Canny Edge Detector

Mikel Rodriguez Introduction:

Edge detection is an essential preprocessing step in many computer vision algorithms. Within this project we implement one of these methods, the Canny Edge Detector.

The Canny edge detector is a popular method for detecting edges that begins by smoothing an image by convolving it with a Gaussian of a given sigma value. Based on the smoothed image, derivatives in both the x and y direction are computed; these in turn are used to compute the gradient magnitude of the image.

Once the gradient magnitude of the image has been computed, a process called non maximum suppression is performed; in which pixels are suppressed if they do not constitute a local maximum.

The final step in the canny edge detector is the hysteresis operator, in which pixels are marked as either edges, non edges and in-between, this is done based on threshold values. The next step is to consider each of the pixels that are in-between, if they are connected to edge pixels these are marked as edge pixels as well.

The result of this edge detector is a binary image in which the white pixels closely approximate the true edges of the original image.

Instructions for executing the edge detector:

I have implemented the edge detector in Matlab. To run the program, set the appropriate paths and run the function in the following manner:

>>canny_edges(MAX,MIN,SIGMA)

In which MAX is the maximum threshold value for hysteresis, and MIN value is the minimum. SIGMA is used for generating the Gaussian kernels.

Results

For:

Max Threshold:1.5

Min threshold 0.05

Sigma :1 Derivative In X Derivative in Y Gradient After non maximum suppression After Hysteresis

For:

Max Threshold:1.5

Min threshold 0.05

Sigma :1 Derivative In X Derivatice in Y Gradient After non maximum suppression After Hysteresis

For: A max threshold of 1, a min of 0.04 and a sigma of 1.2 Derivative in x Derivative in y Gradient Non Maximum Suppression Hysteresis

For: A max threshold of 1, a min of 0.04 and a sigma of 1.2 Derivative in x Derivative in y Gradient Non Max Suppression Hysteresis

Source Code

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% CAP 5405, Fall 2005                                                                                     %

% Canny Edge Detector                                                                                   %

% Mikel Rodriguez                                                                                            %

%                                                                                                                       %

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function canny_edges(max_hysteresis_thresh, min_hysteresis_thresh,sigma);

%%INIT VARIABLES:

%%Convert to double:

ORIGINAL_IMAGE=im2double(ORIGINAL_IMAGE);

%%Save height and width

[H,W]=size(ORIGINAL_IMAGE);

%%Derivatives in x and y

derivative_x=zeros(H,W);

derivative_y=zeros(H,W);

%%Gaussian kernel

size_of_kernel = 6*sigma+1;

Y_GAUSSIAN=zeros(size_of_kernel,size_of_kernel);

X_GAUSSIAN=zeros(size_of_kernel,size_of_kernel);

%%Create gaussian kernels for both x and y directions based on the sigma

%%that was given.

for i=1:size_of_kernel

for iiii=1:size_of_kernel

Y_GAUSSIAN(i,iiii) = -( (i-((size_of_kernel-1)/2)-1)/( 2* pi * sigma^3 ) ) * exp ( - ( (i-((size_of_kernel-1)/2)-1)^2 + (iiii-((size_of_kernel-1)/2)-1)^2 )/ (2*sigma^2) );

end

end

for i=1:size_of_kernel

for iiii=1:size_of_kernel

X_GAUSSIAN(i,iiii) = -( (iiii-((size_of_kernel-1)/2)-1)/( 2* pi * sigma^3 ) ) * exp ( - ( (i-((size_of_kernel-1)/2)-1)^2 + (iiii-((size_of_kernel-1)/2)-1)^2 )/ (2*sigma^2) );

end

end

non_max = zeros(H,W);

post_hysteresis = zeros(H,W);

%%Image Derivatives:

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

reference_row=  r-ceil(size_of_kernel/2);

reference_colum=  c-ceil(size_of_kernel/2);

for yyy=1:size_of_kernel

for yyy_col=1:size_of_kernel

derivative_x(r,c) = derivative_x(r,c) + ORIGINAL_IMAGE(reference_row+yyy-1, reference_colum+yyy_col-1)*X_GAUSSIAN(yyy,yyy_col);

end

end

end

end

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

reference_row=  r-ceil(size_of_kernel/2);

reference_colum=  c-ceil(size_of_kernel/2);

for yyy=1:size_of_kernel

for yyy_col=1:size_of_kernel

derivative_y(r,c) = derivative_y(r,c) + ORIGINAL_IMAGE(reference_row+yyy-1, reference_colum+yyy_col-1)*Y_GAUSSIAN(yyy,yyy_col);

end

end

end

end

%%Compute the gradient magnitufde based on derivatives in x and y:

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

GRADIENT(r,c) = sqrt (derivative_x(r,c)^2 + derivative_y(r,c)^2 );

end

end

%%Perform Non maximum suppression:

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

%%quantize:

if (derivative_x(r,c) == 0) tangent = 5;

else tangent = (derivative_y(r,c)/derivative_x(r,c));

end

if (-0.4142<tangent & tangent<=0.4142)

non_max(r,c)=0;

end

end

if (0.4142<tangent & tangent<=2.4142)

non_max(r,c)=0;

end

end

if ( abs(tangent) >2.4142)

non_max(r,c)=0;

end

end

if (-2.4142<tangent & tangent<= -0.4142)

non_max(r,c)=0;

end

end

end

end

post_hysteresis = non_max;

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

if(post_hysteresis(r,c)>=max_hysteresis_thresh) post_hysteresis(r,c)=1;

end

if(post_hysteresis(r,c)<max_hysteresis_thresh & post_hysteresis(r,c)>=min_hysteresis_thresh) post_hysteresis(r,c)=2;

end

if(post_hysteresis(r,c)<min_hysteresis_thresh) post_hysteresis(r,c)=0;

end

end

end

vvvv = 1;

while (vvvv == 1)

vvvv = 0;

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

if (post_hysteresis(r,c)>0)

if(post_hysteresis(r,c)==2)

if( post_hysteresis(r-1,c-1)==1 | post_hysteresis(r-1,c)==1 | post_hysteresis(r-1,c+1)==1 | post_hysteresis(r,c-1)==1 |  post_hysteresis(r,c+1)==1 | post_hysteresis(r+1,c-1)==1 | post_hysteresis(r+1,c)==1 | post_hysteresis(r+1,c+1)==1 ) post_hysteresis(r,c)=1;

vvvv == 1;

end

end

end

end

end

end

for r=1+ceil(size_of_kernel/2):H-ceil(size_of_kernel/2)

for c=1+ceil(size_of_kernel/2):W-ceil(size_of_kernel/2)

if(post_hysteresis(r,c)==2)

post_hysteresis(r,c)==0;

end

end

end

imwrite(ORIGINAL_IMAGE,'C:\Documents and Settings\User\Desktop\Results\original_image.bmp');

imwrite(derivative_x,'C:\Documents and Settings\User\Desktop\Results\derivative_x.bmp');

imwrite(derivative_y,'C:\Documents and Settings\User\Desktop\Results\derivative_y.bmp');

imwrite(non_max,'C:\Documents and Settings\User\Desktop\Results\non_max_supr.bmp');

imwrite(post_hysteresis,'C:\Documents and Settings\User\Desktop\Results\original_image.bmp');