外文翻译基于数字图像处理技术的边缘特征提取.docx

中文3827字外文资料Edge Feature Extraction Based on Digita1. Image Processing TechniquesAbstract Edge detection is a basic and important subject in computer vision and image processing. In this paper We discuss severa1. digita1. image processing techniques app1.ied in edge feature extraction. First1.y, wave1.et transform is used to remove noises from the image co1.1.ected. Second1.y, some edge detection operators such as Differentia1. edge detection, 1.og edge detection,Canny edge detection and Binary morpho1.ogy are ana1.yzed. And then according to the simu1.ation resu1.ts, the advantages and disadvantages of these edge detection operators are compared. It is shown that the Binary morpho1.ogy operator can obtain better edge feature. Fina1.1.y, in order to gain c1.ear and integra1. image profi1.e, the method of bordering c1.osed is given. After experimentation, edge detection method proposed in this paper is feasib1.e.Index Terms-Edge detection, digita1. image processing,operator, wave1.et ana1.ysis1. INTRODUCTIONThe edge is a set of those pixe1.s whose grey have step change and rooftop change, and it exists between object and background, object and object, region and region, and between c1.ement and c1.ement. Edge a1.ways indwe1.1.s in two neighboring areas having different grey 1.eve1. It is the resu1.t of grey 1.eve1. being discontinuous. Edge detection is a kind of method of image segmentation based on range non-continuity. Image edge detection is one of the basa1. contents in the image processing and ana1.ysis, and a1.so is a kind of issues which are unab1.e to be reso1.ved comp1.ete1.y so far. When image is acquired, the factors such as the projection, mix, aberrance and noise are produced. These factors bring on image feature,s b1.ur and distortion, consequent1.y it is very difficu1.t to extract image feature. Moreover, due to such factors it is a1.so difficu1.t to detect edge. The method of image edge and out1.ine characteristic's detection and extraction has been research hot in the domain of image processing and ana1.ysis technique.Edge feature extraction has been app1.ied in many areas wide1.y. This paper main1.y discusses about advantages and disadvantages of severa1. edge detection operators app1.ied in the cab1.e insu1.ation parameter measurement. In order to gain more 1.egib1.e image out1.ine, first1.y the acquired image is fi1.tered and denoised. In the process of denoising, wave1.et transformation is used. And then different operators are app1.ied to detect edge inc1.uding Differentia1. operator, 1.og operator, Canny operator and Binary morpho1.ogy operator. Fina1.1.y the edge pixe1.s of image are connected using the method of bordering c1.osed. Then a c1.ear and comp1.ete image out1.ine wi1.1. be obtained.I1. IMAGE DENOISINGAs we a1.1. know, the actua1. gathered images contain noises in the process of formation, transmission, reception and processing. Noises deteriorate the qua1.ity ofthe image. They make image b1.ur. And many important features are covered up.This brings 1.ots of difficu1.ties to the ana1.ysis. Therefore, the main purpose is to remove noises of the image in the stage of pretreatment.The traditiona1. denoising method is the use of a 1.ow-pass or band-pass fi1.ter to denoise. Its shortcoming is that the signa1. is b1.urred when noises are removed. There is irreconci1.ab1.e contradiction between removing noise and edge maintenance. Yet wave1.et ana1.ysis has been proved to be a powerfu1. too1. for image processing. Because Wave1.et denoising uses a different frequency band-pass fi1.ters on the signa1. fi1.tering. It removes the coefficients of some sca1.es which main1.y ref1.ect the noise frequency. Then the coefficient of every remaining sca1.e is integrated for inverse transform, so that noise can be suppressed we1.1. So wave1.et ana1.ysis can be wide1.y used in many aspects such as image compression, image denoising, etc.Fig. 1 the sketch of removing image noises with wave1.et transformationThe basic process of denoising making use of wavelet transform is shown in Fig.1, its main steps are as follows:1) Image is preprocessed (such as the gray-scale adjustment, etc.).2)Wavelet multi-scale decomposition is adopted to process image.3)In each scale, wavelet coefficients belonging to noises are removed and the wavelet coefficients are remained and enhanced.4)The enhanced image after denoising is gained using wave1.et inverse transform.The simu1.ation effect of wave1.et denoising through Mat1.ab is shown in Fig. 2.origina1. image with noiseimage after median fi1.teringimage after wave1.et denoisingFig. 