invention The testing result of method.As it is shown on figure 3, edge detection method involved in the present invention is compared with existing edge detection method, can To obtain structural integrity, sharp-edged testing result while effectively suppression noise.

The content that this specification embodiment describes is only the illustration of the way of realization to inventive concept, the present invention's Protection domain is not construed as the concrete form being only limitted to described in embodiment.Protection scope of the present invention is also and in this area skill Art personnel according to present inventive concept it is conceivable that equivalent technologies means.

Claims (2)

1. a color image edge detection method based on color difference, it is characterised in that: the method specifically includes following step Rapid:,

Step 1, smothing filtering denoising；

Input color image is carried out smothing filtering, suppresses noise jamming；This process is expressed from the next:

$I^C(x,y)=\underset{\mathrm{\&Omega;}\left(n\right)}{Smooth}\&lsqb;H^C(x,y)\&rsqb;---\left(1\right)$

In formula, H^C(x, y) is input color image, the Color Channel of C representative image, and C={R, G, B}, R, G and B represent red respectively Green blue three-color；Representing with the Ω (n) smothing filtering as window size, the kind of described smothing filtering includes all Value filtering, gaussian filtering, medium filtering；I^C(x y) is the image after smothing filtering；

Step 2, constitutes neighborhood point pair；

For the image I after smothing filtering^C(x, y), by each pixel, (x, eight y) adjacent pixels are respectively according to level, perpendicular Directly, 45 ° and 135 ° of total of four directions are divided into four neighborhood points pair；These four neighborhood points to be respectively (x-1, y) with (x+1, Y), (x, y-1) and (x, y+1), (x-1, y+1) and (x+1, y-1), (x-1, y-1) and (x+1, y+1)；Obtain pixel (x, y) with The position corresponding relation of its adjacent eight territory pixels；

Step 3, calculates color from result；

(x, the color of four neighborhood points pair y) is from result, the color minute of these four neighborhood points pair to calculate each pixel respectively Do not carry out from computing, be specifically expressed as follows:

I^C(x-1,y)ΘI^C(x+1,y) (2)

I^C(x,y-1)ΘI^C(x,y+1) (3)

I^C(x-1,y+1)ΘI^C(x+1,y-1) (4)

I^C(x-1,y-1)ΘI^C(x+1,y+1) (5)

Wherein, Θ represents that color is a kind of binocular computing from computing, color from computing, is respectively (x for coordinate₁,y₁) and (x₂,y₂) two pixels, its color is as follows from operation definition:

I^C(x₁,y₁)ΘI^C(x₂,y₂)=DR ∪ DG ∪ DB (6)

In formula, DR, DG, DB are respectively

$DR=\left\{\begin{array}{cc}1,& I^R(x_1,y_1)-I^R(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^R(x_1,y_1)-I^R(x_2,y_2)<TH\end{array}\right.---\left(7\right)$

$DG=\left\{\begin{array}{cc}1,& I^G(x_1,y_1)-I^G(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^G(x_1,y_1)-I^G(x_2,y_2)<TH\end{array}\right.---\left(8\right)$

$DB=\left\{\begin{array}{cc}1,& I^B(x_1,y_1)-I^B(x_2,y_2)\&GreaterEqual;TH\\ 0,& I^B(x_1,y_1)-I^B(x_2,y_2)<TH\end{array}\right.---\left(9\right)$

Wherein, I^R(x,y)、I^G(x, y) and I^B(x, y) represents the pixel value of tri-Color Channels of R, G, B respectively, and TH is threshold value；

Step 4, calculates colour-difference component；

According to each pixel, (x, the color of four neighborhood points pair y) is from result, and (x y), calculates to calculate colour-difference component CDM Method is as follows:

$CDM(x,y)=\left\{\begin{array}{cc}255,& CD=1\\ 0,& CD=0\end{array}\right.---\left(10\right)$

In formula, the computing formula of CD is

$\begin{array}{c}CD=I^C(x-1,y-1){\mathrm{\&Theta;I}}^C(x+1,y+1)\&cup;I^C(x+1,y-1){\mathrm{\&Theta;I}}^C(x-1,y+1)\\ \&cup;I^C(x-1,y){\mathrm{\&Theta;I}}^C(x+1,y)\&cup;I^C(x,y-1){\mathrm{\&Theta;I}}^C(x,y+1)\end{array}---\left(11\right)$

In formula, color is higher than or computing " ∪ " from the priority of computing " Θ "；

(x y) is final edge detection results to calculated colour-difference component CDM.

2. according to a kind of color image edge detection method based on color difference of claim, it is characterised in that: Ω (n) needs Noise level according to input picture takes 3x3,5x5,7x7 size.