cluster analysis from edge detection

Question

Anna Marshall 2020-5-3

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/522700-cluster-analysis-from-edge-detection

评论： Anna Marshall 2020-5-14

采纳的回答： Image Analyst

DJI_0022.JPG

在 MATLAB Online 中打开

Hello,

I am looking to perform a cluster analysis (K-means) for an image. I've converted the original image to a binary image and performed an edge detection, which should help with the clustering, but am stuck on how to implement a cluster analysis from the binary image. Any suggestions are greatly appreciated! The code for the binary edge detection is below and the image is attached. The end goal here is to identify patterns of wood orientation.

clear
RGB=imread('DJI_0022.jpg'); %inputs image
I=rgb2gray(RGB); %convers to grayscale
figure
imshow(I)
BW1 = edge(I,'Canny',0.6);
imshow(BW1)

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Image Analyst 2020-5-3

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/522700-cluster-analysis-from-edge-detection#answer_429990

在 MATLAB Online 中打开

Assuming you have a bunch of blobs that are the tree edges, I'd call regionprops() and ask for 'Orientation'. This will give you the average angle of every blob. You can then histogram the angles if you want.

props = regionprops(edgeImage, 'Orientation');
allAngles = [props.Orientation];
histogram(allAngles);
grid on;
xlabel('Angle', 'FontSize', 20);
ylabel('Count', 'FontSize', 20);

16 个评论
显示 14更早的评论隐藏 14更早的评论

Image Analyst 2020-5-3

You can get the angles from the histogram of the orientations. You can look at various peaks in the histogram. Look at it this way: let's say the orientations were completely random from 0 to 360 so that the histogram was flat or almost flat. What would you say k should be? 2? 3? 10? Whatever you pick is arbitrary, and the groupings would be aribtrary and changing from one image to the next. Sure if you had trees only at 30, 60, and 90 degrees, you might say k=3, but what is the likelihood you'll have 3 groupings? Chances are there will be trees at in between angles, so then where would kmeans put the centroids? Maybe at 25, 64, 87 one time, and for another image at 29, 51, 89 degrees. Why would you want that? I still maintain you would want to look at how many trees there are at every angle or in every 10 degree range, depending on how many bins you told it to use. You could use findpeaks() on the histogram counts if you want to find local groupings. But at least with a histogram you have control over the angles get counted. With kmeans you don't have that control. All you can control is how many groups there are. Even if there are no groupings, if you tell kmeans to find 5 angle groups, it will be forced to find 5 whether or not those are valid groups.

Image Analyst 2020-5-3

在 MATLAB Online 中打开

Anna, see this code where I drew a line over the gray scale image where the line color corresponds to the angle:

