How can i get clear straight line edge detection ? I have used following image and i have following edge detect image this is not clear ?

How can i get clear straight line edge detection ? I have used following image and i have following edge detect image this is not clear ?

[Walter Roberson]

Please post the original as well.

[Walter Roberson]

Is the original the one below? If so, then you have the problem that the vertical edges are not straight lines.

Answer 1

[harjeet singh]

may be this code will help you, or do i know what type of edges exactly you want?

clear all
close all

image=imread('b.jpg');
figure(1)
imshow(image)
drawnow

image_g=rgb2gray(image);
figure(2)
imshow(image_g)
drawnow

a = (lagmatrix(image_g',1))';
a(isnan(a))=0;
c=double(a)-double(image_g);
c=c>20;
figure(3)
imshow(c)
drawnow

c=bwareaopen(c,10);
figure(4)
imshow(c)
drawnow

[Selva Karna]

Thank you so much sir. but i Need clear straight line edge detection ?

[Walter Roberson]

selva, what do you mean by "clear straight line edge detection"?

[harjeet singh]

selva, first of all try to reduce the white light coming from one direction which is destroying the edges, than you may try to reduce the number of colors using k-means.

[Selva Karna]

Thank you sir .... I will apply ... can i known about k-means ? its a algorithm or inbuilt commend please refer me sir?

[Selva Karna]

Walter Roberson sir . I Need perfect straight line edge detection sir.. can you help me sir?

[Walter Roberson]

What is a "perfect straight line edge detection"? And why would you want to apply it to an image whose lines are curved like the above image?

Answer 2

[harjeet singh]

hello selva till yet i worked with finding the lines which are area of interest, i have to work on theta to tilt than on right angle, that i will do next time till that have a look this code.

clear all
  close all
  clc
  image=imread('b.jpg');

    %//////////////////////////////////////////

    image_g=rgb2gray(image);
    figure(3)
    imshow(image_g)
    drawnow
    a = (lagmatrix(image_g',1))';
    a(isnan(a))=0;
    c=double(a)-double(image_g);
    c=c>20;
    figure(4)
    imshow(c)
    drawnow
    c=bwareaopen(c,30);
    figure(5)
    imshow(c)
    drawnow
    [m1,n1]=size(c);

    for i=150:m1
        [lab,num]=bwlabel(c(i,:));
        a1(i,:)=[i num];
    end
    a1=sortrows(a1,2);
    r_all=a1(end,1);

    %/////////////////////////////////////////////

    [lab,num]=bwlabel(c);
    for i=1:num
        [r,c]=find(lab==i);
        [r1,c1]=find(r==r_all);
        if(length(r1)>0)
            a=length(r);
            mat(i,:)=[i a];
        end
    end
    mat=sortrows(mat,2);
    mat=flipud(mat);

    %///////////////////////////////////////////////

    img_1(1:m1,1:n1)=0;
    img_1=logical(img_1);

    for i=1:length(mat)
                    [r,c]=find(lab==mat(i,1));
                    at=[r c];
                    [r1,c1]=find(r==min(r));
                    num_1=[1 at(r1(1),2)];

                    [r1,c1]=find(r==max(r));
                    num_2=[m1 at(r1(1),2)];

                    %//////////////////////////////////////////////

                        startp=num_1;
                        endp=num_2;

                        pts=2000;
                         m=atan((endp(2)-startp(2))/(endp(1)-startp(1))); %gradient 
    %                     
    %                     x=m1/2+m1*cos(m*3.14/180)
    %                     y=n1/2+m1*sin(m*3.14/180)

                        if (m==Inf || m==-Inf) %vertical line
                            xx(1:pts)=startp(1);
                            yy(1:pts)=linspace(startp(2),endp(2),pts);
                        elseif m==0 %horizontal line
                            xx(1:pts)=linspace(startp(1),endp(1),pts);
                            yy(1:pts)=startp(2);
                        else %if (endp(1)-startp(1))~=0
                            xx=linspace(startp(1),endp(1),pts);
                            yy=m*(xx-startp(1))+startp(2);
                        end

                        xx=round(xx);
                        yy=round(yy);

                        for j=1:length(xx)
                            if(xx(j)>0 && yy(j)>0 && xx(j)<m1 && yy(j)<n1)
                                img_1(xx(j),yy(j))=1;  
                            end
                        end

    figure(7)
    imshow(img_1)
    drawnow
    %/////////////////////////////////////////////////////////////////
    end

[Image Analyst]

I couldn't run it:

To use 'lagmatrix', you might need:
  lagmatrix - Econometrics Toolbox
Error in test (line 13)
a = (lagmatrix(image_g',1))';

It doesn't seem like the Econometrix Toolbox should be required for an image processing app. Can you make it run without that toolbox?

