CODEWITHSUNDEEP
matlabimage-processingcomputer-visionedge-detection
lakshmen
lakshmen

Reputation: 29094

Canny edge detector in MATLAB

I am trying to perform canny edge detector without calling the canny function in Matlab. I wrote a few functions for the gaussian filter(sigma = 1) and non-maximum suppression. The original image and the resultant image is shown.. Not sure what the error...

The original image is

enter image description here

The output i get is

enter image description here

I have attached the code :

%% Read in
I = imread('fruit.jpg');
figure(1),imshow(I)
I = double(I);
%% Determine Mask Size
sigma = 2;
w = mask_size(sigma);
%% Gaussian Smoothing Filter
[ G,sum ] = gauss_mask(w,sigma);
%% Convolve
I1 = (1/sum) * image_convolution(I,w,G);
figure(2),imshow(I1);
%% Ix(derivative in x-direction)
Ix= delx(I1);
figure(3),imshow(Ix);
%% Iy(derivative in y-direction)
Iy= dely(I1);
figure(4),imshow(Iy);
%% Gradient Magnitude
If = grad_mag(Ix,Iy);
figure(5),imshow(If);
%% Non-maxmimum suppression
It = suppression(If,abs(Ix),abs(Iy));
figure(6),imshow(It);



function [ G,sum ] = gauss_mask( w,sigma )
min = 1;
m = floor(w/2);
sum = 0;
for x = 1: w
    for y = 1:w
        g = x-m-1;
        h = y-m-1;
        k = -(g^2 +h^2)/(2*sigma^2);
        G(x,y) = exp(k);
        sum = sum + G(x,y);
        if min > G(x,y)
            min = G(x,y);
        end
    end
end
B=1/min;
G= B * G;
G = round(G);
end


function [ I2 ] = image_convolution(I,w,G)
m= (w-1)/2;
N= size(I,1);
M=size(I,2);
for i=1:N
    for j=1:M
        if (i > N-m-1 || j > M-m-1 || i<m+1 || j <m+1)
            I2(i,j) = 0;
            continue;
        end
        sum1 = 0;
        for u=1:w
            for v=1:w
                sum1 = sum1+I(i+u-m-1,j+v-m-1)*G(u,v);
            end
        end
        I2(i,j)=sum1;
    end
end
end


function [ Ix ] = delx( image )
mask = [-1 0 1; -2 0 2; -1 0 1];
Ix =image_convolution(image,3,mask);
end

function [ Iy ] = dely( image )
mask = [-1 -2 -1;0 0 0;1 2 1];
Iy =image_convolution(image,3,mask);
end

function [ Imag ] = grad_mag(Ix,Iy)
m=size(Ix,1);
n=size(Ix,2);
for i=1:m
   for j=1:n
            Imag(i,j) =sqrt(Ix(i,j)^2 + Iy(i,j)^2);
   end
end
end

function [ It ] = suppression( If,Ix,Iy )
m=size(Ix,1);
n=size(Ix,2);
for i = 1:m
   for j=1:n
           if (j == 1 || j == n || i == 1 || j == n)
                It(i,j) = 0;
           else if (Ix(i,j)*Iy(i,j)> 0)
               f1 =If(i-1,j-1);
               f2 =If(i,j);
               f3 =If(i+1,j+1);
               It(i,j) = thinning(f1,f2,f3);
                else if(Ix(i,j)*Iy(i,j)< 0)
                    f1 =If(i+1,j-1);
                    f2 =If(i,j);
                    f3 =If(i-1,j+1);
                    It(i,j) = thinning(f1,f2,f3);  
                    else if(abs(Ix(i,j))-abs(Iy(i,j))>5)
                            f1 =If(i-1,j);
                            f2 =If(i,j);
                            f3 =If(i+1,j);
                            It(i,j) = thinning(f1,f2,f3);  
                            else if(abs(Iy(i,j))-abs(Ix(i,j)) > 5)
                                f1 =If(i,j-1);
                                f2 =If(i,j);
                                f3 =If(i,j+1);
                                It(i,j) = thinning(f1,f2,f3);
                                end
                        end
                    end
               end
           end
   end
end

end

function [ w ] = thinning( f1,f2,f3 )
if( f2>f1 && f2>f3)
    w =1;
else 
    w= 0;
end
end

function sz = mask_size(sigma)
sz = floor(6*sigma) + 1;
end

There is a lot of noise... how can i solve the error? i need some help....

Upvotes: 7

Views: 30666

Answers (4)

user4574539
user4574539

Reputation:

Seeing your Output Edge Image, I can say that You should check..

  1. Hysteresis Function is working properly or not
  2. You may take HIGH threshold little higher
  3. you can smooth the image little more.

Upvotes: 0

Wajahat
Wajahat

Reputation: 1

Your problem is at thresholding add a strong thresholder to get rid of false edges.

First you have to smooth image with a Gaussian function. Then find gradient and magnitude of input image. Perform nonmaxima suppression. After that do hysteresis thresholding.

Upvotes: 0

lakshmen
lakshmen

Reputation: 29094

Error is actually at thinning function. 

    if( f2>f1 && f2>f3)   
      w =f2; 
    else      
       w= 0;

Upvotes: 4

0x90
0x90

Reputation: 41022

You should do both:

  1. Smooth image to eliminate noise (conv with a gaussian matrix) before do any manipulation on it.
  2. Take a higher threshold in Hysteresis part of the algorithm:

Take larger T1 when you do this part of the algorithm:

Define two thresholds T1 > T2

for every pixel with value greater than T1 is presumed to be an edge pixel.

Upvotes: 3

Related Questions