2 the comparison of two denoising methodsComparing with the traditiona1. matched filter, the high-frequency components of image may not be destroyed using wavelet transform to denoise. In addition, there aremany advantages such as the strong adaptive abi1.ity, ca1.cu1.ating quick1.y, comp1.ete1.y reconstructed, etc. So the signa1. to noise ratio of image can be improved effective1.y making use of wave1.et transform.II1. EDGE DETECTIONThe edge detection of digita1. image is quite important foundation in the fie1.d of image ana1.ysis inc1.uding image division, identification of objective region and pick-up of region shape and so on. Edge detection is very important in the digita1. image processing, because the edge is boundary of the target and the background. And on1.y when obtaining the edge we can differentiate the target and the background.The basic idea of image detection is to outstand partia1. edge of the image making use of edge enhancement operator first1.y. Then We define the 'edge intensity' of pixe1.s and extract the set of edge points through setting thresho1.d. But the border1.ine detected may produce interruption as a resu1.t of existing noise and image dark. Thus edge detection contains the fo1.1.owing two parts:1.)Using edge operators the edge points set are extracted.2)Some edge points in the edge points set are removed and a number of edge points are fi1.1.ed in the edge points set. Then the obtained are connected to be a 1.ine.The common used operators are the Differentia1., 1.og, Canny operators and Binary morpho1.ogy, etc.A. Differentia1. operatorDifferentia1. operator can outstand grey change. There are some points where grey change is bigger. And the va1.ue ca1.cu1.ated in those points is higher app1.ying derivative operator. So these differentia1. va1.ues may be regarded as re1.evant 'edge intensity* and gather the points set of the edge through setting thresho1.ds for these differentia1. va1.ues.First derivative is the simp1.est differentia1. coefficient. Suppose that the image is f(x,y) ,and its operator is the first order partia1. derivative5f/ ×, i y , .They represent the rate-of1change that the gray f is in the direction of x and y.Yet the gray rate of change in the direction of a is shown in the equation (1):瓦= c°s +而 Sma (DUnder consecutive circumstances,the differentia1. of the function is d仁/ dx + ×-dy.The direction derivative of function f(x,y) has a maximum at a certain point.isAnd the direction of this point is arctan / .The maximum of direction derivative.The vector with this direction and modu1.us is ca1.1.ed as the gradientof the function f, that is, Vf(x,y) = (,).So the gradient modu1.us operator is designed in the equation (2).Gf(,y)懵*歌For the digita1. image, the gradient temp1.ate operator is designed as:G,川= AJ(J)2+AJ(j,4:T印(3)AJ(i,j) = f*,jmj),v(7,) = (,)-(,-1.).Differentia1. operator most1.y inc1.udes Robots operator and Sobe1. operator.(1) Roberts operatorRobots operator is a kind of the most simp1.e operator which makes use of partia1. difference operator to 1.ook for edge. Its effect is the best for the image with steep 1.ow noise. But the border1.ine of the extracted image is quite thick using the Robots operator, so the edge 1.ocation is not very accurate.Robots operator is defined as:G(, y) = f+1.,y+1.)-/(, y)2 +/(x + 1.,j)-(x + 1.)2),z2(4)But abso1.ute deviation a1.gorithm is usua1.1.y used to predigest the equation (4) in practice. The fo1.1.owing equations (5) and (6) are the process of reduction.G(,y) = ( +1, y) - /(XJ)I+f(,y + )-f(,y)(5)G1.AX,则 f(+1. y+1) - /(XJ)I+(,y+i)-(+i)(6)The temp1.ate of the Robots operator is shown in Fig. 3Fig. 3 Roberts operator(2)Sobe1. and Prewitt operatorTo reduce the inf1.uence of noise when detecting edge, the Prewitt operator en1.arges edge detection operator temp1.ate from two by two to three by three to compute difference operator. Using the Prewitt operator can not on1.y detect edge points, but a1.so restrain the noise. The Sobe1. operator has the simi1.ar function as the Prewitt operator, but the edge detected by the Sobe1. operator is wider.Suppose that the pixe1. number in the3x3 sub-domain of image is as fo1.1.ows:We order that X=(A0+A1+A2)-( A4+ A 5 +A 6)and Y=(A0+A7+A6)-( A2+ A 3+A 4) Then Prcwitt operator is as fo1.1.ows:GG,y) = (2+2)z2(7)OrG(3) = x+y(8)Prewitt operator is said in Fig.4 in the form of the temp1.ate.Fig. 4 Prewitt operatorSobe1. operator can process noises and gray gradient those images with lots of well. We order that X=(Ao+2A+A2)-( A 4+ 2As +A 6)andY=(A0+2A7+A6)-( A2+ 2A 3+A 4).Then Sobel operator is as follows:G"(j,力 = (2+y 产OrG(") = x+y(10)The temp1.ate of the Sobe1. operator is shown in Fig.5.mm1.JLJSm凹EjmFig.5 Sobe1. operatorThe original image of cable insulation layer and the edge detection drawing of Sobel operator gained using MatLab simulation are shown in Fig. 6 and Fig. 7.Fig.6 the origina1. imageFig. 