Image Processing: Counting Lines in a Range of Angles

Hi,

I'm working on a problem right now that involves a method to automate scanning drill core data from a mining property and identifying volume of vein and the angle at which the vein crosses the core axis.

With help from this forum and online, I've gone from the original image to the cleaned image. The cleaned image identifies the veins in the photo quite well.

If possible, it would be great if I could remove the rectangular sample tags and mini whiteboard. This step isn't completely necessary.

The main thing that I now need to do is somehow measure the angle of the veins to horizontal. I need to fit these objects with a line then identify lines between say 15 and 75 degrees (I don't want horizontal or vertical lines). Once they're identified, I need to record the angle to horizontal and count the number of them.

I'm trying to use the hough transform, houghpeaks and houghlines functions but it seems to be only grabbing the long, horizontal lines.

Thanks

Answer 1

[Amir]

Hi Cole. Please try this code. I hope it can give you some ideas

    clc
  clear all
  close all
  [filename, pathname] = uigetfile('*','File Selector');
  I = imread(strcat(pathname,'\',filename)); % for example FileName='MyImage.jpg'
  I=im2bw(I);
  BW = edge(I,'canny',0.1);
  [bw, loc2]= imfill(BW,'holes');
  % http://www.mathworks.co.uk/help/images/ref/regionprops.html
  rp = regionprops(bw,'All'); % you can specify the parameters which you need
  ObjOrient=zeros(size(rp,1),1);
  CenterX=zeros(size(rp,1),1);
  CenterY=zeros(size(rp,1),1);
  for i=1:size(rp,1)
      ObjOrient(i)=rp(i).Orientation;
      CenterX (i)= rp(i).Centroid(1);
      CenterY (i)= rp(i).Centroid(2);
      % you can other properties (for example area, perimeter etc here)
  end

Final=[ObjOrient CenterX CenterY];
imshow(I);
hold on
for i=1:size(Final,1)
    text(Final(i,2),Final(i,3),num2str(Final(i,1)),...
    'HorizontalAlignment' , 'center',...
    'VerticalAlignment' , 'middle');
end
title('Coordination of Particles - between -90 and +90');
figure
hist(Final(:,1));
title('Histogram');
xlabel('Angle - Degree');
ylabel('Number of objects');

[Cole]

Amir,

Thank you! That worked really well for generating a histogram of angles which is definitely information that I need.

Do you have any advice on how to further clean the image before running the code that you posted? If possible, I would like to be able to remove the rectangular objects. In the example above, there are 9 small rectangular sample tags and 1 whiteboard. If I could remove those, I would have pretty much everything that I need.

Just out of curiosity, why do you think Hough lines wouldn't work well on my example above? I tried running hough, houghpeaks and houghlines following along with the examples on Matlab's help and it kept plotting the end points in pretty much the same place. My guess is that my veins are straight enough lines for the hough algorithm to pick up but I'm not sure how to relax that requirement

[Amir]

Just a quick question Cole. I will back soon :). Do you have many images like this that you would like to analyse or you have only one.

[Cole]

Hi Amir,

If I can get this to work well enough, I have over 20,000 images to analyse :)

This is a side project as part of my MSc thesis work. I'm using a discrete fracture network to model quartz vein data. One of the hardest parts is getting accurate veining information. The dataset that I'm using for a case study has information on intensity but not very good information on distributions of angles and thickness.

If I can automate this well enough to work, I can process photos of 200,000 meters of drill core and generate vein volume plus angle and thickness distributions which I need for the fracture simulation.

This is my first foray into image processing. It's very interesting though!

Cole

[Amir]

Hi Cole Sorry for delay. I wanted to think about this and write a code (if I can) for you. But these days I am very busy and couldn't find any free time to think about this. I think you need to combine object detection (code above) with the colour detection. i.e. for those objects which have -5<orientation <5 (horizontal)find the colour histogram of ORIGINAL image and if white colour is dominant then you can say most likely that object is a label. Hope this helps.