[harjeet singh]

hello sir, use this code without lagmatrix i am still working on the theta to produce exact angle as it is in original image.

clear all
close all
clc
image=imread('b.jpg');

%//////////////////////////////////////////

image_g=rgb2gray(image);
figure(3)
imshow(image_g)
drawnow
dummy(1:size(image_g,1),1)=0;
a=[dummy image_g(:,1:end-1)];

c=double(a)-double(image_g);
c=c>20;
figure(4)
imshow(c)
 drawnow
c=bwareaopen(c,30);
figure(5)
imshow(c)
drawnow
[m1,n1]=size(c);

for i=150:m1
    [lab,num]=bwlabel(c(i,:));
    a1(i,:)=[i num];
end
a1=sortrows(a1,2);
r_all=a1(end,1);

%/////////////////////////////////////////////

[lab,num]=bwlabel(c);
for i=1:num
    [r,c]=find(lab==i);
    [r1,c1]=find(r==r_all);
    if(length(r1)>0)
        a=length(r);
        mat(i,:)=[i a];
    end
end
mat=sortrows(mat,2);
mat=flipud(mat);

%///////////////////////////////////////////////

img_1(1:m1,1:n1)=0;
img_1=logical(img_1);

for i=1:length(mat)
                [r,c]=find(lab==mat(i,1));
                at=[r c];
                [r1,c1]=find(r==min(r));
                num_1=[1 at(r1(1),2)];

                  [r1,c1]=find(r==max(r));
                  num_2=[m1 at(r1(1),2)];

                  %//////////////////////////////////////////////

                      startp=num_1;
                      endp=num_2;

                      pts=2000;
                       m=atan((endp(2)-startp(2))/(endp(1)-startp(1))); %gradient 
  %                     
  %                     x=m1/2+m1*cos(m*3.14/180)
  %                     y=n1/2+m1*sin(m*3.14/180)

                      if (m==Inf || m==-Inf) %vertical line
                          xx(1:pts)=startp(1);
                          yy(1:pts)=linspace(startp(2),endp(2),pts);
                      elseif m==0 %horizontal line
                          xx(1:pts)=linspace(startp(1),endp(1),pts);
                          yy(1:pts)=startp(2);
                      else %if (endp(1)-startp(1))~=0
                          xx=linspace(startp(1),endp(1),pts);
                          yy=m*(xx-startp(1))+startp(2);
                      end

                      xx=round(xx);
                      yy=round(yy);

                      for j=1:length(xx)
                          if(xx(j)>0 && yy(j)>0 && xx(j)<m1 && yy(j)<n1)
                              img_1(xx(j),yy(j))=1;  
                          end
                      end

figure(7)
imshow(img_1)
drawnow
%/////////////////////////////////////////////////////////////////
end

[Selva Karna]

Thank you sir. and thank your spent time for me[GOLD time] but ... some other image did not clear output sir? so i have attached that image ? help me sir?

[Selva Karna]

This code dose not give straight line edge in some one more image ? how improve get good performance ?so help me sir harjeet singh . I have attached that image ?

Answer 3

[Selva Karna]

Thank you sir .... I will apply ... can i known about k-means ? its a algorithm or inbuilt commend please refer me sir?

Answer 4

[Image Analyst]

You might be able to use hough(), houghlines() or RANSAC. Look up ransac in the MATLAB help - I think there's a function estimateGeometricTransform() that uses it.

Answer 5

[Selva Karna]

<<

>> I have attached one more image . this image not give clear straight line corrresponding image object count ,, so how to improve help me sir........

[harjeet singh]

selva please upload atleast 10 images of the same so that i do make changes in the code to run on all images

[Image Analyst]

That image looks absolutely nothing like the one Walter posted above from your duplicate question. This one is virtually all black with just a single line. Personally I wouldn't even bother doing anything with the image you posted here in your so called "Answer". Actually this should not even be here since it's not an "Answer" to your original question posted at the very top of this page. You should have posted it as a comment (perhaps under Walter's comment), or as an edit to your original question.

Also, in your duplicate question, people have suggested improving your image capture situation since it's always better to start with a good image than with a bad one and then try to fix it up. I don't remember you responding to them. Can you improve your images? If not why not? Have you considered contacting a professional machine vision company, or a professional photographer to help you with that? It would make your task unbelievably easier.

And you've asked 72 questions but have only ever accepted one of them. Accepting a question gives the person who took the time to answer you "reputation points" so you might want to thank some of the people who have helped you in the past by accepting their answers. It's kind of the etiquette here in the forum, along with not asking duplicate questions of course.

[Selva Karna]

Fine sir..