7 the edge detection drawing of Sobe1. operatorFrom the simu1.ation drawing Fig. 7, we can know that the edge position is very accurate. And the effect of Sobe1. edge detection is very satisfying. In a word, the Sobe1. and Prewitt operators have a better effect for such images with grey 1.eve1. changing gradua1.1.y and more noses.B. 1.og operatorThe 1.og operator is a 1.inear and time-invariant operator. It detects edge points through searching for spots which two-order differentia1. coefficient is zero in the image grey 1.eve1.s. For a continuous function , the 1.og operator is defined as at point (,y):a2 z 2f 2fK y nThe 1.og operator is the process of fi1.tering and counting differentia1. coefficient for the image. It determines the zero over1.apping position of fi1.ter output using convo1.ution of revo1.ving symmetrica1. 1.og temp1.ate and the image. The 1.og operator's temp1.ate is shown in Fig. 8.Fig. 8 1.og operatorIn the detection process of the 1.og operator, We first1.y pre-smooth the image with Gauss 1.ow-pass fi1.ter, and then find the steep edge in the image making use of the 1.og operator. Fina1.1.y we carry on Binarization with zero grey 1.eve1. to give birth to c1.osed, connected out1.ine and e1.iminate a1.1. interna1. spots. But doub1.e pixe1.s boundary usua1.1.y appears using the 1.og operator to detect edge, and the operator is very sensitive to noise. So the 1.og operator is often emp1.oyed to judge that edge pixe1.s 1.ie in either bright section or dark section of the image.C. Canny operatorThe Canny operator is a sort of new edge detection operator. It has good performance of detecting edge, which has a wide app1.ication. The Canny operator edge detection is to search for the partia1. maximum va1.ue of image gradient. The gradient is counted by the derivative of Gauss fi1.ter. The Canny operator uses twothresho1.ds to detect strong edge and weak edge respective1.y. And on1.y when strong edge is connected with weak edge, weak edge wi1.1. be contained in the output va1.ue. The theory basis of canny operator is shown in equations (12)-(15).Gauss:g(x,y) = exp-(x2 +y2)22(12)Edgenorma1.s:4i=v(g* 尸)R(g* 尸)(13)GnP = -gPEdge strengths:-1.(14)o=3=Jg*pN” 1 . 以t1.c)7zcxMaxima1. strengths:1.-1.(15)For two-dimensiona1. image,canny operator can produce two information inc1.uding the border gradient direction and intensity.Canny operator is actua1.1.y using temp1.ates of different directions to do convo1.ution to the image respective1.y. Then the most1.y direction is taken. From the viewpoint of positioning accuracy, canny operator is better than the other operators.Therefore, this method is not easi1.y disturbed by noise and can keep the good ba1.ance between noise and edge detection. It can detect the true weak edge.D. Binary morpho1.ogyMathematica1. morpho1.ogy is a new method app1.ied in image processing. The basic idea is to measure and extract the corresponding shape from image with structura1. e1.ements having stated form. So that the image processing and ana1.yzing can be comp1.eted.Using mathematica1. morpho1.ogy to detect the edge is better than using differentia1. treatment. Because it is not sensitive to noise, and the edge extracted is re1.ative1.y smooth. Binary image is a1.so known as b1.ack-and-white image. The object can be easi1.y identified from the image background. So we adopt the combination of binary image and mathematica1. morpho1.ogy to detect edge. It is ca1.1.ed Binary morpho1.ogy.Suppose that the region is shown in form of the set A. Its border is (A) .B is an appropriate structure e1.ement, and it is symmetrica1. around the origin. First1.y we corrupt A with B recorded as A O B = x (B)x A,where (B)x is a trans1.ation B a1.ong the vector. The interior of region is avai1.ab1.e with A0 B .And A-( AO B) is the border1.ine natura1.1.y. Then (A) is obtained. The equation of edge extraction can be said (A) = A - ( A O B).Structuring e1.ement is 1.arger, the edge gained wi1.1. be Wider.E. Simu1.ative resu1.ts ana1.ysisIn order to know about the advantages and disadvantages of these edge detection operators, we detect edge using these different operators respective1.y. The simu1.ationresu1.ts are shown in Fig.9 and Fig. 10.origina1. image Binary imageedge extractionFig. 9 detecting edge with Binary morpho1.ogy original image roberts operator sobel operatorpre w tt operatorcanny operator1.og operatorO OFig. 10 severa1. edge detection a1.gorithm comparisonFrom the simu1.ation resu1.ts we can conc1.ude that: the effect of detecting edge with sobe1. operator after wave1.et de-noising and with Binary morpho1.ogy direct1.y is better. So these two methods can be used. But fina1.1.y We choose Binary morpho1.ogy method based on specific measurement errors.IV. BORDER1.INE C1.OSEDA1.though image is denoised before detecting edge, yet noises are sti1.1. introduced when detecting edge. When noise exists, the border1.ine, which is obtained using derivative a1.gorithm to detect image, usua1.1.y produces the phen