% Initialization steps.  Brute force cleanup of everything currently existing to start with a clean slate.
% Just for the demo.  If you put this into your own function you'll want to get rid of the close and clear commands.
clc;    % Clear the command window.
fprintf('Beginning to run %s.m ...\n', mfilename);
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
rgbImage = imread('DJI_0022.jpg'); %inputs image
hFig = figure;
subplot(2, 2, 1);
imshow(rgbImage)
axis('on', 'image');
impixelinfo;
title('Original RGB Image', 'FontSize', 20);
grayImage = rgb2gray(rgbImage); %convers to grayscale
subplot(2, 2, 2);
imshow(grayImage)
axis('on', 'image');
impixelinfo;
title('Gray Scale Image', 'FontSize', 20);
edgeImage = edge(grayImage, 'Canny', 0.6);
subplot(2, 2, 3);
imshow(edgeImage)
axis('on', 'image');
title('Canny Edge Image', 'FontSize', 20);
hFig.WindowState = 'maximized';
drawnow;
% Get coordinates and plot a line over them.
props = regionprops(edgeImage, 'PixelList', 'Orientation', 'Centroid');
lineLength = 100; % Whatever you want.
subplot(2, 2, 2); % Switch to the gray scale image.
hold on;
lineColorMap = jet(181); % One color for every 1 degrees.
for k = 1 : length(props)
% 	thisList = props(k).PixelList;
% 	fprintf('\nFor blob #%d of %d:\n    ', k, length(props))
% 	for k2 = 1 : size(thisList, 1)
% 		thisx = thisList(k2, 1);
% 		thisy = thisList(k2, 2);
% 		fprintf('(%d, %d), ', thisx, thisy);
% 	end
	% Get the centroid
	x = props(k).Centroid(1);
	y = props(k).Centroid(2);
	% Get the angle.
	angle = props(k).Orientation;
	% Get the line endpoints.
	x1 = x + lineLength * cosd(angle);
	x2 = x - lineLength * cosd(angle);
	y1 = y + lineLength * sind(angle);
	y2 = y - lineLength * sind(angle);
	thisColor = lineColorMap(round(angle + 90) + 1, :);
	fprintf('Drawing line from (%.1f to %.1f) to (%.1f, %.1f)\n', x1, y1, x2, y2);
	line([x1,x2], [y1,y2], 'Color', thisColor);
	if mod(k, 100) == 0
		drawnow;
	end
end
% Plot the histogram of angles.
subplot(2, 2, 4)
allAngles = [props.Orientation];
histObject = histogram(allAngles);
grid on;
xlabel('Angle', 'FontSize', 20);
ylabel('Count', 'FontSize', 20);
title('Histogram of angles', 'FontSize', 20);

Anna Marshall 2020-5-4

cannyoverlay.PNG

Interestingly, it does look like the Canny filter lines up well with the wood orientations in the image overlay. Looking back at the code, I'm wondering if the code to find the centroid angles before plotting those lines is what is creating a discrepancy in the colored lines?

Image Analyst 2020-5-4

在 MATLAB Online 中打开

You might try getting the PixelList, like I already showed you in your other post, and then put those x,y values into polyfit to get the angle. Something like

% Get coordinates and fit to a line.
props = regionprops(edgeImage, 'PixelList', 'Centroid');
for k = 1 : length(props)
    thisList = props(k).PixelList;
    fprintf('\nGetting angle for blob #%d of %d.\n    ', k, length(props))
    thisx = thisList(k2, 1);
    thisy = thisList(k2, 2);
    coefficients = polyfit(thisx, thisy, 1); % Fit to a line
    angles(k) = atand(coefficients(1));
end

请先登录，再进行评论。

Answer 2

Anna Marshall 2020-5-4

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/522700-cluster-analysis-from-edge-detection#answer_430040

I'll give it a try! Thanks!

6 个评论
显示 4更早的评论隐藏 4更早的评论

Anna Marshall 2020-5-4

在 MATLAB Online 中打开

Getting rid of the k2 makes more sense! Thanks. The funny thing is, I keep running into an error in the props=regionprops line now that 'orientation' is removed. I'm using the following as code for this section, but the error is in props = regionprops(edgeImage, 'PixelList', 'Centroid ')

% Get coordinates and plot a line over them.
imshow(grayImage)
axis('on', 'image');
title('Line Grouping', 'FontSize', 20);
hFig.WindowState = 'maximized';
drawnow;
props = regionprops(edgeImage, 'PixelList', 'Centroid '); 
lineLength = 20; % Whatever you want.
hold on;
lineColorMap = jet(181); % One color for every 1 degrees.
for k = 1 : length(props)
    thisList = props(k).PixelList;
    fprintf('\nFor blob #%d of %d:\n    ', k, length(props))
    thisx = thisList(:, 1);
    thisy = thisList(:, 2);
    coefficients = polyfit(thisx, thisy, 1); % Fit to a line 
    angles(k) = atand(coefficients(1 )); 
    thisColor = lineColorMap(round(angle + 90) + 1, :);
    fprintf('Drawing line from (%.1f to %.1f) to (%.1f, %.1f)\n', x1, y1, x2, y2);
    line([x1,x2], [y1,y2], 'Color', thisColor);
    if mod(k, 100) == 0
		drawnow;
    end
end

Image Analyst 2020-5-4

在 MATLAB Online 中打开

Anna, this code works but it looks like you're still missing quite a few edges with that filter.

