How to count black pixels in a region of an image that can only have 1 white neighbor pixel

I have a binary image (black background and white pixels). How do i count the number of blackpixels witch in it's neighbourhood of 5*7 (5 lines and 7 colums) can only have a white pixel.

I can try using bwhitmiss() but i would need 35 matrixs with one white pixel in each "matrix slot".

Can you tell me an easyest way to achieve this?

Answer 1

[Ashish Uthama]

You could try leveraging conv2.

Here is the idea:

in = [
    0 0 0 0 0 0 0 0 
    0 1 0 0 0 0 0 0 
    0 0 0 0 0 0 0 0 
    0 0 0 0 0 0 0 0
    0 0 1 0 0 0 0 0 
    0 0 0 0 0 0 0 0 
    0 0 0 0 0 0 0 0 ]

k = [
    1 1 1
    1 1 1
    1 1 1]

r = conv2(in,k,'same')

The result r will have 0 at all locations which DO NOT have 1's in the immediate 3x3 neighborhood.

(Have a look at conv2 and play with the positions of 1's in in to get a better idea)

[João Viveiros]

I think i didn't explain well.

Can you tell me how i can create a matrix like:

A=imread('image.png');

B=zeros(3,3)

Now i want to create:

SE1=[1 0 0

0 0 0

0 0 0]

SE2= [0 1 0

0 0 0

0 0 0]

SE3= [0 0 1

0 0 0

0 0 0]

SE4=[0 0 0

1 0 0

0 0 0]

SE5=[0 0 0

0 1 0

0 0 0]

.... basicaly i want to generate "n" matrixes with the one always in a diferent spot. But i want all the possibilities.

In this example i have 9 spots where my 1 can appear. I want as many matrix as 1's that can appear.

At last i will have 9 structural elements without writing them down one by one.

In the end i can use:

C=bwhitmiss(A,SE(),~SE())

And with this i will be able to know how many black pixels in regios 3*3 with only one pixel in it.

Was i clear? my english is a bit rusty.

Thanks again in advance

[Image Analyst]

Yeah, we understood. We were just suggesting conv2() as a better approach than calling bwhitormiss 9 (or 35) times. Did you run my code below yet?

[João Viveiros]

Yes i ran your code on my image but i think it only considers the white pixels that are centered in a 5 by 7 window isolated. I need the white pixels in the other positions to be considered as well. For example pixels like this one:

[0 0 0 0 0 0 1

0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 0]

I think conv2 only works when the pixel is centered like this:

[0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 1 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 0]

right?

As long as one pixel is detected in one of those positions of the 5 bu 7 matrix it will be considered for the total.

It's tricky. Get the idea?

Thank you very much again.

[Image Analyst]

OK. Well you accepted, so I guess you got it working and that's all that counts.

[Ashish Uthama]

João, conv2 pads with zeros around the edges, and it ought to work. convole these two images with [1 1 ; 1 1] to see whats going on and to understand the 'neighborhood' being used.

[João Viveiros]

I made it.I use this aproach and i think i made it:

clc;clear all;close all;

A=imread('image.png');

a=5;b=7;c=a*b;

SE = reshape(eye(c), a, b, c);

SE(:,:,(c+1)/2)=SE(:,:,(c+1)/2)*0;

B=0;

for n=1:c

B=B+bwhitmiss(A,SE(:,:,n),~SE(:,:,n));

end

sum(sum(B==1))

imshow(B);

Thanks again.

Answer 2

[Image Analyst]

Here is the solution in case you want to compare yours to mine:

clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
workspace;  % Make sure the workspace panel is showing.
fontSize = 20;

% Create an image of all zeros.
imageArray = false(50, 70);
% Get up to a tenth of the pixels as white dots.
numberOfSinglePixels = 100 %int32(rand * numel(imageArray) / 10)
scrambledCoords = randperm(numel(imageArray));
coordinatesToSet = scrambledCoords(1:numberOfSinglePixels)';
imageArray(coordinatesToSet) = true;
% Display the image
imshow(imageArray, 'InitialMagnification', 300);
title('Original Image', 'FontSize', fontSize);
% Enlarge figure to full screen.
set(gcf, 'units','normalized','outerposition',[0 0 1 1]);

% Count up the dots in a sliding 5x7 window.
windowRows = 5;
windowCols = 7;
sumImage = conv2(double(imageArray), ones(windowRows, windowCols), 'same');
% Extract only those pixels where the window is centered
% at dots in the input image.
sumImageMasked = sumImage .* double(imageArray);
% Count those places where the count is exactly 1.
numberOfDots = sum(sumImageMasked(:) == 1)

% Get the locations of those dots in a 5x7 black neighborhood.
[dotRows dotCols] = find(sumImageMasked == 1);
% Plot crosses on those dots.
hold on;
plot(dotCols, dotRows, 'r+', 'MarkerSize', 25);
caption = sprintf('Red crosses over the dots that are in an isolated %x by %d window',...
  windowRows, windowCols);
title(caption, 'FontSize', fontSize);

message = sprintf('%d of the %d pixels are in a black neighborhood of %d by %d\n',...
  numberOfDots, numberOfSinglePixels, windowRows, windowCols);
fprintf('%s\n', message);
msgbox(message);