How to get Areas of Specified Objects in an Image?

I have been trying to take an image and divide it up into regions, and then calculate various properties of those regions. A sample image is linked: <http://tinypic.com/view.php?pic=esq2it&s=7>

   I have been running the following code:

    close all;
    clear all;
    RGB = imread('vir1036.tif');
    threshold = graythresh(RGB);
    BW = im2bw(RGB,threshold);
    L = bwlabel(BW);
    imshow(BW)
    figure
    imagesc(L)

L looks like this: <http://tinypic.com/view.php?pic=2cd998w&s=7>

I am new at MATLAB and still learning about all of these commands. As you can see, there is some noise, but bwlabel has identified most of the regions I was targeting (the regions in various colors of blue). I would like to be able to find the area of each of those regions separately. Is there a way to clean up this image so that it find all of the regions I was targeting, and do you have any suggestions for how to find the areas of each of those regions?

Thank you in advance for any help!

Answer 1

[Sean de Wolski]

Edit Per Comments

%%Answers 7/21
I = imread('coins.png');
Ibw = I>100;

Edited Again

Ibw = bwareaopen(Ibw,20); %remove eveyr object with less than 20px.
CC = bwconncomp(Ibw);

imshow(Ibw);
H = msgbox('Click on Object of Interest');
uiwait(H);
while 1
    [x y] = ginput(1);
    Pxidx = sub2ind(size(Ibw),round(y),round(x)); %pixel linear index
    Objidx = cellfun(@(x)any(x==Pxidx),CC.PixelIdxList); %what object?
    if ~any(Objidx)
        H = errordlg('Doh! You didn''t select a valid object');
        uiwait(H);
        H = msgbox(sprintf('Let''s try again\nClick on Object of Interest'));
        uiwait(H);
    else
        break;
    end
end    
ObjArea = numel(CC.PixelIdxList{Objidx}); %get area
msgbox(sprintf('Area is %i Px',ObjArea)); %show area

[Cynth]

I tried

stats = regionprops(L,'Area')

stats = regionprops(L, 'CovexArea')

they returned

stats =

104x1 struct array with fields:

ConvexArea

stats =

104x1 struct array with fields:

Area

How do I use region props to find for example the region of the blue region in the lower left hand corner

[Nathan Greco]

You can use ginput to have the user point to the region, get the x/y coordinate (column/row), and compare that to the labeled image. When using these coordinates on the labeled image, you should get a number. This number is the index you want to use in your stats struct: stats(index).Area

Hopefully this helps.

[Sean de Wolski]

You could also use bwconncomp to get the pixelidxlist. Then use ginput -> convert to linear index with (sub2ind) and then ismember with cellfun on the CC.PixelIdxList, to find which region it's in by index. Then the area will correspond to that one

[Cynth]

Both your comments were very helpful! I was able to get the area of each region by the number of pixels. Unfortunately, I am doing this for a number of pictures, and the index of a region keeps changing because bwlabel labels some of the noise. Is there anyway to get bwlabel to only label pixels over a certain number?

Example:

if size(pixels)>20,

L = bwlabel(BW);

end

[Sean de Wolski]

Use bwareaopen to get rid of small stuff (noise).

[Cynth]

weirdly at 50 it gets rid of smaller regions I want but no small dots. At 10 nothing happens.

[Sean de Wolski]

I'll bet you'd be surprised at how small an object actually is if it only has 10px. Did you see the above code?

[Cynth]

I just did. It works very well, but you also have smaller regions that did not get removed by bwareaopen (the coin near the top right). Why doesn't that disappear?

[Sean de Wolski]

Because it's greater than 20 pixels in size... These are the sizes of every coin:

cellfun('prodofsize',CC.PixelIdxList)

ans =

2693 1899 2725 1772 2796 2563 2648 2598

1935 1906

I'd have to up the ante in bwareaopen.

[Cynth]

I tried 800 and even the smallest ones still didn't go away.

[Sean de Wolski]

cellfun('prodofsize',CC.PixelIdxList)

run this and see how big your objects actually are.

Or do a histogram of that - etc. Post an image, but I think the code above is basically what you want.

[Cynth]

I meant in your image. Were you able to make those small areas disappear? I could not even at 800.

[Sean de Wolski]

In my image none of the areas were below 800!!! They were the sizes that I showed you four comments ago.

[Cynth]

I see I'm looking at what you caned an invalid object (when you click on it it says that you did not select a valid object. I believe I figured it out by using imcomplement before using bwareaopen! Thank you so much! I'm sorry for the misunderstanding!

Answer 2

[MUHAMMED EROL]

Hello Sean de Wolski how do I calculate the translation of these objects?

Answer 3

[MUHAMMED EROL]

PERİMETER?

[Image Analyst]

Muhammed, see my Image Segmentation Tutorial.

In short

props = regionprops(binaryImage, 'Perimeter');
allPerimeters = [props.Perimeter]

[MUHAMMED EROL]

THANKS