% Initialization steps.  Brute force cleanup of everything currently existing to start with a clean slate.
% Just for the demo.  If you put this into your own function you'll want to get rid of the close and clear commands.
clc;    % Clear the command window.
fprintf('Beginning to run %s.m ...\n', mfilename);
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
rgbImage = imread('DJI_0022.jpg'); %inputs image
hFig = figure;
subplot(2, 2, 1);
imshow(rgbImage)
axis('on', 'image');
impixelinfo;
title('Original RGB Image', 'FontSize', 20);
grayImage = rgb2gray(rgbImage); %convers to grayscale
subplot(2, 2, 2);
imshow(grayImage)
axis('on', 'image');
impixelinfo;
title('Gray Scale Image', 'FontSize', 20);
edgeImage = edge(grayImage, 'Canny', 0.6);
subplot(2, 2, 3);
imshow(edgeImage)
axis('on', 'image');
title('Canny Edge Image', 'FontSize', 20);
hFig.WindowState = 'maximized';
drawnow;
% Get coordinates and plot a line over them.
imshow(grayImage)
axis('on', 'image');
title('Line Grouping', 'FontSize', 20);
hFig.WindowState = 'maximized';
drawnow;
props = regionprops(edgeImage, 'PixelList', 'Centroid'); 
lineLength = 20; % Whatever you want.
hold on;
lineColorMap = jet(181); % One color for every 1 degrees.
angles = zeros(1, length(props));
for k = 1 : length(props)
    thisList = props(k).PixelList;
    fprintf('\nFor blob #%d of %d:\n    ', k, length(props))
    thisx = thisList(:, 1);
    thisy = thisList(:, 2);
    coefficients = polyfit(thisx, thisy, 1); % Fit to a line 
    angles(k) = atand(coefficients(1 )); 
	% Get the centroid
	x = props(k).Centroid(1);
	y = props(k).Centroid(2);
	% Get the line endpoints.
	x1 = x + lineLength * cosd(angles(k));
	x2 = x - lineLength * cosd(angles(k));
	y1 = y + lineLength * sind(angles(k));
	y2 = y - lineLength * sind(angles(k));
	thisColor = lineColorMap(round(angles(k) + 90) + 1, :);
	fprintf('Drawing line from (%.1f to %.1f) to (%.1f, %.1f)\n', x1, y1, x2, y2);    line([x1,x2], [y1,y2], 'Color', thisColor);
    if mod(k, 100) == 0
		drawnow;
    end
end
% Plot the histogram of angles.
subplot(2, 2, 4)
histObject = histogram(angles);
grid on;
xlabel('Angle', 'FontSize', 20);
ylabel('Count', 'FontSize', 20);
title('Histogram of angles', 'FontSize', 20);

Anna Marshall 2020-5-4

Thank you!! That works great. I'm going to play around with different filtersm particularly the ridgeline-finding filters and see if there is one that might work a bit better than the canny filter. Thanks again for all the suggestions!!

Anna Marshall 2020-5-14

orientationpatterntest.PNG

@Image Analyst- I have another question to throw your way! I've been playing around with this code and one thing that I'm looking to try and do is add borders that group together "clusters" of the same colors aka same angles. For example, something like the attached sketch. I've tried the boundary function, but can't get it to work quite right. Do you have any ideas on what might work?

请先登录，再进行评论。

cluster analysis from edge detection

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

16 个评论
显示 14更早的评论隐藏 14更早的评论

更多回答（1 个）

6 个评论
显示 4更早的评论隐藏 4更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

cluster analysis from edge detection

0 个评论 显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

16 个评论 显示 14更早的评论隐藏 14更早的评论

更多回答（1 个）

6 个评论 显示 4更早的评论隐藏 4更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

16 个评论
显示 14更早的评论隐藏 14更早的评论

6 个评论
显示 4更早的评论隐藏 4更早的评论