Calculating mean angles of edge detection

Hello,

I'm a new user of Matlab and am looking to calculate the mean and median angles of an image after performing an edge detection. I am using the following edge detection script adapted from the MathWorks edge detection example code and am looking to take it a step further in order to calculate the mean and median angles of the edge detection for each image I input. I eventually am looking to change the code from a specific image to a moving window. Additionally, I'm curious whether it is possible to calculate the number of horizontal, vertical, and diagonal edges from the edge detection output?

Thank you!

clear
RGB=imread('DJI_0022.jpg'); %inputs image
I=rgb2gray(RGB); %convers to grayscale

figure
imshow(I)

%%
%https://www.mathworks.com/help/images/ref/edge.html
%two different types of edge detection
BW1 = edge(I,'Canny');
BW2 = edge(I,'Prewitt');

figure
imshowpair(BW1,BW2,'montage')

%%
%https://www.mathworks.com/matlabcentral/answers/8591-edge-direction
%vertical
BWv = imopen(BW1,[1 1 1]');
%horizontal:
BWh = imopen(BW1,[1 1 1]);
%top left to lower right
BWd = imopen(BW1,[1 0 0; 0 1 0; 0 0 1]);

%%
%plot results
figure
subplot(3,1,1)
imshow(BWv);
title('vertical')

subplot(3,1,2)
imshow(BWh);
title('horizontal')

subplot(3,1,3)
imshow(BWd);
title('diagonal')

[Anna Marshall]

This is the image I am currently using for reference!

[darova]

• I'm curious whether it is possible to calculate the number of horizontal, vertical, and diagonal edges from the edge detection output?

Yes. THe answer is yes. Use bwlabel

[L,n] = bwlabel(BWh,4);         % '4' is connection

Little explanation:

I'd use 4 connected object for horizontal and vertical lines. And 8 connected object for diagonals

See more here: bwlabel

[Anna Marshall]

Thanks so much! I tried it out and bwlabel works great. Any ideas on how to calculate mean angles for each direction (horizontal, vertical, and diagonal)?

Answer 1

[darova]

Try regionprops with orientation property

Example

BWd1 = 1.0*BWd;
data = regionprops(BWd,'orientation','pixelidxlist');   % get orientation and pixels list of each region
for i = 1:numel(data)
    idx = data(i).PixelIdxList;                         % pixel indices of current region
    ori = data(i).Orientation;                          % orintation of region
    BWd1(idx) = ori;                                    % assign angle value
end
imagesc(BWd1)
colorbar

[Anna Marshall]

Thank you! The orientation of the region is a neat calculation.Is the orientation calculating from 0-360 degrees? I did notice when I ran it for each of the horizontal (ori value of 0), vertical (ori value of 90), and diagonal images (ori value of -45) , which were more "standard" values than I might expect.

[darova]

• Is the orientation calculating from 0-360 degrees?

No. As you can notice it's -180 .. +180

• I did notice when I ran it for each of the horizontal (ori value of 0), vertical (ori value of 90), and diagonal images (ori value of -45) , which were more "standard" values than I might expect.

Please see values of orientation

data(:).Orientation;

[Anna Marshall]

Thanks!

[Anna Marshall]

@darova- I have a belated follow-up question on this. In thinking about the goal a bit more, I think it makes more sense to gather all orientation values (i.e. the gradient of all horizontal values, the gradient of all vertical values, etc.) Is there a way to get all the orientation values as an output instead of just the mean orientation values?

Thanks!

[darova]

I don't understand what you say

I = imread('im.png');
I1 = ~im2bw(I);

data = regionprops(I1,'orientation','centroid');
cent = cat(1,data.Centroid);
ori  = cat(1,data.Orientation);

imshow(I1)
for i = 1:length(ori)
    h = text(cent(i,1)+5,cent(i,2)+5,num2str(ori(i)));
    set(h,'color','y')
end

Can you make a simple drawing of what the result you want to see? Here is my sketch

[Anna Marshall]

Absolutely. Your drawing is what I am trying to achieve. I have code that adds up the number of edges at each specified orientation (horizontal, vertical, left diagonal, right diagonal) and then calculates the mean angle of each orientation. What I would like is to modify my code to instead calculate EACH angle and report those values. Below is the code I have for average orientation from my binary edge detected image:

%vertical
BWv = imopen(BW1,[1 1 1]');
%horizontal:
BWh = imopen(BW1,[1 1 1]);
%top left to lower right
BWdleft = imopen(BW1,[1 0 0; 0 1 0; 0 0 1]);
%top right to lower left
BWdright = imopen(BW1,[0 0 1; 0 1 0; 1 0 0]);
%calculate horizontal # of edges
[H,Hnum] = bwlabel(BWh,4); 
%calculate vertical # of edges
[V,Vnum] = bwlabel(BWv,4); 
%calculate diagonal left # of edges
[DL,DLnum] = bwlabel(BWdleft,8); 
%calculate diagonal right # of edges
[DR,DRnum] = bwlabel(BWdright,8); 

%% Diagonal Left orientation
BWdl1 = 1.0*BWdleft;
% get orientation and pixels list of each region
data = regionprops(BWdleft,'orientation','pixelidxlist');   
for i = 1:numel(data)
    % pixel indices of current region
    idxDL = data(i).PixelIdxList;     
    % orientation of region
    oriDL = data(i).Orientation;   
    % assign angle value
    BWdl1(idxDL) = oriDL;                                    
end
imagesc(BWdl1)
colorbar

%% Vertical Orientation
BWv1 = 1.0*BWv;
data = regionprops(BWv,'orientation','pixelidxlist');   % get orientation and pixels list of each region
for i = 1:numel(data)
    idxV = data(i).PixelIdxList;                         % pixel indices of current region
    oriV = data(i).Orientation;                          % orintation of region
    BWv1(idxV) = oriV;                                    % assign angle value
end
imagesc(BWv1)
colorbar

%% Horizontal Orientation
BWh1 = 1.0*BWh;
data = regionprops(BWh,'orientation','pixelidxlist');   % get orientation and pixels list of each region
for i = 1:numel(data)
    idxH = data(i).PixelIdxList;                         % pixel indices of current region
    oriH = data(i).Orientation;                          % orintation of region
    BWh1(idxH) = oriH;                                    % assign angle value
end
imagesc(BWh1)
colorbar
%% 
%% Diagonal Right orientation
BWdr1 = 1.0*BWdright;
data = regionprops(BWdright,'orientation','pixelidxlist');   % get orientation and pixels list of each region
for i = 1:numel(data)
    idxDR = data(i).PixelIdxList;                         % pixel indices of current region
    oriDR = data(i).Orientation;                          % orintation of region
    BWdr1(idxDR) = oriDR;                                    % assign angle value
end
imagesc(BWdr1)
colorbar

[darova]

What about this?

[Anna Marshall]

That's very close. Because my image has so many angles, I'm hoping to have the values in the workspace as a table rather than printed on the image. What are m and n in your script? I've attached my edge detection script with orientations for reference.

[darova]

• What are m and n in your script?

Please see this line

[m,n] = size(I2);

What should i do with your script?

[Anna Marshall]

Do you have any suggestions on how to get these angles to print as a table in the workspace instead of on the image?

[darova]

print it like this

sprintf('angle: %4d; center = (%3d, %3d)\n',round([ori cent])')

[Anna Marshall]

Neat- that works great for printing! Thank you!

[Anna Marshall]

Do you know of any functions that convert the angles from -90 to 90 to 0 to 360 degrees?

[darova]

Try wrapTo360

[Anna Marshall]

Thanks! I'll give that a try

Answer 2

[Image Analyst]

If you have the Computer Vision Toolbox, try extractHOGFeatures() to get the histogram of oriented gradients.

[Anna Marshall]

The extractHOGFeatures() tool is neat just from working throught the examples. Is there a way to clip the tool to a specific area of the image or train it so that it is only extratcting the HOG features for the area of interest. For example, in the image attached, I am interested in looking at the logs but not the vegetation and trees. Thank you!